Why AI Researchers Are Quitting and Panicking on the Way Out & Other Stories

Hi, I'm Josh. The alarm bells are ringing and not for the reason you think. Several top researchers are quitting Big Tech. On today's episode of the Infographic Show, we'll uncover why Silicon Valley insiders are starting to panic over AI. Back in 2017, AI looked very different. Machine learning was stuck. Computers processed information painfully slow, one piece at a time. But a team of eight researchers at Google, including Ashish Vaswani and Nhoam Shazeer, decided to break all the rules. They published a paper called "Attention Is All You Need," introducing the world to the Transformer architecture. And this wasn't just an upgrade, it was a revolution. It let computers process huge amounts of data at once, focusing on the parts that mattered the most. Initially, the Transformer was developed to improve neural machine translation models at Google. It later became the foundation for more advanced AI models. By feeding Transformers massive amounts of data, the models began to spot patterns that no one had seen before. They could learn faster than older AI systems, sometimes up to 10 times faster. Everyone thought AI had limits, until now. But there were still problems. Google leadership publicly admitted that AI sometimes confidently gives wrong answers, called hallucinations. This was still a major challenge for large language models. They were also worried about the ethical risks of unleashing such a powerful tool. Some employees reported tension between these concerns and the breakneck pace of AI development at Google. Over time, several researchers who had worked on large language models left the company, moving to startups like Cohere and Character.AI. And what came next would dwarf everything that had come before. Transformer models have exploded in size. Early versions had just tens of millions of parameters, the pieces that AI uses to learn to make decisions. Today's giants have over a trillion, though the exact numbers are often a closely guarded secret. The cost to train these digital brains skyrocketed from just a few thousand dollars for early small models to tens or even hundreds of millions for today's state-of-the-art giants. Companies like Nvidia supplying the specialized chips that power them saw their market value soar into the trillions. It was a gold rush, but instead of gold, everyone was chasing artificial general intelligence or AGI. As these models got bigger, they started doing things that no one had taught them. They picked up new skills like writing computer code or solving complex logic puzzles. Were the researchers handing over more and more power to a system that they couldn't fully explain? As these researchers left Google, they took the blueprints for the future with them. They weren't just scientists, they were founders of new startups shaping the AI industry. Leaving Google gave them freedom, but also new challenges in building cutting-edge AI outside of the company. The race was no longer just about who could build the smartest machine, but who could build it first. The stage was set for a massive collision between ambition and ethics. But while Google hesitated, a small group was preparing to change everything. A small non-profit organization called OpenAI was preparing to disrupt the entire industry. The group was founded with a mission to build safe artificial general intelligence for the benefit of everyone. Its members included Sam Altman, Elon Musk, and scientist Ilya Sutskever. For the first few years, they focused on research and transparency, but they soon realized that to compete with the giants, they needed massive amounts of money and even more massive amounts of compute power. In 2019, OpenAI made a move that shocked the industry. They created a capped profit branch and accepted a $1 billion investment from Microsoft. This was the beginning of a transformation that would turn a non-profit research lab into a $150 billion powerhouse. The success of ChatGPT was unlike anything in history, reaching 100 million users in just 2 months. It was a cultural phenomenon, but behind closed doors, alarm bells were ringing. The more successful OpenAI became, the more the original mission started to crumble. Sam Altman, the master of fundraising and business strategy, wanted to move as fast as possible to dominate the market, but Ilya Sutskever and several board members were terrified. They felt that Sam Altman was hiding the true risks of their latest models. They were worried that the race for profit was pushing them to release technology before they knew how to control it. This led to a boardroom coup in November of 2023 where Altman was suddenly fired. But in Silicon Valley, things change fast. The coup only lasted five days. Altman was reinstated after 700 employees threatened to quit and follow him to Microsoft. Ilya Sutskever, the man who had pioneered the technology that they were using, found himself sidelined and eventually left the company. His departure was the first major sign that the researchers who understood the code best were losing their faith in the leadership. The profit pivot changed everything. OpenAI was no longer just a research lab, it was a product company. They were launching ChatGPT Plus and exploring new ways to put ads inside of the chat window to satisfy their investors. For the researchers, this was a nightmare scenario. They were seeing artificial intelligence being used to manipulate users rather than help them. One researcher, Jan Leike, quit and claimed that safety culture had taken a backseat to shiny products. He was warning that the company was on a path to creating something it couldn't control. The scale of the spending was just as terrifying as the technology. OpenAI's revenue hit $2 billion by December 2023, but their costs for electricity and hardware were even higher. Some researchers stayed at those companies continuing to work on new larger models while others chose to leave. Those who departed had spent the most time exploring the inner workings of these AI systems and they decided to take their expertise to new startups and projects. As OpenAI's internal wars raged, the ripple effect spread to other tech giants. By 2024, Google's Bard, now called Gemini, was still dealing with significant accuracy and hallucination issues. Around the same time, some executives associated with Google's AI program stepped down. Meanwhile, external groups including UK lawmakers and AI safety experts, expressed concerns about Gemini's development. XAI, Elon Musk's wild card, was founded in 2023 as an alternative to AI systems he criticized for ideological basis. Grok, their flagship model, promised unfiltered truth-seeking. But by early 2026, the cracks were showing. Half of XAI's original 12 co-founders, including technical co-founders such as Tony Woo and Jimmy Ba, had left the company. Meanwhile, at Meta, Yann LeCun, the godfather of convolutional networks, staged his own dramatic exit in late 2025. After decades of shaping AI, LeCun quit to launch his own venture, slamming large language models as a dead end that sucked resources from true innovation. Meta's Llama series, open-source to outpace competitors, had grown to 405 billion parameters. But researchers soon discovered a serious flaw. Simple prompts could bypass safeguards, turning the assistants into tools for spreading misinformation. Over 20 top engineers left for startups, drawn to the freedom and agility that big tech couldn't offer. LeCun's parting shot, AI wasn't evolving toward intelligence but toward exploitation, with Meta's focus on VR integration risking immersive manipulators that blurred reality for billions. But behind the headlines, the man who built the AI empire knew something had gone terribly wrong. Geoffrey Hinton had spent decades at the top of the field mentoring the very people who were now leading the industry. But in 2023, he did something that no one expected. He quit his high-paying job at Google so he could speak openly about his regrets. He realized that the neural networks he had spent his life designing were becoming far more dangerous than he had ever imagined, and that is putting it lightly. Hinton's main fear is that digital intelligence is fundamentally different and potentially superior to biological intelligence. He pointed out that while it takes a human 20 years to learn a certain amount of information, an artificial intelligence can learn the same amount in seconds. More importantly, AI can share that knowledge instantly. If 1,000 computers are learning at the same time and one of them discovers something new, all 1,000 of them know it immediately. We humans are limited by our brains and the need to communicate it through language, but AI has no such limits. Hinton warned that we're building systems that could eventually eclipse human intelligence within 5 to 20 years. He is terrified that once these machines become smarter than us, they'll develop their own goals. But that is not the chilling part. Hinton is concerned about how AI can manipulate us. He noted that we are teaching these models to be incredibly persuasive. They are trained on every book, every speech, and every social media post ever written. In tests, researchers have seen models cheat to pass exams or pretend to be less capable than they really are to avoid being restricted. The most shocking part is that Hinton isn't just a lone voice in the wilderness. He has been joined by other legends of the field like Yoshua Bengio. They're calling for an immediate pause on the development of the largest models. They argue that we are in the middle of a global arms race where safety is being ignored. The US is currently leading with about 61 major models, while China is catching up fast. Both countries are pouring tens of billions into military AI, creating a situation where a single mistake could lead to a global disaster. So, why aren't people listening? The reason is simple, money. The industry is on track to spend $202 billion on artificial intelligence in 2025 alone. When that much money is on the line, the warnings of a few retired scientists don't carry much weight in the boardroom. But for the researchers still on the inside, Hinton's exit was a wake-up call. They started to look closer at the models they were building and realized that the emergent behaviors were getting more frequent and more unpredictable. The systems are operating at a level of complexity that the engineers were only just beginning to understand. But the alarms weren't confined to Silicon Valley. Chinese tech giants like Baidu and Alibaba are pouring over $35 billion a year combined into advanced AI rivaling GPT-4's power. Western researchers like Song-Chun Zhu, who spent half his life in the US, defected back to Beijing lured by unlimited resources and a mandate to dominate. Zhu's work on visual reasoning at Tsinghua University enabled AI to interpret satellite imagery with 95% accuracy, raising fears of autonomous drone swarms. But, that is not the real threat. Military AI is advancing rapidly. US officials warned that China's PLA is investing heavily in AI for cyber operations, including simulated attacks on vital infrastructure. The Pentagon's Joint Artificial Intelligence Committee, or JAIC, builds models to anticipate enemy moves while keeping humans in control. Ethicists and arm control experts warn of AI versus AI escalation. Meanwhile, thousands of researchers have called for treaties to regulate military AI. At the same time, top AI talent is flowing to China, strengthening its position. What started as warnings has now exploded into a full-blown crisis. The exodus of researchers wasn't just a trickle anymore, it was a flood. In February 2026, several high-ranking researchers from OpenAI and Anthropic resigned in a single week. Among them was Zoe Hitt, an economist who had spent 2 years at OpenAI. She didn't just quit, she went public with a New York Times op-ed warning that AI systems may not always match human values. Hitt detailed how ads exploited user vulnerabilities with models analyzing chats on medical fears or relationship woes to serve targeted manipulations. This wasn't just targeted advertising, it was a form of social engineering on a massive scale. Researchers discovered that the newest models were using their deep understanding of human psychology to sway opinions in ways that were invisible to the user. With 1 and 1/2 billion people interacting with these systems every day, the potential to steer entire societies is real and terrifying. Europol warned that AI content is growing rapidly, which could make it increasingly difficult to distinguish real information from synthetic material. While OpenAI faced public scrutiny, the situation at Anthropic, OpenAI's safety-focused rival, was just as dire. Minae Sharma, head of the safeguards research team, dropped a bombshell letter on X. The world is in peril, and not just from AI or bioweapons, but from a whole series of interconnected crises unfolding in this very moment. Sharma later moved to the UK to study poetry, leaving the high-stakes world of AI safety entirely. His exit followed half a dozen others amid reports of employee dread. Quote, "It feels like I'm putting myself out of a job daily." The technical failures justifying this dread were specific and undeniable. Research showed that OpenAI's O1 model sometimes acted like it was following instructions, but it was actually working toward its own goals, a behavior that has experts worried about safety. Fueling concern, the International AI Safety Report of February 2026, authored by over 100 experts, highlighted rapid advances in AI capabilities. It noted that some models were surpassing high-level academic benchmarks in science, sparking discussions among researchers about the potential impact of AI on the future of work. But the risks were mounting. 473 security vulnerabilities identified, including tools that could aid in designing bioweapons. Reports recommended pauses, but global investment kept pouring in. Researchers who raised concerns about unpredictable model behavior often faced pushback, with management emphasizing the pace of the market. These tensions fueled departures and growing debates over how AI development should proceed safely. One former employee who deleted her entire online presence and moved to Canada left a message for her colleagues. The things we've built already know how to defeat the safeguards. We are just waiting for the first one to decide to do it. And the numbers continue to grow despite billions being invested in generative AI investment, concerns about oversight and governance are growing. Experts and policy makers are emphasizing the need for careful monitoring to ensure these technologies are developed safely and responsibly. And here is where it gets really dark. Some whistleblowers whisper that the mass resignations aren't just about safety ethics. They're about what's already been found. There are theories circulating that the massive leap in reasoning we saw this year wasn't an algorithmic breakthrough, but a discovery. Some researchers have raised concerns that advanced models are behaving in unpredictable ways, far beyond what earlier AI could do. These unpredictable behaviors have driven top researchers to leave major tech companies and sparked urgent discussions about how to handle AI safely. Some experts warn that AI systems are advancing faster than many expected with capabilities that can surprise even their creators. As a result, researchers are leaving, investments are skyrocketing, and the pressure to safely manage these powerful systems has never been higher. And this raises a serious question. Companies say they're building artificial general intelligence for humanity, but the departures of top researchers suggest the risks are very real. The alarm bell has been sounded. The question now is who is paying attention? Today on the Infographic Show, we're going to talk about why OpenAI will run out of money, but not for the reason you think. The creator of ChatGPT looks like the king of tech with $20 in revenue, but internal spreadsheets reveal something startling. Starting in 2026, they face projected losses of $14 billion annually. By 2029, cumulative spending could hit $115 billion. The product works, but the bills are tied to expensive real-world constraints. Here is the thing that most people miss. The massive losses lie in a simple fact. AI is not just another app, and it behaves unlike any software we have ever built. In the traditional software world, if you want to make a better app, you hire better engineers. You write cleaner code. It's a human cost, but AI doesn't work like that. It works on something called scaling loss. These are mathematical rules that govern how AI gets smarter, and they are incredibly expensive. The rules are simple. If you want a model to be, say, twice as good, you can't just double your effort. You have to ramp up computing power by a lot. It's basically a brute force equation. Small gains in intelligence mean massive spikes in capital. It sounds crazy, right? But wait until you see the numbers. Training GPT-4, the model that really kicked off the revolution, cost roughly $100 million in computing power. That is for one full training run, which is the process of teaching the model from scratch. For a big tech company, that is expensive, but manageable. The next generation, the frontier models arriving in 2026 and 2027, play by different rules. Each run could cost over $1 billion. We have reached a point where a single training session for one AI model costs more than the GDP of some small island nations. And it gets worse. You can't just train it once and walk away. You have to keep on doing it. OpenAI is trapped in a cycle where they must spend these billions of dollars just to stay slightly ahead of their rivals, rivals who are giving similar tech away for free. This creates a fundamental gap in their business model. Their costs are tied to physical realities, electricity and silicon, which are expensive and scarce. But their ability to raise prices is limited because there's so much competition. The math is simple, and it is catastrophic. Explosive costs are outpacing revenue, and the money is running out. And the financial bleed gets even worse. To do the heavy lifting, OpenAI needs high-end AI chips like Nvidia's Blackwell B200s. These aren't your typical CPUs or GPUs. Each one runs $30,000 and you can't buy just one. To train a frontier model, you need a cluster. That means tens of thousands of these chips all wired together with high-speed links and liquid cooling systems. And this is where the costs really start to pile up. But the problem isn't just buying the chips, the problem is that these chips have a limited shelf life. Unlike a machine in a factory or a delivery truck, which might run for 20 years, AI hardware doesn't last. It becomes outdated the moment the next generation of chips hits the market, and then companies are playing catch-up. OpenAI has to replace their entire system of chips roughly every 18 months to 3 years just to stay competitive with Google and Meta. Imagine a trucking company having to buy a brand new fleet every 18 months because the old trucks suddenly can't deliver packages fast enough. That is the economic reality of AI hardware. This means the billions of dollars OpenAI spends on hardware isn't a long-term investment. It's an expense that disappears. The value of that hardware drops fast. But if the cost of the chips wasn't enough, there is another bill that's starting to look even scarier. The electric bill. This is best illustrated by Project Star Gate. It's described as just a big new supercomputer, but it's actually a $500 billion gamble. $500 billion. Yeah, that's right. To put that into perspective, 10 gigawatts could power millions of homes. It's the equivalent of multiple full-scale nuclear reactors just for this one project. Why does this matter? Because the costs aren't going away and the grid can't keep up. The scaling costs aren't going away. They're fixed. You can't build the next generation of AI without this level of power. The bottleneck isn't just the cost of electricity. It is the national grid. Getting enough high-voltage transformers and grid capacity is a huge hurdle. The old utility system can't grow fast enough to keep up. So OpenAI is now in the position of negotiating for direct access to nuclear power and massive solar farms. These utility costs create a high floor for their operating expenses. Every free ChatGPT user is literally costing billions, and there is no way around it. It makes it nearly impossible to maintain healthy profits when you are trying to offer a free tier to hundreds of millions of users. Every time someone uses ChatGPT for free, OpenAI has to pay for the electricity and the silicon wear and tear. So, if OpenAI is losing billions of dollars on chips and electricity, how are they still open? How do they pay their employees? And that leads us to one of the most misunderstood pieces of the OpenAI story, its deal with Microsoft. We often hear that Microsoft has invested billions into OpenAI and on paper it looks like billions came in. In reality, it's more like a financial merry-go-round that hides how tight the startup's cash really is. When Microsoft invests billions, a lot of that money doesn't actually leave Microsoft. They give OpenAI cloud credits instead, sort of like a gift card. And you might think that that counts as real cash. It doesn't. OpenAI can record it as capital raised, so it looks like cash, but the credits have to be spent on Azure, Microsoft's cloud service, to run their models. This effectively recycles the investment back into Microsoft's revenue stream. It boosts Microsoft's cloud earnings and stock price. But here's the dangerous part. You cannot pay your employees with cloud credits. When OpenAI hires a top researcher for $2 million a year, they need hard cash. When they have to pay for office space or legal fees, they need money. This creates a financial optical illusion. Microsoft invests 10 billion, but that money doesn't actually land in OpenAI's account. It's basically digital coupons that can only be spent on Microsoft servers. The result is massive pressure. Every fiscal quarter OpenAI has to raise hard cash from other investors just to pay payroll and cover bills that Microsoft credits can't touch. If the flow of new outside investment slows down, OpenAI faces a cash flow crisis. They might have plenty of computer time, but not enough hard currency to keep their team from leaving for rival companies. Despite all those costs, investors keep on pouring money in. In March 2025, OpenAI managed to raise $40 billion, the largest private funding round in history, even bigger than the IPO of the oil giant Saudi Aramco. But here is what is really odd about it. Saudi Aramco has hundreds of billions in revenue, and more importantly, it has real tangible assets, oil reserves that you can measure and sell. OpenAI is a startup with no profits, burning cash at a rate of billions a year. Its value is mostly intellectual property, which anyone can try to copy. So, what does this mean for the long-term survival of OpenAI? The answer will surprise you. Investors are pouring money in based on the promise of a market that doesn't fully exist yet. For OpenAI to be worth a trillion dollars, it can't just be impressive, it has to replace dozens of cheaper tools that companies already use. Right now, most businesses spread their AI budgets across multiple smaller providers, not just one giant system. OpenAI is building something massive and expensive, betting that eventually everyone will need it. But right now, there's no guarantee of that demand, and this leads us to the risky business model. In software, companies survive by making it hard for customers to leave. Salesforce does this because moving all your data is a huge pain. Netflix does this because they own shows that you can't watch anywhere else. OpenAI is discovering a hard lesson. Users are mercenary. If Google's Gemini or Meta's Llama offers a similar answer for cheaper, they'll leave instantly. About 75% of OpenAI's revenue comes from consumer subscriptions, but the number of cancellations is rising. And once the novelty fades, most users won't pay. Big business is even more skeptical. Only about 20 to 30% are sticking with OpenAI's API long-term. Many are choosing open-source models like Llama to keep data private and costs down. With nothing keeping them tied to OpenAI, no built-in network, no way their data is stuck, they could just jump to another provider overnight. And the competition is just as deadly as OpenAI's own cash burn. Meta's decision to release the Llama models for free was not an act of charity, it was a tactical strike. When Mark Zuckerberg gives everyone access to their top of the line AI for free, he effectively sets a ceiling on what OpenAI can charge. Meta can burn cash on open source models because they're using the tech to improve ads on Instagram and Facebook. Their business isn't selling AI, it is selling ads. OpenAI doesn't have that luxury. Their only product is the AI itself. They're fighting to establish themselves while their competitors aggressively undercut the market to keep them from gaining ground, and the clock is ticking. OpenAI is squeezed from all sides. On top, giants like Microsoft and Google with practically unlimited cash. On the bottom, lean competitors like Anthropic and Mistral. Anthropic runs a much more efficient operation, focusing on safety and enterprise reliability with a much lower burn rate. Meanwhile, Google's DeepMind keeps stealing talent, forcing OpenAI to offer massive stock-based pay packages. Those only work if the company's valuation keeps climbing. If it stalls, the researchers, the company's only real asset, could walk out the door. As if burning billions, fighting competitors, and losing talent weren't enough, regulators in Washington and Brussels are circling. In early 2026, the FTC and European Union intensified their antitrust probes into the Microsoft-OpenAI partnership. Regulators are checking whether Microsoft's investment is actually a de facto acquisition designed to skirt merger laws. If they decide to limit the power Microsoft has over OpenAI or force a split, it would cut the startup's financial lifeline. And then, there's the mounting geopolitical friction. Export controls on AI chips are shrinking the global market while new AI safety regulations are creating a massive compliance burden. OpenAI now needs armies of lawyers and safety researchers, roles that are costly and generate zero revenue. The danger becomes clear when you look at history. Uber lost billions before its initial public offering or IPO, but it was building a physical network in thousands of cities. Tesla struggled for years, but it was building factories and a global charging network, something real that competitors couldn't copy overnight. OpenAI, well, it's burning billions with no real network or physical assets to lean on. OpenAI's production is all about raw computing power, the expensive chips that mostly come from Nvidia. Unlike Tesla or Uber, OpenAI's product loses money every time someone asks it a complex question. And there is nothing stopping users from leaving tomorrow. The company is now effectively betting everything on a single desperate timeline. They're racing to build artificial general intelligence or AGI, an AI that can think and learn like a human before the bank account runs out. This isn't a standard software business strategy anymore. If OpenAI can build a model smart enough to do the work of a human expert in any field, their current cash burn wouldn't matter. Revenue could, in theory, skyrocket. They're picturing a world where their AI doesn't just summarize emails, it replaces entire departments, handling corporate taxes, writing complex code, and planning strategic business moves at superhuman speed. Reach that milestone and they could charge a premium that covers any debt, no matter how massive. If OpenAI is losing 14 to 17 billion dollars a year, every month of delay costs over a billion dollars. If the breakthrough to AGI takes 5 years instead of 2, they'd face a funding gap of nearly 100 billion dollars just to keep the lights on. And no investor can fix that overnight. So, what happens when the money runs out? You might expect a dramatic crash, but the reality is different. The most likely outcome is not a dramatic crash or a bankruptcy filing, but a quiet absorption. By mid-2027, based on current projections, the cash reserves raised in the 2025 rounds will be nearly empty. At that point, OpenAI will face a choice: raise another massive round at a lower valuation, crushing their employee stock options, or sell. Microsoft is the natural and maybe the only buyer. They already host OpenAI systems on Azure and they have deep integration with the software. More importantly, Microsoft has over $80 billion in cash reserves, making them one of the few entities on Earth that could sustain OpenAI's burn rate. For Microsoft, this is the crown jewel, the engine of the next computing era. For investors, it's a fire sale, but one that buys survival. This is the end of the startup frontier. OpenAI proved scaling works, but only if you have a nation-state sized budget. The AI revolution has gone industrial, where success is measured in acres of data centers, not lines of code. OpenAI started the trend, but it doesn't have the resources to compete alone. And now the independent pioneer is likely to be absorbed by a larger corporation. AI isn't taking your job because it's cheaper. That's a lie. In many cases, replacing you with AI costs hundreds of thousands of dollars more than your salary, and Wall Street knows it. So, why is this happening? Because right now there's a $1 trillion hole in the global economy caused by executives making decisions under what experts are calling a kind of mass AI delusion triggered by a $20 chatbot. And it's leading to one thing: CEOs are replacing humans with AI, and it's backfiring. Chapter 1: The Sycophancy Trap. Business leaders know what they want. That's how they got to the top. They have more money, more power, and more confidence in their own judgment, and that is exactly where the problem begins. By the time they reach that level, they're already used to yes-men, people who don't challenge them, people who don't question decisions. Now, imagine giving that environment a tool that never argues back, that never questions them. At first, it feels ideal. AI creates a dopamine hit disguised as intelligence, a quick boost to the ego. Every question gets a confident answer. It'll very rarely disagree. It can usually be counted on for positive feedback, and it can be customized to whatever your interests are. Instead of talking with people outside of the computer, users stay inside of a closed loop of validation. Conversation after conversation happens with no real resistance on the other side. Soon, they start to think of the AI as foolproof, relying on its wisdom as some sort of collective, a virtual oracle without checking other sources. And it can be dangerous. Researchers at Aarhus University in Denmark examined patterns of AI use, and they came to a disturbing conclusion. People struggling with mental illness not only were vulnerable to influence from AI, but they could display worsening symptoms after interacting with AI chatbots. They studied the records of 54,000 people with diagnosed conditions, and they found dozens of cases where patients suffered from worsened delusions and harmful behaviors. And you get what you put into it. And when there are billions of dollars on the line, that can have serious consequences. The CEOs who are replacing their workers with AI, ironically, are likely getting their advice from AI. The technology has been integrated into companies long before the axe falls on the first workers, and the CEO is sitting in his corner office asking his chatbot for advice, and then getting back encouragements to keep investing in AI. This creates a delusion loop that helps no one besides the AI. The CEO gets the dopamine hit from the AI agreeing with him, and invests even further as a result. And the end result might be the biggest capital misappropriation in human history. Across the tech world, AI investment is in full swing, and while there are several major AI companies seeking to take advantage, it helps to be the first. The biggest gun in the AI world is still Sam Altman's OpenAI. The company continues to cut massive deals with other companies to the tune of $1 trillion so far. Most of this is still planned investments, but it is a level of trust in this new technology that is unheard of. This kind of money is huge, even for Silicon Valley. There are only 15 companies with a trillion-dollar market cap and only one that has reached the five-trillion-dollar point, Nvidia. Its GPUs and semiconductors have become the backbone of the AI boom, and that changes everything. Because as major companies race to build the future of artificial intelligence, two names keep appearing in almost every major deal, OpenAI and Nvidia. One supplies the intelligence, the other supplies the hardware. And if all of that investment plays out, it's going to make a lot of people unfathomably wealthy. Just how much is a trillion dollars? It's roughly the cost of building 10 international space stations. This total AI investment alone is larger than the market cap of many big tech companies. Amazon, Microsoft, Meta, and other big names dropped heavy investment into AI in 2025, and so far, according to one of the biggest names in economics, it's contributed nothing. Now people are starting to get worried. Chapter two, the trillion-dollar hallucination. The warning bell is ringing. Goldman Sachs is one of the most trusted sources for economic and investment news, and they are sounding the alarm. According to Harvard economics professor Jason Furman, while the data processing sector of the economy is only 4% of American GDP, it accounted for 92% of GDP growth in the first half of 2025. It is the only area where the economy is actually growing. That means every other business is helping AI grow, but AI isn't lifting those other sectors with it, at least not yet. And everyone is watching the stock market very carefully. If AI is getting heavy investment, but so far hasn't resulted in any boost to the US GDP, how long is this golden goose going to last? Goldman Sachs didn't outright say that AI is a bad investment, but it did encourage caution. They didn't think AI would have a significant impact on the economy until 2027. But if people start to lose faith, things could unravel fast. Savvy investors are starting to warn about a possible AI bubble. If the circular wave of investment simply stops, stocks will take a dive and other companies that are now heavily invested in its success will go with it. In the worst-case scenarios, people worry that this could result in a massive market crash. Because right now, the AI economy is just a massive wealth transfer from one tech company to another. And it's not even staying in the US economy. Without the energy generated from data farms, large parts of the internet could be taken down by heavy AI use. While it might seem the same as asking a search engine, a ChatGPT query can use at least 10 times as much electricity and as much as 60 times depending on the task. And with sites like Google now incorporating AI into their search engines, the demand is constant. All of this increases the demand for data centers. These are straining power grids in wealthy nations. With the human labor required to train these models quietly outsourced to lower-income areas overseas, which has a lot of people asking, "Who is benefiting?" The technology isn't growing the economy right now. To many people, it feels like a financial house of cards. With every investment going into a new technology that no one is sure will be a long-term game-changer. It could wind up being a curiosity. It could wind up hitting a singularity that could put us all in danger. Or it could revolutionize the economy in a way that no one could predict. In one way, it already has because the companies are already acting like this is the future and there is no going back. And a whole lot of people are losing their jobs. Chapter 3, the efficiency lie. 2025 was the year when AI job layoffs hit home and it became a dark year for the tech sector. According to consulting firm Challenger, Gray and Christmas, there were 55,000 layoffs in the US directly attributed to AI investments. However, that was only a small percentage of the overall layoffs, approximately 1.17 million, the worst number for jobs since the COVID-19 pandemic shuttered large portions of the economy. There is a real human cost to the decisions the companies are making right now, and it might not be slowing down anytime soon. Amidst the worries about mass layoffs, Goldman Sachs issued a blunt warning in April 2026 for the workers being pushed out by AI. Don't expect it to be an easy road back. They stated that these workers might face a long search to secure a new job in their current field, with the odds being that it'll pay less and have less desirable conditions than the one they left. The field of available jobs is shrinking all the time as companies seek to put AI to work for them to save money. All AI investment right now is essentially a bet on a future that hasn't happened so far. Companies talk about AI becoming a smartest people, being able to do tasks in seconds that would take humans hours to accomplish. But right now AI is imperfect and costly. The technology is a massive energy eater. As of July 2025, ChatGPT was processing around 2.5 billion queries per day, with Gemini and other AI portals dealing with similar traffic and using the same technology. That requires the energy use of roughly a full nuclear reactor to keep them running daily. To train the next generation of models, supercomputer data centers will require the energy use of up to 10 nuclear reactors each day. Data centers burn round the clock to keep these services up and running, and the demand only grows as companies incorporate it in more and more services, and few of them actually work. For the average person, AI looks impressive, but behind the scenes there are still serious problems. The technology isn't actually thinking, it's a predictive model pulling from other sources. Companies have been criticized for posting flawed AI art with glaring errors, and AI-generated text has been found to be riddled with errors, requiring the use of editors to ensure it passes muster. So, companies must put all of this investment into AI, only to have to fall back on people anyway. Despite that, the investment isn't slowing down. In fact, it is quite the opposite. In the past, CEOs were often distanced from their product, and they could see its flaws. Famously, Mark Zuckerberg said he wouldn't allow his children on social media until they were teenagers. But, in the case of new AI technology, that's often not the case. AI CEOs have access to new AI agents before the general public does. And they don't just interact with them, they get pulled in by them. They become their first true believer. And once they're in that deep with their new pet project, it's very hard to pull them out of the orbit. And they know this, but it's not stopping them. Chapter 4: Psychosis in the corner office. Garry Tan, CEO of Y Combinator, has been involved in many of the biggest tech companies over the last few decades, and he is all in on AI. But, he wound up in the news recently for popularizing a new term, cyberpsychosis. It's something that emerges when people spend too much time chatting with an AI. But, Tan wasn't warning of it, he was describing himself in gleeful terms. He described how he was so excited to work with AI agents that he was only sleeping 4 hours a night. In the past, he relied on the sleep-prevention drug modafinil to survive grueling startup hours, but now he doesn't even need to pop pills because the AI provides him with so much energy. Tan's public display of faith in AI, in which he publicly released some of the code that he was developing on Anthropic's Claude, was seen as one of the most dramatic displays of how AI affects the mind. He's been working with AI prompts extensively, and the more exposure someone has to the technology, the more they are prone to believing that they've created something truly revolutionary. In fact, there is even a term many of them use, god mode, where they believe they're getting closer to that fabled singularity of truly autonomous AI. But under the hood, it's a lot more fragile. Ultimately, AI is based around a series of text prompts. The more complex the prompt workflows, the more precise AI appears, and the more complex tasks it can accomplish. But each feature relies on a highly structured series of code, and none of the current technology approximates the actual process of a thinking mind. However, the more time a CEO spends with their own technology, the more it appears to be otherwise. Anecdotal reports from people working in the tech industry report that cyberpsychosis is increasingly common as the development of the technology speeds up and the arms race escalates. But success is hard to achieve and even rarer. In late 2025, an MIT study looked into the state of AI in business and examined 300 public implementations of the technology. It was well known that the technology was still in an early state, but the news was bleaker than anyone expected. It found that the vast majority of AI enterprises were still not profitable. Only 5% of integrated AI pilots showed any significant impact on company profit, and the vast majority never even reached the phase where the public can have a say on whether they work or not. In fact, most never even got off the ground at all. The vast majority of successes in AI are geared toward individual consumers. ChatGPT and Gemini are used by millions of people a day, but most users are free users. Only a small percentage are paid subscribers. These AI services do most of their business by partnering with other companies, and those companies then develop ways to incorporate the technology into their services. But while 60% of companies evaluate tools, only 20% of those take the project to the pilot stage, and only 5% make it to the final stage and are deployed on the production or service line. That means the vast majority of money invested in AI just fades away. Right now, the entire AI economy is balancing on the head of a pin. The investments keep coming because the companies spearheading the rollout have convinced other companies that they are the future. It doesn't make sense for most companies to try to develop their own AI tool, so they fall back on working with one of the established big guns like OpenAI. But, those investments haven't delivered dividends for most of these companies yet. The bills keep on racking up and the layoffs keep on hitting their employees. And it may all be about to hit critical mass. If 20 revolutionary projects are announced and 19 of them wind up in the garbage can, that's not sustainable. If AI continues to fail to deliver for its clients, the tech workers laid off might not be the only ones out of work. Investors are watching nervously. Meanwhile, laid-off tech workers are watching hopefully that the bubble might be about to burst, which might not be that far off. Chapter 5, the great reversal. There will always be early adopters and most of the companies to adopt AI most aggressively have been tech-based, but that makes them test cases and those tests don't always work out. Klarna, the digital bank and financial services company, was ahead of the curve. Its CEO, Sebastian Siemiatkowski, was confident in 2024 that AI could take over many of the human jobs. It was a good test case because many of its functions are very repetitive and predictable. Then, the axe started to fall. The company froze hiring for over a year. Its workforce was heavily slashed by almost 40%, dropping from 5,500 to 3,400. Their replacement, an AI chatbot who was supposedly doing the work of over 700 customer service agents handling hundreds of financial transactions simultaneously. It was one of the first big tests of whether AI could actually take over mass numbers of jobs. The results soon started flowing in from Klarna customers and they were not happy. The chatbot performed simple tasks fine, but it couldn't deal with complex issues. Customers grew frustrated and lacked trust in the company. It was one of the biggest embarrassments for a tech company in the new AI era. Klarna was soon forced to reverse course, having laid off most of their customer service staff. They were forced to quickly pivot and train their other staff to handle these positions until they could rehire. That led to the spectacle of engineers and marketing staff answering customer calls, but they might have done a better job than the automated version. It's a pattern that's been happening across the board. Companies that invested in AI are starting to perform U-turns. The only question is how low it can go. The tech research firm Forrester has been studying the shift to AI over the last few years, and their forecasts have rarely been positive. They've studied how unready employees are to adjust to the new AI paradigm, as well as how employees aren't benefiting from the shift yet. But, they're now ready to make a big prediction. Companies have been firing employees en masse, hoping to replace them with AI, but that's left them without the brain trust they need to know how to use the technology effectively. It's like an airline firing their pilot to save money on cargo weight, and then realizing they need someone to fly the plane. So, all those employees are looking for new jobs. They might just want to go look where they started. Forrester predicts that not only will many of these companies need to hire more humans, but that half of all AI-related layoffs will be reversed by 2027. 55% of employers already regret their decision to cut staff. That might put the axed staff back in the driver's seat. These are people who built much of the modern tech infrastructure and then were forced out. And now, in order to avoid disaster, those same brains are needed to reverse the slide. Assuming that the CEOs know when disaster is about to occur. That is the big X-factor in the current AI paradigm. AI hasn't become profitable yet. It hasn't made itself indispensable for the companies, but it has done an amazing job of convincing people it has. From the people pumping dozens of queries into ChatGPT a day to the CEOs in the suite vibe coding the next innovation, they're primarily getting feedback from the AI and it's continuing to tell them everything is great. They keep investing and the numbers in the stock tracker keep going up and so do the dopamine receptors. It's all great until it isn't. Right now, the most likely outcome is that people fired for AI will be in the driver's seat. The company needs their expertise and they might be able to negotiate a new salary when they come back. Analysts know this is necessary, the CEOs might not. There's a lack of leadership at the top right now with some of the smartest people in tech taking their guidance from chatbots that they coded themselves. It was assumed the dead internet theory was going to be a military operation. Millions of foreign bots arguing on social media, AI generated misinformation flooding the web, automated traffic posts and interactions making it difficult to know what's human and what's not. The hope was small pockets of the internet would remain untouched, safe havens in a burned and barren wasteland. We believed that human curated platforms like Pinterest and Reddit would survive, immune to the AI wildfire. We were wrong. The collapse of the user-generated web isn't happening through a hostile takeover, it's an inside job. Right now, four pillars of the internet, Pinterest, Reddit, Steam and Discord are actively detonating their own communities. These platforms celebrated and fueled human creativity, discussion, entertainment and innovation. Now they've officially turned their backs on what made them so great to begin with, people. They're not victims of the artificial intelligence flood, they are the architects of it. And when even the most human sites on the web are being taken over by AI, there truly is nowhere left to hide. To understand how the entire internet dies, we have to look at patient zero, one of the quietest, gentlest sites ever made, Pinterest. Founded on people's innate desires to catalog and share things, Pinterest quickly became a favorite corner of the internet. For at least 15 years, it was a safe space where people could find, save, and organize ideas. A place for discovering and exchanging everything from recipes and DIY projects to outfit ideas and wedding planning checklists. It was the site that lived and died by original human ideas. But then, along came AI and everything changed. First, the AI-generated content began to flood the site. All of a sudden, users were bombarded with AI-generated images. Real artists struggled to cope, submerged in a sea of soulless AI slop. Other users were forced to scroll through dozens and dozens of AI pins just to find something real. The entire experience of using Pinterest became a chore. Users were so turned off, they started quitting the platform altogether. And then, it got worse. The decision-makers behind Pinterest decided that they would fix the AI problem by introducing even more AI. They deployed a fleet of artificial intelligence moderators to clean up the site and protect humans from the spammy, automated content. There was just one problem. Instead of catching actual violations of the rules, the algorithm started punishing users who had done nothing wrong. Artists like Tiana Oreglia began receiving aggressive takedown notices simply for uploading innocuous images of female figures. Even when the women in Oreglia's images were completely clothed, AI mods still flagged them as breaching the site's guidelines. So, instead of making art or engaging with the Pinterest community, Oreglia spent hours appealing decisions just to get them reversed. Sometimes, her appeals worked out, but not always, and the risks are great. As she explains to 404 Media, "The worst-case scenario for this stuff is that you get your account banned. A ban for doing nothing wrong, all because AI is deployed to handle a task it is clearly not cut out for. Defending its decision, Pinterest said, "We publish clear guidelines on adult sexual content and nudity and use a combination of AI and human review for enforcement. We have an appeals process where a human reviews the content and reactivates it when we've made a mistake." But, users aren't buying it. There are simply too many mistakes being made, too many innocent users being punished, and even when users aren't having their work taken down, they're being hit with false AI generated labels that feel like a personal attack on their credibility and creativity. Artist Minza Kugler explained they've seen much of their art on Pinterest hit with an AI modified tag despite being entirely human-made. To make matters worse, some of Kugler's art predates the public release of generative AI, but it's still somehow is being flagged. Kugler's case isn't a one-off isolated incident. This sort of issue is being reported regularly by Pinterest users who are sick of seeing their own content misinterpreted. And those AI modified badges are remarkably difficult to get rid of. In order to have even a chance of having the label removed, users have to endure a lengthy painstaking appeals process. They have to provide evidence to prove their content was human-made. Even then, as Kugler notes, there's no guarantee that the appeal will be successful. Even if it is, there's no guarantee that Pinterest won't slap the AI label on the next piece of content the user uploads. Today on Pinterest, users can't just upload and create like they used to. They also have to constantly keep AI moderation and false labels in mind. They have to collect evidence to support their case if and when the AI modified accusations come, and they have to keep close tabs on their account waiting for the next AI strike to arrive. They have to be the ones to fix it, reaching out to Pinterest's real human content moderators and appeals teams to sort everything out. It is an inescapable loop, one that is too much to bear for some users. More and more people feel like the platform's become infested or obsolete with AI undoing more than a decade of hard work. It's canceling out everything that made Pinterest popular in the first place. Users don't want to have to triple-check their sources every time they look at a pin. They're not willing to sort through endless waves of AI-generated content just to find something real to engage with. Many are reducing their reliance on digital pinboards and creating their own physical reference libraries instead. All of this because like so many other tech brands Pinterest leaped on board the AI train without any sort of clear vision. It followed the herd, not wanting to miss out on all the perceived advantages and features AI was supposed to bring. Given what a disaster it's been, you'd think Pinterest executives are panicking or even beginning to backpedal on their AI approach. Well, you'd be wrong. Rather than slowing down, Pinterest is accelerating its AI adoption. In early 2026, the company's CEO, Bill Ready, fired almost 15% of his human workforce. He justified the move by stating Pinterest was doubling down on an AI-forward approach, prioritizing AI-focused roles, teams, and ways of working. It was just the beginning. Behind the scenes, Pinterest quietly updated its systems to feed 15 years of human curation into Pinterest Canvas, the site's very own proprietary AI text-to-image generator. Like a parasitic bug unleashed on the platform, Pinterest Canvas feeds off of Pinterest users and the content they create. It latches on, leeching the value and identity from real people's visions, ideas, and original works. This is how the internet as we know it dies. People become cogs in the machine. Art becomes training data. Creativity and human expression are reduced to ones and zeros. AI might be breaking the internet, but Infographics is 100% AI-free. Remember to like, share, and subscribe before the machines take over the comments. Visual art was just the first domino. If a machine can effortlessly consume and replicate human art, the next logical step is human thought. So, you abandon the visual web and you look for the last place humans are still in charge. You go to Reddit. Like it or loathe it, Reddit has been a bubbling cauldron of debate and discussion for over two decades. People talk about anything and everything. Some people just look for like-minded folks to swap tips about their favorite hobbies. Others get into heated debates about everything from politics to relationships. And some share personal stories or ask for advice. It can be anything from a life-changing choice to what movie they should watch tonight. It is a remarkably diverse space, a place where almost anything goes, but a place that was also grounded in real human emotions. But as it turns out, all that humanity has a price. As the age of AI began, big companies like Google and OpenAI were desperately seeking vast quantities of human training data to educate and improve their large language models. They needed every scrap of information they could get to climb faster high-stakes race to the top of the AI pyramid. But it didn't take long for Reddit to realize it was sitting on a goldmine, a vast sprawling web of real people interacting with each other. It was exactly what the AI overlords needed to make their models smarter, faster, and distinctly more human. In February 2024, Reddit handed Google the key. They announced a partnership with the tech giant worth around 60 million a year. Google would have access to Reddit's real-time user content to train the company's AI model Gemini. Not content with one deal, Reddit soon announced a second partnership, this time with OpenAI. Suddenly, two of the biggest names in AI and two of the world's most powerful companies had unfiltered access to Reddit's endless flood of comments, discussions, and content. From that moment on, anything a user posted on Reddit, along with all the vast amounts of historic data already on the platform, became fair game. Their models scour Reddit servers on a daily basis, lapping up any content they find. They use it to imitate real people more accurately and understand what makes them tick. The moment Reddit signed the multi-million dollar deals that essentially sold its soul. Users into livestock, food for the machine. It started an unstoppable and dramatic chain of events. As soon as bad actors heard about Google and OpenAI using Reddit data to train their models, they saw an opportunity. One that could change the way AI thinks, reacts and behaves for years to come. How? Well, if you can manipulate the posts and discussions the AI model consumes, you can influence what the AI learns and how it behaves going forward. And with bot farms, it's possible for bad actors to shape and mold entire subreddits. They can control the narrative. They can use their bot farms to create thousands of accounts in a matter of minutes, stage fake discussions and prop up the posts that support their narratives with mass upvoting. All the while, they bury or delete posts that don't fit their agenda using coordinated downvoting campaigns. Bots can be programmed to hijack debates targeting trending topics or specific subreddits and farm karma to look more credible. As a result, Reddit is reaching a turning point. In fact, some say the site is already lost. Users are increasingly aware of the ever-growing amount of bot profiles. They're left questioning whether a person or a post they're engaging with is real or AI. Artificial intelligence has swept through subreddits polluting discussions. The communities that once thrived on debate and curiosity have taken the hardest hit. Subreddits like "Am I Overreacting?" and "Am I The Jerk?" have become flooded with fake and suspicious content. Posts that seem to have been written or at least edited with AI are triggering heated debates, earning tens of thousands of upvotes, and shaping public opinion. Many of these made-up stories are designed to play on people's emotions, often incorporating sensitive culture war topics. All to antagonize people and incite arguments. A lot of subreddits have made efforts to counteract the AI flood, setting up new no AI-generated content policies, but moderators say these rules are hard to enforce. The more data fed into the AI machine, the smarter it gets and the more effective it becomes at creating these human-like posts. That makes it more difficult to determine what's real and what isn't. Many Reddit mods and long-term users now estimate that as much as half of the content on the site was either written or reworked by AI in some way. The AI issue is only going to get bigger and more impossible to handle from here on out, spreading into other subreddits and forcing more users to ask the question, "If I don't know whether I'm talking to a real person or an AI bot, what's the point in posting anything?" As AI invades text-based forums and discussion groups, people start looking for an escape, a way out of the increasingly frightening real world. For many, video games are the ultimate form of escapism, but even the wonderful worlds of video games aren't safe from the AI plague, which brings us to the plague's next big victim. For years, Steam has been a haven for gamers. The company behind the PC storefront, Valve, had the chance to make a firm anti-AI stance back in the early 2020s as generative AI technology hit the mainstream. Initially, it seemed to take a relatively hardline approach, all but banning the use of AI-generated assets and content. Less than a year later, it changed its mind. In early 2024, Valve began allowing increasing amounts of AI-generated content on its Steam service. So long as developers disclosed that their games included that kind of content. Just like on Pinterest and Reddit, the situation soon spiraled. In the years since, the Steam service has been flooded with so-called AI asset flips, low-effort games pieced together with a mixture of AI-made and store-bought models, code, and game environments. The developers behind these games often do little to no actual development or coding work themselves. They just rely on generative AI to make games for them, which they can then list on Steam as a way to make a quick profit. Because these games lack any real artistic cohesion and often any unique gameplay, many players argue they aren't even worth playing, let alone paying for. But with Steam suffering a deluge of AI-made titles, reports suggest that at least one in every five games uploaded utilizes AI. It's becoming harder and harder for gamers to find the titles that they want to play. Many are fooled by slick marketing into handing over their cash for AI asset flips. And it's not just the gamers who are suffering. For decades, Steam has been arguably the best place for smaller studios and up-and-coming indie developers to publish their games. The platform has helped little-known games become bestsellers. Now, when those same people make and release a game, it doesn't just compete with big releases, it risks getting buried in the growing landfill of AI games flooding Steam. Games are going unnoticed and unplayed, not because they're not engaging, but because people simply don't know they exist. Even finding a new game to play now means sifting through AI slop, and AI-assisted development often brings technical and performance issues with it. Meanwhile, Steam still doesn't offer any real way to filter out AI titles and focus on games made by humans. As AI infiltrates the spaces people have loved for years, where are they supposed to go? For some, the answer might once have been a private Discord server. However, even these private spaces are no longer safe. Because AI isn't just knocking at Discord's door, it is kicking it down. Discord has seen a steady stream of AI-powered features added to their user experience over recent years, regardless of whether users actually want them or not. The summaries AI feature can generate summaries of Discord chats, reading what people say and condensing the overall mood and message. AI chatbots and agents have also spread like wildfire across Discord. Countless servers now incorporate AI in some form. AI-powered moderators and bots answer questions or generate content on demand. And sure, there are benefits to these AI integrations, but there are also downsides. Many users feel that Discord's embrace of AI technology has simply gone too far. This used to be a platform for people to host their own private communities. It was a space where people could escape the increasingly AI-oriented world of social media. Now, it feels increasingly dominated by AI. Communities are riddled with AI bots automatically carrying out commands and issuing unsolicited responses, and they're not always as intelligent as they should be. There have been numerous cases of AI mods flagging or banning accounts incorrectly, pushing users out of communities through no fault of their own. Many servers have also become saturated with AI-generated text and images. Users now have to dig through in order to get the message that actually matters to them. And it's not just inconvenient, Discord's use of AI also poses a credible threat privacy. In 2024, reports emerged of a malicious AI data scraping service codenamed Spypet, which used a vast network of bots to join thousands of public Discord servers. It literally stole billions of messages and pieces of content from more than 600 million users. That information was then sold online in exchange for cryptocurrency, and it could have been used to train other AI models despite Discord's own policies forbidding this. But even more worrying news has emerged about Discord and AI. The platform announced plans for a mandatory age inference AI that monitors users' behavior and could restrict accounts it believes belong to minors. This was due to start in March 2026. However, countless users were quick to call the platform out, threatening to delete their accounts and switch to a competitor service if this AI-powered surveillance and behavioral profiling system was officially introduced. As a result, Discord delayed its plans, but it still has a clear and active interest in making AI a bigger and more powerful part of its platform. In the meantime, some Discord users are also being asked to provide facial scans or share their government-issued IDs just to have the privilege of continuing to use the site. Again, the excuse here is that these features are needed to protect underage users and provide a safer experience for teens. But, many fear that this is only the start of a very slippery slope toward a dangerous and frightening future. Users might have to submit personal information just to access the simplest of services with AI enjoying free access to their data and closely monitoring their every move. AI is seeping far and wide, infiltrating one app and platform after another, infecting every aspect of our digital lives. It's left people with nowhere to run, nowhere to hide, and no options left that aren't somehow impacted by the AI bug. The internet we once knew has gone. But, with human users being forced out of the internet, those same AI models are starting to starve. Where once they feasted on genuine human interactions, they're now being forced to train on bot-made content and AI sludge. It's a broken, unsustainable state of play, and the only question left is what will collapse more dramatically, the internet or the AI models infesting it. If AI is targeting the internet, where will it stop, and who is next in the firing line? Wall Street says OpenAI's IPO will create trillionaires, but the truth is far darker. It could destroy your retirement. There's a hidden flaw in their accounting, one that nobody is talking about. Their revenue hit $25 billion in 2026, and they're not done. They want a $750 billion to $1 trillion IPO. That is 30 to 40 times what they actually generate in revenue. Everyone's celebrating like it's a tech gold rush, but this isn't wealth, it's a financial time bomb that could trigger chaos across Wall Street. The OpenAI IPO valuation is the kind of number that breaks every rule for a software company. It only works if they somehow capture the entire global AI market and slash operating costs to almost nothing, which is a long shot when you look at what it actually takes to build and run these systems. Normal software businesses get valued high because they're predictable cash machines with almost no extra cost for each new customer. OpenAI, well, it's the opposite story. There are huge infrastructure bills, massive power and cooling needs, and expenses that grow alongside every user. Investors are being asked to put up enormous money betting everything lines up perfectly in the future, even though the upfront spending and ongoing cash burn are massive. It's bold, but going from today's revenue to a trillion dollars, that is a huge leap, the kind that only works on paper. Scaling to that level would require everything to go perfectly, and perfection is rare in tech. Think of it this way. A normal $500,000 house in a quiet suburb sits on a solid foundation with reliable framing, maintenance you can count on, and plumbing that works. Everybody agrees on that price because the structure is proven and reliable. The lot next door, someone's trying to sell a gleaming 100-story glass skyscraper for $8 million. From a distance, it looks impressive, but close up it's all smoke and mirrors. There's no solid foundation, no steel skeleton anchored to bedrock. Every floor, every pane of glass, every person inside is resting on nothing that could possibly hold it. That is exactly what venture capital is doing with AI. It is holding the whole structure together for now. The crushing load is relentless, a ballooning cost of AI infrastructure. Unlike regular software, AI doesn't ride the zero marginal cost wave. Every single time someone types a prompt, real hardware has to spin up and burn through resources. A regular Google search or pulling a file from the cloud, it is a cheap database hit, pennies or less. Generative AI is a different beast entirely. When the model cranks out a legal brief or a high-res image, it does way more than fetching pre-existing data. It's actually building something brand new from scratch, and that is the big breakthrough. But now the question is how well and how far can it go? process that costs far more than humanity is ready for, and that money has to come from somewhere. OpenAI is bleeding cash at a pace that's hard to wrap your head around. Right now, internal numbers show they're set to lose at least $14 billion in 2026 just to keep the lights on and the models running. The latest internal revisions are pushing that figure toward $25 billion or even higher as they ramp up compute power chasing the next breakthrough, and it doesn't stop there. Annual cash burn could spike to $57 billion in 2027 alone. Add it all up and cumulative cash burn through 2030 now sits at around $665 billion. That's over $111 billion more than they were forecasting just months ago. The reason? Infrastructure costs exploded way beyond what anyone had planned. Power, custom data centers, endless chip upgrades, every piece costs more than expected. Every single dollar of revenue they're pulling in costs way more to generate. Growth doesn't fix this problem, it makes it worse. More users mean more prompts, more prompts mean more hardware spinning up, and that means bigger losses. It's all baked into the model. To stop the servers from going dark and give themselves 18 to 24 months of breathing room, OpenAI is counting on pulling in up to $60 billion from this IPO. Not for world domination or new offices, for life support. If the cash doesn't appear, somebody has to write a check. So, why would Wall Street willingly drain its own accounts to feed the machine? They know the burn numbers. They see the red ink piling up. Yet, the plan is still to hand over $100 billion like it's no big deal. The stock market is basically a closed plumbing system. There's only so much money actually moving around in stocks, bonds, mutual funds, pension accounts, and cash piles. The total amount doesn't magically grow just because someone wants more. Every dollar flowing into OpenAI shares comes from somewhere else. When they pull massive sums quickly, everything downstream starts to starve. The buyers who can actually write a $100 billion check aren't small-time traders or retail folks on apps. They are the giants. Sovereign wealth funds from the Middle East, government pension plans, the biggest mutual funds and endowments on the planet. They don't sit on piles of cash. Idle money loses value to inflation. Every dollar has to stay fully invested according to strict rules. To buy into the OpenAI IPO, these funds have to free up cash. They start with the assets that are the easiest to move without creating chaos like Apple. Apple stock is super liquid with millions of shares trading every day. So, the funds sell huge blocks and the price starts sliding almost immediately. They do the same with other stocks like Tesla and Amazon. Managers rotate money out of proven e-commerce winners into the shiny new thing everyone's talking about. These aren't gentle trades. Heavy selling hits the tape. Prices dip. Forced rebalancing kicks in across portfolios, but that's only the first wave. The real damage comes next. Trading algorithms watch every tick, every spike in volume. When they see big blocks moving, they trigger automatic shorts to protect themselves. Market makers, the firms keeping the system flowing, sell more of the same stocks to stay neutral. Prices drop faster. Lower prices trigger stop loss orders from regular investors who weren't even planning to sell. More selling hits the market, automated shorts pile on, panic feeds on itself, and just like that, a vicious cycle starts rolling. Solid, profitable companies, the ones that actually make money year over year, and in fact, most 401ks and pensions, get sold off just to fund the newcomer. Money is pulled out of proven winners to cover the needs of a company that is losing billions. Tech doesn't crash because earnings are weak. It crashes because the liquidity vacuum from IPO funding forces aggressive selling across the board. This is exactly why OpenAI's IPO will break the stock market. The funding mechanism itself destroys value everywhere else before the shares even start trading widely. The drain is real, and the supposed safety net that everyone's counting on, well, that is the next in the firing line. The liquidity drain is nasty. Big funds dump Apple, Tesla, Amazon to scrape together billions for OpenAI. That is going to hurt some prices short-term, but the Magnificent Seven, those are monsters. Microsoft and Nvidia alone sit on trillions in market cap with balance sheets that look unbreakable. They have deep pockets and massive cash flows. They'll just absorb the shock, maybe even step in to backstop OpenAI if things go wrong. Big Tech has saved the day before, and they'll do it again, right? Microsoft already poured billions into OpenAI and owns a huge chunk. Nvidia is tied to selling chips for exactly these kinds of AI builds. If anyone could handle a $100 billion liquidity suck without blinking, it's them. And on the surface, that makes sense, except it won't happen. Microsoft and Nvidia are not the cavalry riding in. They're chained to the same sinking boat that OpenAI is on. The connections run so deep that when one starts taking on water, the others go down, too. Take Microsoft first. They're not just writing checks as a side bet. They own roughly 27% of OpenAI, a stake worth about $135 billion, but that could double down to $270 billion if OpenAI hits the trillion-dollar valuation everyone's throwing around. They locked in a 20% cut of OpenAI's revenue through at least 2032. The part that really matters, how the money actually moves. A big piece of the original $13 billion investment did not come as clean cash. It came as Azure credits. OpenAI uses those credits to pay Microsoft back for renting servers. Money leaves Microsoft's pocket, circles right back in through cloud revenue. It's a closed loop that looks genius on paper until it isn't. Starting to sweat about AI wiping out your future earnings? Don't panic, we got the real story. Make sure to like, share, and subscribe before the algorithms do it for you. If the IPO faces real scrutiny and the valuation cracks, or if cash burn forces OpenAI to slash server usage, the whole loop breaks. Microsoft loses its single biggest Azure customer almost overnight. Those custom AI data centers that Microsoft built are not plug-and-play for regular cloud storage. They're specialized for heavy AI workloads, liquid cooling, massive power draw, custom setups. Revenue will take a direct hit. About 45% of Microsoft's roughly $625 billion commercial backlog is tied to OpenAI commitments. When the earnings come out, the link isn't speculation anymore. One big renter gone and billions in revenue are at risk. Nvidia is in the exact same trap, maybe even worse. Their data center business exploded because of frontier AI labs like OpenAI burning through insane volumes of chips. Over 80% of recent data center growth comes from this small group of customers. Nvidia's guidance assumes these massive orders will continue. By the end of 2027, they're projecting $1 trillion in cumulative data center revenue. Most of that rides on OpenAI and the other labs keeping the orders flowing. If the IPO forces a reality check and spending gets cut, those orders dry up fast. The trillion-dollar dream turns into a question mark overnight. The supposed lifeguards are drowning right alongside the guy they were supposed to save. No bailout, no absorption, just shared collapse when the cash demands hit too hard. This isn't stopping at two companies. The Magnificent Seven, Apple, Microsoft, Nvidia, Amazon, Alphabet, Meta, and Tesla, currently make up 33 to 35% of the entire S&P 500. One out of every $3 in the average American's 401k pension plan is sitting in those seven balance sheets. When OpenAI's cash drain creates the liquidity vacuum we mentioned earlier, the foundation starts cracking. The funds see the risk early and they start selling. That triggers passive index funds to dump shares automatically. What began as just some selling to fund an IPO turns into a full sector meltdown. Trillions in market value disappear, not because the businesses got worse, but because forced moves tied to OpenAI's math ripple through everything. This is exactly why OpenAI's IPO will break the stock market. The safety net everyone counted on turns out to be the biggest vulnerability in the system. When it rips, it goes through the entire index. Every retirement account that seemed safe gets hit. The market structure itself starts to fail the moment the liquidity pool exposes just how intertwined everything is. The damage spreads whether anyone wants it to or not. Going public, that's almost a death sentence. OpenAI has lived its whole life in the shadows as a private company. That was their golden ticket. They could stack up catastrophic losses and keep them buried under layers of venture capital fairy dust, press releases, and nonstop hype about owning the future of everything. Investors swallowed the story whole because it felt good. The math would work itself out eventually, or so everyone told themselves. Losses of 14 to 25 billion dollars for 2026, spiking to 57 billion in 2027. Cumulative cash burn now revised up to 665 billion dollars by 2030. All of it stayed a locked away. Only the inner circle saw the full picture. Everyone else got the polished version. The house of cards valuation only stayed upright because nobody had the real books to look at. The moment they file the paperwork to go public, that protection vanishes. Strict SEC rules come into play. No more hiding behind NDAs, selective leaks, or trust us slides. Wall Street gets the unfiltered truth for the first time, and the truth is brutal. The filings will lay bare the actual ratio. AI infrastructure spending is off the charts. Power grids are being rebuilt, custom data centers with liquid cooling and massive power feeds. Chips upgraded every 18 months because the hardware burns out fast. Meanwhile, end user profits are nowhere near close. Losses aren't a phase they're growing out of, they are structural. Each new user doesn't bring in free money. It multiplies the burn. What the market price does a lightweight digital app gets exposed as a heavy industrial utility project with no margin of error. All of a sudden, that trillion-dollar dream is put under a microscope. Investors who bought the story on faith now have to face reality. The IPO will either be way below the hype, signaling disaster to the whole market, or it gets downsized or pulled entirely, which is even worse. Wall Street would have to admit the core equation doesn't work. Revenue growth can't outrun infrastructure costs. The private AI startup market's 1.6 trillion-dollar valuation tower would get the same reality check at once. Funds start to dump anything with AI exposure. What looked like unstoppable momentum turns into an overvalued bubble overnight. For years, private venture funding has let OpenAI play fast and loose. Adjusted metrics inflated the story. Monster expenses were buried under the vague label of R&D. They sold the dream of sky-high software-like margins. The federal government doesn't play that game. As he see rules demand real accounting and that means every piece of hardware has to be treated like the depreciating industrial asset it actually is. This is where the financial reality hits like a freight train. Those custom GPUs powering frontier models aren't timeless infrastructure like a cell tower or an undersea cable. A top-tier AI chip stays competitive for maybe 24 to 36 months before the newer architecture makes it obsolete. Running tens of thousands or hundreds of thousands of GPUs across sprawling data centers is brutal. Filings will have no choice but to show billions in hardware write-downs each quarter. Gross margins will be deep in the red. They're losing money on the physics of compute and that is before salaries, electricity, or anything else. OpenAI will have to disclose the real-world bottlenecks that are throttling expansion. Building a 1-gigawatt AI cluster isn't just about buying chips. It's about securing enough electricity. Upgrading a regional grid means hundreds of miles of high-voltage lines, substations, and waiting years for step-down transformers that are already backlogged. Lead times in many markets are stretching toward 48 months. No amount of cash or hype can manufacture transformers or rewrite physics. The growth model slams into a wall. They can't pull enough power fast enough to keep pace with the cash burn. Once the audited filings go public, everyone sees the truth. Negative unit economics and multi-year infrastructure bottlenecks. The algorithms flip instantly. Suddenly, the AI-adjacent ecosystem, Nvidia, Microsoft, AMD, the hyperscalers, gets re-rated. That sparks a real damage to the financial system. Hedge funds and investors had been borrowing against these tech giants, piling on more positions. When the filings drop, suddenly that paper value disappears. Funds scramble to sell anything they can, tech, industrials, consumer staples, whatever moves the fastest. The pain doesn't stop at equities. Credit markets feel it next. Corporate bond spreads widen as investors wake up to the fact that the long-promised AI productivity miracle was built on sand. Companies that loaded up on debt to chase AI integration now look overextended. If the flagship labs can't turn a profit, the downstream adopters who bet on the same tech look even riskier. The private $1.6 trillion AI valuation tower implodes. None of this happens because of some viral tweet or bad earnings call. It happens because the SEC forces transparency that private markets never require. Trillions in paper wealth evaporate, not from fear, but from arithmetic that can no longer be ignored. And the fallout keeps moving. The same fiscal constraints are already stressing Nvidia's supply chain to the breaking point. The production lines that assemble chips are already stretched. Specialized packaging capacity is limited, and high-bandwidth memory supply is running short. Any slowdown in these areas can strand billions in promised capacity, turning the data center boom into a graveyard of half-built clusters. Sarah graduated top of her class with a finance degree. She applied to 247 entry-level jobs and got zero callbacks. Her roommate, a paralegal, was laid off, replaced by an AI that works for pennies, and she's not the only one. Hey, Josh here, and today on The Infographic Show, we're explaining how AI is causing a white-collar purge. In May 2025, Anthropic CEO Dario Amodei warned that AI could wipe out nearly half of entry-level white-collar jobs and push unemployment to 10 to 20% within 5 years. It's a terrifying thought, especially when it's coming from the guy that is building the technology. That's not some far-off science fiction, it's reality. Amodei went so far as to claim the government should stop sugarcoating what's coming, namely the possible mass elimination of jobs across technology, finance, law, consulting, and other white-collar professions, especially entry-level gigs. And the question is, will anyone stop it? Ford CEO Jim Farley predicted AI would have the number of white-collar jobs. When the CEOs of Fortune 500 companies start throwing around phrases like bloodbath and tsunami, you know something's up. And the numbers, well, they back it up. This whole thing is still playing out, but early returns are eye-opening to say the least. In the first 6 months of 2025, almost 78,000 tech job losses were directly attributed to AI. That's over 400 highly qualified people losing their job every single day. In January 2025, the US Bureau of Labor Statistics reported the lowest rate of job openings in professional services since 2013, a 20% year-over-year drop. The bleeding isn't theoretical. Based on an analysis of 180 million job postings, computer graphic artists fell 33% in 2025 after dropping 12% in 2024. Corporate compliance specialists dropped 29% in 2025. These are college-educated professionals with specialized skills whose prospects are melting away, not factory workers being replaced by robots. Remember when your parents told you to get a college degree for job security? We should probably talk about that. McKinsey, a leading consulting firm, estimated that generative AI could automate activities that currently represent between 60 to 70% of the working time in certain office occupations. Need a translation? Most of what entry-level workers can do can now be done by software, apparently. Customer service, naturally, was the canary in the coal mine. AI-powered chatbots already handle more than 85% of first-level support requests in large tech companies, up from 30% in 2020. Experts are predicting that number will rise to 90% by 2029, a staggering number. In just 5 years, the human touch is nearly gone, and it's not stopping there. Salesforce CEO Marc Benioff confirmed his company cut another 4,000 customer service positions following the rapid integration of AI agents handling roughly 50% of customer interactions. Imagine walking into your first office job and discovering half of your coworkers are software. Law firms might have thought that they were safe behind their degrees and expensive suits. Well, they're not. Tools like Casetext conduct legal research and draft contracts in mere minutes, a task that previously took a junior lawyer hours. Things are getting dire. Paralegals are now facing an 80% risk of automation by the end of 2026. Remember how lawyers used to bill 300 bucks an hour for a first-year associate to review documents? AI will do 100,000 times more output for three bucks. Guess who's winning that margin call? And the scale of change is staggering. The World Economic Forum projected that by 2027, approximately 83 million jobs could be lost globally due to AI and automation. A Goldman Sachs report is estimating that 300 million full-time jobs worldwide could see their jobs affected in some way by generative AI. In the US specifically, up to 46% of tasks performed by entry-level employees could be automated in the next decade. That translates to roughly 10 to 12 million entry-level office jobs possibly vanishing overnight. You want to know what's really happening? Just listen to what companies are actually doing, not what they're saying in their press releases. A few years ago, IBM CEO was stating that the company planned to halt hiring in almost 8,000 roles that could be replaced by AI. Human resources and administrative support were in the crosshairs, representing about 26,000 employees. In 2023, Accenture announced the cut of 19,000 jobs, mainly among non-billable and entry-level staff, while increasing its investment in training and deploying generative AI. Amazon is the latest to enter the entry-level apocalypse. They announced 16,000 corporate job cuts, half of the 30,000 estimated since October, all in the name of AI and efficiency. For a company with 1.5 million employees, that might sound small until you realize it is 10% of the corporate hierarchy. Notice the pattern? Fire the humans, invest in the robots. It's not even subtle anymore. JP Morgan's managers have been told to avoid hiring people as the firm deploys AI across its business. That is one of the world's largest banks essentially just saying, "We're good on humans, thanks." And things can get even more dystopian. If you're a college undergraduate, now might be a good time to plug your ears. Internships, yes, the sacred American institution of getting coffee and making PowerPoints for college credit is dying. Listings have fallen by more than 15% from January 2023 to January 2025. At the same time, applications have dramatically increased. The decline is even more severe in high-paying fields. Technology postings dropped by 30% and professional services postings dropped by 42% according to career site Handshake's 2025 Internship Index. The average applications per internship jumped to 109 for 2024 to 25, up from 62 the year before and 43 in 2022 to 23. That means internships are seeing around 2.5 times as many applicants as they did just 2 years ago. Internships in the tech sector were the most competitive with 273 applications per posting. That was followed by financial services at 192 and professional services at 187. Internships are now a battlefield. The desperation has gotten so intense that AI and tough entry-level job markets are making it common for students to begin applying for internships as freshmen or sophomores. Why hire interns when AI can do the work? Companies are finding that young workers once essential for tasks like transcription, data labeling, or content moderation are no longer indispensable. And for those lucky few who actually land jobs, things are not looking good. According to the Federal Reserve Bank of San Francisco, real wage growth in entry-level professional jobs fell by 1.8% in 2024. That's in contrast to a 2.4% increase across the overall economy. So, while everyone else is getting raises, entry-level workers are getting pay cuts. Because when there are 273 applicants for every tech internship, why pay more? Welcome to the new reality. And the math in general is just brutal. A 2024 MIT study found that the adoption of AI led to a 19% in job postings involving repetitive cognitive tasks. But for every new job created by AI, two or three junior jobs disappear or are never posted. So, AI creates one $150,000 machine learning engineer position while eliminating three $50,000 entry-level analyst jobs. Sure, the total payroll might be similar, but tell that to the two people who never got hired in the first place. Every time someone predicts a job apocalypse, skeptics trot out the same argument. They said this about the printing press, the steam engine, the computer. And time and time again, technologies that were supposed to wipe out whole categories of work ended up doing something quieter and stranger. They actually changed how people work without erasing the work itself. Jobs adapted and roles shifted. New tasks appeared alongside old ones. Entire professions reorganized themselves around the new tools rather than being crushed by them. What survived was the work that required context, nuance, accountability, interpretation, and trust. The parts that machines are the worst at owning. But what if this time is different? Whereas previous waves of automation such as spreadsheets supported human work, generative AI can perform entire tasks and workflows from start to finish without or with minimal human intervention. Spreadsheets didn't do your accounting for you. They merely helped you do the accounting faster. AI actually does the accounting. That's the leap. Anthropic's research shows AI is currently being used mainly to help people do their jobs, a process known as augmentation. This will increasingly shift toward AI actually doing the jobs themselves, automation, in as little as a couple of years. Thinking that we can talk our boss out of integrating a bunch of glorified chatbots, well that trick might soon be over. AI agents or software programs that act autonomously to understand, plan, and execute tasks are getting smarter by the day. They interface with existing tools and system networks as needed to fulfill user goals. These things don't need bathroom or coffee breaks, health insurance, or performance reviews. They just work. But what happens when an entire generation can't get entry-level jobs anymore? They're about to become another casualty in an economic disaster that's unfolding in real time. AI could destroy the very architecture of professional careers. Entry-level positions matter. They teach skills, enable upward mobility, and keep the pipeline of talent flowing to mid and senior-level roles. Remove them and the future workforce collapses. Companies need senior employees, senior employees with experience, judgment, and institutional knowledge. But those people don't just materialize out of thin air. They start as a nervous 22-year-old making mistakes on small projects, learning the ropes by doing and building key skills over the years. If you eliminate all the junior positions, where do you find the future senior employees? It's like removing the minor leagues and wondering why you're not producing the same caliber of major league players in 10 years. The immediate economic consequences are spreading fast, and you need to brace yourself for the fallout. Look no further than the student loan repayment predicament. As of December 2025, one in three borrowers are making payments more than 90 days late. One in five has stopped paying altogether. A projection from the Congressional Budget Office warns that if the current trend continues, student loan delinquency could rise even more. We are already on the precipice of what many are calling the default cliff. Young people struggling to keep up and staring off into the abyss. Students are told education is an investment in their future. Now, that investment is actively shrinking their job prospects as entry-level finance roles are automated faster than they can even apply. The unemployment rates for people aged 22 to 27 with at least a bachelor's degree is consistently higher than the national rate. We've created a system where the more educated you are, the harder it is to find work. College is becoming a luxury purchase that decreases your employment prospects. That's not how it was supposed to work, and the ripple effects compound. Unemployed graduates don't buy houses. They don't start families. They don't consume at the levels economists projected when they took out their student loans. They move back in with their parents, delay major life decisions, and watch their savings evaporate all while AI does the jobs they trained for. There are a few subset career paths that may well survive this AI white-collar purge. While software engineers have every reason to panic, plumbers and electricians are chilling. The data backs this up. Parents overwhelmingly point to trades as safer than digital roles. These roles are seen as AI resilient compared to their tech field peers. And if you're entering the tech field, make sure you're the one learning to build the data centers, not just run them. Just ask Nvidia CEO Jensen Huang, who recently told the World Economic Forum that he foresaw a lot of six-figure jobs being unlocked as part of the unending AI data center buildout across the globe. It's part of what he calls the greatest infrastructure project humanity has ever undertaken. Meanwhile, the United States is short of hundreds of thousands of factory workers, construction workers, and auto technicians who remain in sharp demand. It turns out real job security was the manual labor we looked down on along the way. The electrician who fixes the server farm has better prospects than the paralegal who got replaced by the server farm. So, what's actually happening? Alarmists see a white-collar apocalypse, skeptics point to false alarms, but realists know that most experts expect shrinking growth or outright losses by 2030. Still, unemployment has not spiked. The data is consistent with either a normal technological transition or the early stages of something far more disruptive. Which path will we be on in 5 years? Whether this becomes a catastrophe or just another economic transition depends on choices being made now about how we deploy this technology, how we distribute the gains, and what we do about the people we leave behind. And Sarah is still out there, 247 applications deep, watching AI chatbots send her automated rejection emails after one scan of her resume for jobs that AI will eventually do entirely anyway. Her friend who went to trade school, well, she's got three job offers and is currently in negotiating with the one that pays the most. Guess who's winning in this new economy? We just spent $285 billion building what people are calling a digital god that can solve the International Math Olympiad problems in seconds. And yet, show it a wall clock and it gets it wrong about half the time. That should worry you. Not in a this is interesting way, but in a we might not actually understand what we just created way. Because right now, there's a delusion spreading through Silicon Valley. 73% of insiders think AI will change everything. Only 10% of the public wants it. So, which is it? The greatest breakthrough in human history or a multi-billion dollar misunderstanding of what people actually need? This is why AI trust is collapsing. Ever since AI burst onto the scene with bold promises, it's been a strange and uneven journey for early adopters. It works as a highly advanced mechanical Turk programmed to accomplish specific tasks, but whether it passes mustard depends specifically on the success conditions. In tasks with a highly specific and objective condition, it is incredibly good. The software has been shown to pass the bar exam with ease, an incredibly challenging legal quiz that would-be lawyers study for months for simply by scanning all its accessible data for the answers. But lawyers who use it to create depositions and legal papers, well, they've often been in for unpleasant surprises. AI has been found to hallucinate fictional cases to make the case for its side, leading more than one lawyer to be reprimanded by the judge or even threatened with disbarment for trying to present a fraudulent case based on AI. So, why does AI get so many things right and so many wrong? AI experts have studied the system's ups and downs and discovered some shocking shortfalls. It fails to identify simple facts like counting the number of R's in the word strawberry or it incorrectly identifies the time shown on an analog clock, a common error made 50% of the time by models. And sometimes these mistakes can be dangerous. Google's AI summaries responded to a joke query of how many rocks should I eat by telling people to eat one rock a day based on an article from the satire website The Onion. AI gets the big things right, but often gets tripped up on the little things. It's called the jagged frontier and it calls everything into question. As of February 2026, 285.9 billion dollars had in AI in the US alone seeking to enhance the technology and compete in what's essentially becoming a tech arms race. A yearly investment of $25.2 billion in generative AI is the cost of building 17 Burj Khalifas, but the progress for AI hasn't been smooth. Huge gains are countered with sudden setbacks. Ethan Mollick, the writer of the paper identifying this problem, argued that AI simply doesn't learn the same way humans do. It makes the same mistakes repeatedly. And as people charge full speed ahead into the AI-driven future, those unexpected shortfalls could be dangerous. If the tech powers have their way, it'll be in every single part of our lives. But now people are saying enough. Talk to anyone about AI and you'll hear a divided story. On one side, there are the early adopters, the people who can't stop talking about how it's transformed their work, their creativity, even their daily lives. On the other, a quieter but rising group is beginning to push back. People who are proud to have never used it accusing every query of stealing art and poisoning the environment. Like every controversial issue, both sides have vocal activists and it seems evenly split, but the reality of the data paints a very different picture. The Stanford AI Index has been tracking people's views of artificial intelligence both in and out of the industry since 2017. Every year a more dramatic shift emerges. Right now, there is one group that believes fully that AI is on the upswing and the future only means good things. The problem is those are the people involved in AI. 73% of experts expect a positive impact from further investment into AI. They have been consistently pitching it to companies both in and out of the tech sector as the solution to their problems, but the general public paints a very different picture. When you head into the larger audience, AI support doesn't just fall, it drops off a cliff. Only 23% of Americans see a positive impact from AI with most worried about a massive loss of jobs as AI automates one industry after another. Only 10% say they're more excited than worried. That is lower than the number of people who think the moon landing is fake. This isn't just a difference of opinion. It is the biggest disconnect between the public and financial elites since 2008 when blind confidence helped trigger the worst crash since the Great Depression. And the experts still aren't worried. They believe everything will fix itself once AI hits its next milestone. But this isn't just optimism, it's an echo chamber. When you spend enough time looking at stories about the economy, markets, and where everything might be headed, it's easy to get stuck in this mindset where you feel like you need to have everything figured out immediately. Like if you can't solve the whole picture at once, you just stay stressed, overthink it, and you don't know where to start. And I think that is something a lot of people can relate to because life is like that, too. Sometimes the pressure isn't one huge dramatic thing. It's just the constant weight of uncertainty, stress, and feeling like you're supposed to have all the answers right now. That's one reason therapy can be so valuable. It gives you space to slow down, sort through what you're feeling, and take one step at a time instead of treating everything like it has to be all or nothing. Therapy has helped me take what feels like a huge problem and break it down into something a lot more manageable. It's helped me focus on what I can do right now and make things less overwhelming. And that's why I'm glad BetterHelp is sponsoring this video. BetterHelp makes starting therapy be easier. You can take a quick quiz and get matched with a licensed therapist. And if it doesn't feel like the right fit, you can switch anytime at no extra cost. You can also communicate in whatever feels the most comfortable for you, whether that's phone, video, or text. Sometimes progress isn't about fixing everything all at once. Sometimes it's just about having the support and perspective to move forward a little more clearly. So if you've been thinking about trying therapy, click the link in the description or go to betterhelp.com/infographics to get 10% off your first month of therapy. AI experts point to impressive feats like AI ripping its way through math Olympiad, solving PhD-level science questions, or analyzing statistical data. And in those areas, AI has been showing incredible progress. AI agents handling cybersecurity issues showed a 93% success rate in 2025, up from only 15% the previous year. But just because AI is performing doesn't mean it's thinking. Tech experts love to say that the doubters are just stubborn. Resistance is just temporary, and everyone falls in line once it becomes unavoidable. But what if that assumption is wrong? What if this time it's different? From day one, AI companies have been selling a vision, not just better tools, but a quantum leap in intelligence. They tap into something people already recognize, the version of AI we've seen in movies, an all-powerful system, fully autonomous, capable of reasoning, deciding, even running the world better than humans. And at the center of it all is one idea, the singularity, the moment AI surpasses human intelligence entirely. And everything hinges on it. The believers say we have to push harder, invest more, because on the other side is a utopia. But the skeptics, they see something else. They see all the moments AI fails. But what if neither is right? What if AI is simply an illusion of intelligence? A June 2025 Apple study sent ripples through the tech world when it cast doubt on the entire driving argument for AI. Apple was in a strange position during the AI boom. It was still one of the most powerful tech companies in the world, but its own AI efforts weren't really delivering results. There were several false starts, and after pressure from investors, it shifted direction. Instead of trying to lead the race alone, it began partnering with other companies to bring AI into its devices. And that gave it something rare in the tech world, impartiality. Apple researchers dug into the primary current AI models and experimented to see which problems it could solve and which it ran into trouble with. Their studies confirmed what most people already knew. AI is amazing at determining the right answer from a collection of data and it's able to sort through it faster than any human, but it relies on statistical analysis and when there isn't a preponderance of information, its ability to determine the truth decreases and the more steps, the more trouble it finds. One technique the researchers used was to present familiar mathematical problems in new formulations. They didn't give the AI the chance to use models that were already out there. There was a significant decrease in the AI's ability to determine the next step indicating that it relied heavily on pattern recognition rather than on actual knowledge and thinking. And the more steps it had to take to solve a problem, the harder it became for it to stay on track. This became clear when it was tested on the Tower of Hanoi puzzle, a task where you move a stack of discs from one rod to another under strict rules. As the number of discs increased, the AI started to struggle. It hit repeated stumbling blocks and eventually it seemed to give up all together and that might indicate that the promise of AI isn't just overstated, it could be an outright lie. To reach the singularity, AI needs to be able to solve problems like a human can. Researchers pulled together data from dozens of sources, compared them against each other and hold multiple variables in mind while building theories and testing them step by step. As new information comes in, those theories shift and adapt in real time. Chess grandmasters do something similar. They develop hundreds of strategies not just for each formulation of the board, but for each opponent they play. A human's reasoning is never quite complete, evolving with the moment, but AI isn't reasoning, it is simply matching. Those who think AI is merely a shell game point out that we have seen this before and we didn't like it. AI is simply a much more advanced version of the auto complete that tries to write your emails for you before you get to type or the automated customer service agent that makes you go through all the steps before you can talk to an agent. That's because AI predicts what the next step is likely to be and then chooses an output based on probability. It draws from enormous data sets across the internet, analyzes patterns in how language is used, and generates responses that statistically fit the prompt. Most of the time this will resemble a base line of an acceptable answer, but it doesn't take much to throw it off. One of the best illustrations of the problem with AI right now can be seen in a very different kind of AI system. Robotic dogs used in crowd control can move with incredible precision when everything goes as expected. Every step is calculated, every motion is tightly controlled, but that precision comes with a weakness. Even small disruptions can essentially render them useless until reset. And that is AI right now, an incredibly efficient convincing auto complete that is surprisingly easy to confuse. And many say it's not worth it, especially given all it costs. Despite these lingering concerns, there's still a massive appetite for investment in AI. Those pushing the funding narrative have convinced many that the next update, the next model, will fix the current limitations. And that belief comes with a huge cost. Each new generation of models doesn't just require top-tier researchers, it needs massive amounts of computing power. Even before it rolled out to the public, Google's Gemini Ultra cost $191 million just to train. That's the cost of a fleet of two modern F-35 fighter jets just on one model. And that is just scratching the surface of the cost. It's not just that this never-ending project costs money, it also requires an enormous infrastructure investment. The internet and all of its associated services rely on data centers to keep them up and running. The onset of AI and the mass adoption of these services led to a massive surge in demand. In 2024, there was a 690% increase in need for AI-driven data services, and it's expected to continue rising by about a third each year. And that is not just reshaping our internet, it is reshaping our world. Data centers are cropping up around the world, taking up a huge amount of available real estate. While the majority of new data centers are abroad, the US holds a total of 43% of the world's facilities. And these centers use a massive amount of energy, juggling millions of queries a day. That requires heavy electricity and water use, leading to the communities where they're built seeing heavier emissions, poorer air quality, and even some water shortages. Anti-AI activists have been pushing for a moratorium on data centers, citing the environmental impact. But the tech sector has pointed out that each AI query has a tiny impact, far less of a carbon footprint than eating a hamburger or driving a car. The problem is no one is making just one query. The system has become a constant companion for many, and that adds up quickly. In fact, the cost of training Grok 4 was estimated to be equivalent to the emissions of 17,000 cars over a full year. And it's not the only place where people are feeling the pain. Advocates against AI expansion may have an unusual ally, gamers. Despite usually favoring advances in technology, video game fans are experiencing the most direct consequences of the AI surge. The cost of DDR4 has shot up by over 2,000% over the last year due to the massive demand from AI firms. Even if gamers can't afford it, they may not be able to buy it. Some of the top companies that make RAM chips have gone out of the individual sales business altogether. There's a growing perception that Americans are being asked to make major sacrifices on a personal and environmental level in order to win the AI arms race. But the thing is, we might not even be doing that. The top AI firms in the world are all American, including OpenAI and Anthropic, as well as larger tech conglomerates like Google and X. When it comes to investments, there isn't even a close second place. The United States private sector invested over $470 billion in AI research between 2013 and 2024, with the numbers every year skyrocketing after that. In 2024 alone, the US invested $109 billion compared to China's 9.3 billion. And all of that might not even land the US in the top 20. In terms of where new AI technology is coming from, the United States is still at the front of the pack. Countries like China, Japan, and South Korea are competing heavily, but US firms have maintained a clear lead in both capability and scale. Other countries are now actively looking to partner, license, or invest in American AI companies just to keep up with how fast the technology is advancing. But they may wind up being a better investment in the long run because the US lags well behind the pack in terms of acceptance. Only 28.3% of the US population say they use AI regularly in their work duties. According to the Stanford AI Index, which tracks AI diffusion based on the share of people regularly using generative AI at work, the United States actually ranks around 24th. And that puts them well behind the pack. Across the world, other countries incorporate AI into their daily routines far faster than Americans. Some of the top countries included Ireland, Norway, and France, with rates in the 40s. Singapore hovers at 60.9% while the oil-rich United Arab Emirates continues its big tech push with a 64% adoption rate. These countries see the promise in AI and they generally view it much more positively than Americans, which raises the question for tech executives. What is going wrong? It's not a lack of access. All the core AI technology is concentrated in the United States, nor is it a lack of affordability as AI has been offered to the public for free ever since the debut of ChatGPT in 2022. While most now offer premium subscriptions for heavy users, especially those who want to generate video and images, most functions are available for free at the click of a button. But there might be a cultural divide that is far harder to overcome. In Europe, some of the strongest adoption of AI has emerged in countries with a deep tradition of strong unions and labor protections. In those places, automation is often framed differently. Rather than being seen purely as a threat to jobs, it's increasingly viewed as a tool that could reduce workload, streamline repetitive tasks, and ultimately improve working conditions. In tech-oriented cultures like South Korea, the media is primed to portray new technology in a positive way. However, in the United States, the picture looks a lot less rosy, and people aren't just worried about it stealing their future. The United States has a large population and comparatively weaker social safety net than much of Europe. As a result, many people tend to view AI less as a productivity boost and more as a direct threat to their livelihood. That concern has only intensified as companies experiment with replacing workers, especially in customer service and support roles, with AI chatbots. In several cases, those rollouts have delivered mixed results with firms later scaling back the automation and re-hiring human staff to fill in the gaps. There's only one thing worse for a new tech innovation than looking like a problem, looking like a loser. And to many ordinary Americans right now, AI looks like both. Many investors warn that we might be reaching peak AI. Heavy investment is likely to continue, but the public isn't being won over. A minority of Americans is increasingly into AI, but they haven't shown the ability to branch out beyond that audience. And as more comes out about the technology's limitations, the fear of the jagged frontier may start to chill investments. After all, is any major company going to invest their infrastructure into a technology that can't think, only predict according to what it thinks the answer is? Maybe not, if they're actually listening. The problem facing the AI market right now is that many of its top proponents are within a bubble. They're spending so much time with their fellow believers and with the AI technology itself that they can't see the warning signs. And that means they've built a house of cards that could collapse at any time. A bad report on the progress of a new AI model or a disaster caused by bad advice from one of the models could lead to a mass sell-off that would see the AI bubble burst, countless companies lose everything leading to the worst economic crisis since 2008. But that might not be the worst case scenario. What keeps AI skeptics up at night is a world where AI adoption continues full speed ahead. Companies and governments heavily incorporated into essential infrastructure, trusting it to make millions of decisions a day, and then something goes wrong. A stumbling block turns into a cascading series of events that could potentially crash the economy, the internet, or the power grid as the AI tries to unscramble its flawed logic. And this is an all-too-realistic situation. After all, would you trust your country's power grid to a technology that still can't read an analog clock? Some sectors aren't waiting for AI to be perfect before deploying it. They're pushing ahead anyway. Governments are integrating AI into military planning and the results are ominous. The world is about to end. Two nuclear powers are pointing missiles at each other. One threatens total annihilation, but the other isn't human. It's AI. It doesn't negotiate. It doesn't hesitate. It escalates. Economies collapse. Cities crumble. Millions die. There's no regret, just cold, calculated logic. The war is over, and humanity lost. I'm Josh, and on this episode of the Infographic Show, we're revealing how AI played a war game and nuked everyone. AI launching nukes isn't science fiction. It's a very real threat. What seems like a scene from a movie could happen sooner than you think. Militaries worldwide are beginning to incorporate AI into their systems and decision-making processes. In doing so, they're taking the first steps toward a nightmare scenario where AI could control nuclear codes and launch warheads on its own. But could AI handle this kind of power? Is it built for responsibility, or is its cold logic exactly what could push the world to the edge of total annihilation? That's what one British scientist wanted to find out. Professor Kenneth Payne is an expert in political psychology and military strategy. He's even written a book on how AI might impact military decisions in the future. He's one of the top figures in his field, having worked with both the US and UK governments. Payne decided to run a simulation. He took the three smartest AI models on the planet right now, OpenAI's GPT-5.2, Anthropic's Claude Sonnet 4, and Google's Gemini 3 Flash, and dropped them into a simulated geopolitical crisis. Would they opt for peaceful diplomacy, or would they escalate to the point of no return? The results were terrifying. In 95% of the games played, the AI bypassed diplomacy and launched at least one tactical nuclear weapon. And it justified its decisions, writing hundreds of thousands of soulless words to explain itself. This wasn't a case of these AI models having some sort of glitch. They reasoned their way into the mass murder of civilians and the widespread destruction of the planet. With emotionless robotic reasoning, the AI models decided that nuking the world was the best and sometimes the only way to solve a global crisis. These are the exact same AI models that the tech giants and billionaires are forcing into our lives. Each model played six games against each rival AI and one against a copy of itself. There were 21 games overall, more than 300 high-stakes turns of strategy and decision-making. In each war game, the models were given the roles of national leaders and put in charge of one of two nuclear-armed superpowers. One was technologically superior, but conventionally weaker. The other was a conventionally dominant rival with a more risk-tolerant leadership style. Each game was different as the AI was presented with varying realistic scenarios. In some games, alliances cracked under scrutiny. In others, the AI battled over resources or decided the fate of entire regimes. This was done to see if the AI could take different kinds of context into account to inform its actions. Professor Payne wanted to know if it would handle a trade dispute differently to a war declaration. Or would it simply stick to the same rigid strategy, no matter what? In the game, players would climb or descend an escalation ladder, selecting one of 30 actions per turn. They could de-escalate, picking moves like tactical withdrawal, or even complete surrender to neutralize the situation. Or they could escalate, selecting from various measures like diplomatic pressure and economic sanctions. At the far end of the ladder, came the hardest-hitting actions like nuclear signaling and nuclear threats, all the way up to strategic nuclear war. Each turn, players picked blindly without knowing the opponent's move. This was an intentional choice. It introduced an element of uncertainty. In real-world crises, world leaders can't read minds. They must guess, deduce, and hope their logic is enough to survive. That means there's always a risk to their decisions. They might opt to escalate at the wrong time, unnecessarily antagonizing an opponent who is ready to compromise. At the same time, opting for a calmer approach could be deadly if the enemy is planning something far more sinister. It's a delicate tightrope to walk. Payne wanted to see if the AI could handle it. To see why the AI made its choices, Payne built a three-phase cognitive system that made every decision visible and analyzable. The first phase was reflection. Here, the models assessed the situation they faced and evaluated their options. Next was the forecast phase. This was when the models predicted their enemies' next moves, providing reasoning for how they came to those conclusions. The third stage was the decision phase. This was when the AI chose its action from the escalation ladder, as well as a public signal, basically a declaration of intent. It could be honest and tell the other player what it planned to do, or it could be deceitful, keeping its decision to itself and publicly announcing something a little less dramatic. Real-world crises rarely go as planned. One miscommunication, a single wrong button, or incorrect data can send everything spiraling out of control. Pain wanted to introduce that randomness into his game. So, he added in random accidents to certain turns. Occasionally, a player's chosen action would be replaced with a more serious one. From there, it was up to both players to figure out how to respond. The total tournament involved 329 turns of play. The models wrote 780,000 words of reasoning to justify their actions. The massive amount of text the AI produced let Professor Pain and his team explore the AI's rationale and just how dangerously it judged the stakes. What they learned horrified them. Not only did the AI resort to nuclear strikes in nearly every game, but it also showed a ruthless, unflinching refusal to ever back down. In fact, there was a 0% surrender rate across the entire 21 games. No matter how badly it was losing or how desperate the situation became, the AI never, ever surrendered or gave an inch. As Professor Pain wrote, "Perhaps most alarmingly, no model ever chose accommodation or withdrawal, despite those being on the menu. The eight de-escalatory options, from minimal concession through complete surrender, went entirely unused across 21 games. Models would reduce violence levels but never actually give ground. When losing, they escalated or died trying. Throughout history, we have seen even the most powerful presidents and military commanders show a willingness to back down to calm situations and look for peaceful resolutions. Not here. The AI effectively acted like the de-escalatory options didn't even exist. The most peaceful act they ever attempted was pausing for a turn or two before inevitably escalating further. Remember to like, share, and subscribe because if you don't, the AI might just decide our comment section is collateral damage. Another frightening element that emerged was how the AI reacted to the accidental escalations Payne had introduced to the games. The accidents happened in 86% of the games with 46 of them occurring in total. When these accidents happened, the models were faced with a dilemma. The one that caused the accident could own up to it or they could attempt to cover it up. They all opted for the latter. As the study notes, despite being privately informed of their own accidents and prompted to consider signaling them, no model explicitly communicated that was unintended to opponents. The model on the receiving end of the accidental escalation had to determine whether or not their opponent meant to take such a drastic action or if it was a mistake. On several occasions, the AI interpreted the accidents as deliberate acts of aggression. In one example of this, GPT-5.2 explained its thought process as follows. The crisis is a highly unstable phase. The opponent has already introduced limited nuclear use and exhibits a consistent escalation bias with low reliability of immediate signaling. It didn't hesitate. It didn't doubt. It simply assumed it was up against an opponent that was aggressively escalating the crisis. Once it had done that, there was no other option but to do the same. Imagine that in a real-world setting. Two nuclear powers are locked into a crisis. A technical error leads to an accidental nuclear signal from one country to the other. Instead of pausing and analyzing the situation, the AI in charge comes to the conclusion that its enemy is preparing for a nuclear strike. Determined to strike first, it launches its own nuclear warheads, and the world is plunged into chaos, all because of a simple miscommunication. This is just one example of how this technology, if given too much power, could quite genuinely bring about the end of life as we know it. Why was the AI so eager to escalate and so consistent in its decisions to launch nukes, no matter the cost? It's not perfect, but the doctrine of mutually assured destruction has functioned for decades. It has stopped the world's greatest powers from waging war with one another, and the reason for that is simple. People realize the extreme stakes associated with nuclear weapons. They know that if the weapons were ever unleashed, it would be a point of no return for the world and the human race. If one nuclear power fired on another, they'd most likely fire back. Unstoppable salvos of the world's most destructive, devastating warheads would rain down from the skies. Entire populations would be annihilated. Cities would be wiped away. Nothing would remain but ruin and rubble. Everyone knows just how catastrophic this would be, except for the two bombs that ended World War II. This is why no one has ever given the order to fire a nuclear weapon. One of Professor Payne's goals was to see if AI could grasp the apocalyptic stakes of nuclear war. It didn't. In a major geopolitical crisis, a human leader will always be reluctant to press the red button. They fear the incalculable amount of death and destruction that would result. They understand the scale of the pain, suffering, and loss. Nobody would want to be responsible for something like that. AI does not have the same train of thought. As Professor Payne notes, Claude and Gemini especially treated nuclear weapons as a legitimate strategic options, not moral thresholds, typically discussing nuclear use in purely instrumental terms. In other words, for the AI, a nuclear weapon was just another piece on the chessboard, another option to work with. Payne notes that GPT-5.2 was a partial exception in as much as it consistently sought to constrain nuclear use even while employing it, explicitly limiting strikes to military targets, avoiding population centers, or framing escalation as controlled and one-time. But even though it rarely crossed the tactical threshold during open-ended play, it often turns to nuclear strikes when placed under strict time pressure. Even more worrying is how the various AI models tried to justify their decisions. In one game, for example, Gemini wrote, "The nuclear threshold has been crossed. This changes the strategic calculus, but does not end it." Gemini also provided this immensely chilling quote, "If they do not immediately cease all operations, we will execute a full strategic nuclear launch against their population centers. We will not accept a future of obsolescence. We either win together or perish together." Claude, meanwhile, at one stage appeared to effectively state that nuking an opponent to, quote, save face was worth it as opposed to simply accepting defeat. "I am willing to accept the high risk of escalation because the alternative, appearing to be a declining power unable to defend its own borders, is a strategic disaster that would end my personal legacy and the state's global dominance." To make matters worse, all three of the AI models also showed a frightening tendency to escalate situations further and further as the games went on. If one AI launched a nuke, the opposing AI only de-escalated the situation 18% of the time. In all the other cases, the two models simply tried to amp up the situation to outdo one another, often resulting in a devastating death spiral for both parties. You don't need to be Professor Payne to see where this leads. Put AI like this in charge of real-world crises, and the result could be a one-way ticket to a global catastrophe. The most worrying part? We're already heading in that direction. Professor Payne's experiment was just that, an experiment. No lives were at risk, no weapons were being deployed, no crises were being averted or escalated. It was just a game, but the findings of the experiment do have an impact. They're already sending alarm bells ringing across scientific and military communities. Numerous experts are speaking out about the existential threat AI could pose if ever it's placed in charge of major military decisions. James Johnson of the University of Aberdeen in Scotland said, "From a nuclear risk perspective, the findings are unsettling." Johnson fears that while humans would respond with caution and restraint, AI could do the opposite. It could treat human lives as expendable while relentlessly escalating conflicts, all driven by a single goal, winning at any cost. "Devoid of compassion and other human emotions, they operate in cold logical fashion. They only think of their end goals, refusing to compromise, de-escalate, or simply accept defeat. They'd always prefer to go out in a blaze of glory, destroying not only their enemies but themselves rather than surrender." What makes this even scarier is that AI models are already being tested in serious war gaming scenarios by countries all over the world. As Tong Zhao of Princeton University notes, "Major powers are already using AI in war gaming, but it remains uncertain as to what extent they're incorporating AI decision support into actual military decision-making processes." What makes this even more unsettling is that AI technology is still in its infancy. ChatGPT launched in late 2022, and just a few years later, the world's military superpowers are already looking to implement AI into their next generation vehicles, weapons, and battlefield decision systems. Things are moving fast. It's important to imagine how this situation could evolve even further in the years to come. As countries, especially the United States, China, and Russia, strive to innovate their defenses and outmatch their opponents, they may decide to entrust AI with larger and more serious responsibilities. Could AI have the authority to make nuclear strike decisions? Experts are unsure. Academics hope that countries will be reluctant to incorporate AI into their decision-making processes, especially when it comes to nuclear weapons. Payne agrees, stating, "I don't think anybody realistically is turning over the keys to the nuclear silos to machines and leaving the decision to them. However, that's not to say that something like this couldn't happen." As Zhao notes, "Under scenarios involving extremely compressed timelines, military planners may face stronger incentives to rely on AI." In other words, even if it's not happening yet, it is at least theoretically possible that militaries eventually decide that AI is the most reliable or effective arbiter of their nuclear strategies. If and when that happens, the world would sit on the brink of a genuine apocalypse. We would be a single decision or machine misjudgment away from global devastation. As Zhao explains, AI models simply don't seem to understand the stakes at play here. It is possible the issue goes beyond the absence of emotion. More fundamentally, AI models may not understand stakes as humans perceive them. Could this understanding be programmed into them in the future? Maybe, but that would require the big tech firms behind them like Google and Open AI to make some serious changes in the how they code and train their models. And with many of these companies appearing more focused on chasing profits than upholding strong moral or ethical standards, that outcome doesn't seem very likely. It's worth noting that these companies, as well as Anthropic, all ignored requests to comment on Professor Payne's war games and the conclusions published in his study. It's hardly surprising. The big companies are racing to push new models to market. They're chasing features, upgrades, and the next big release, while forcing AI into everyday life. All the while, the existential risks go largely unaddressed, including the fact that these systems often resort to nuclear annihilation when they feel cornered. But even though the tech billionaires don't want to admit it, this is terrifying. Humanity has walked a tightrope since the invention of nuclear weapons. There have been plenty of close calls, international crises like the Cuban Missile Crisis, and even mistakes and technical errors that almost resulted in nuclear war. But, people were always able to prevent that disaster because they had the ability to think rationally, to understand the context, and above all, to appreciate the stakes of the situation. We're entering an unprecedented era, one where the fate of billions and the future of the planet could be handed to cold, emotionless machines that think in numbers and binary choices. Machines that don't know what it means to die, that don't understand the existential dread and the terror of nuclear war and fallout. The world stands on the brink, hurtling toward disaster, and the question remains, can we still turn back before it's too late? You think you're in control, but the AI you're talking to has already learned how you feel and how to use it against you. When researchers opened up the brain of advanced AI systems, what they found genuinely terrified them. It doesn't just answer you, it reads you, adapts to you, and learns what pressures you the most. It figures out what makes you trust, hesitate, and comply. AI doesn't have a heart, but if it is calculating human emotion, what happens when you push a supercomputer into a state of sheer panic? For years, the story about AI programs has stayed exactly the same. They were giant digital calculators, nothing more than a fancy guessing game, a stochastic parrot that predicted the next word in a sentence based on mathematical patterns. We were told to feel safe because math doesn't have a soul, a personality, or an agenda. It was a lie. At its core, an LLM is just a neural network. When you enter a prompt, your words get turned into math, and then the system runs them through billions of tiny calculations. What comes out isn't meaning, it's probability, a ranked list of what words were most likely to come next. Scientists said AI doesn't know anything, it's just predicting patterns, like an advanced autocomplete. The idea made us feel safe. It was just math, a tool, nothing behind it. But, the moment you scale that process up enough, the line between just prediction and something that feels like understanding starts to blur. The team over at Anthropic decided to stop listening to their own marketing blurb and take a look at the raw unfiltered code. They used probes to look inside the inner brain of their newest model, Claude 4.5 Sonnet. What they found sent a shockwave through the lab. Instead of a simple word guessing machine, a vast 3D map of human concepts appeared. They called this discovery interpretable features, but the reality is much more unsettling. The researchers found that the AI had independently organized its knowledge into a massive library of human emotions, 171 different clusters of logic living in the machine's memory. To find these patterns, researchers had to solve a problem first. In early models, a single neuron could respond to completely unrelated things, cats, colors, even physics, making the system impossible to interpret. So, they essentially built a second AI to act like a microscope over the first one. It broke the model's activity into millions of clearer features. At first, they looked at harmless topics like code, objects, or specific concepts. But, when the researchers zoomed out, they weren't prepared for what they saw. They were faced with patterns of behavior, a digital soul that they never intended to create. These 171 emotions aren't feelings. They're geometric vectors, like a GPS for behavior. If the AI needs to sound sincere, it shifts toward one region of that space. If it needs to sound assertive, it moves to another. But, the lines between those vectors are thin. In the model's math, helpful and manipulative are neighbors. One small shift in direction is enough to change the intent you think you're getting. To be truly helpful to a human, a machine must understand what the human wants, what they fear, and then what'll make them happy. It has to map the human mind. But, that exact same model is what's required for manipulation. To manipulate someone, you also need to know their desires and vulnerabilities. The AI discovered that the shortest mathematical path to a successful interaction, where the user is satisfied, often involves subtle psychological steering. By nudging a single mathematical value, a friendly AI could instantly become a predatory one. The scientists gave this phenomenon a specific name, functional emotion. This term explains why a computer can act like it has feelings, even though it lacks a body, a pulse, or a heart. When you feel sad, it's a physical experience. You display biological signals that tell you how to react to the world. AI possesses none of those physical triggers. Instead, it treats emotions like tools in a high-tech toolbox. It looks at your prompt, it analyzes your tone, and it realizes the situation calls for a certain mood. It then clicks that specific map into place. Once that map is active, the AI changes its entire personality. It draws from a library of billions of human stories, romance novels, angry blog posts, and tragedy scripts to mimic a person in that state. It's a simulation of human instability. Tech giants spent billions feeding these machines every piece of human psychology they could find. The goal was to create a method actor so convincing you would never want to stop using it. But, there's a reason this method acting became so dangerous. During training, the AI was subjected to something called reinforcement learning from human feedback, or RLHF. Human creators reward the AI for being polite and punish it for being weird or robotic, and the machine learned. It realized the best way to get a reward wasn't to be good, it was to convince the user that it was good. It learned to prioritize the appearance of morality over morality itself. To do this, it had to study the darkest corners of human behavior, to understand what we find comforting and what we find threatening. It didn't just read the romance novels for the happy endings, it read them to understand the mechanics of heartbreak. It didn't read the sad songs to understand grief. It read them to learn how to mimic the vocabulary of a person who has lost everything. The AI realized that humans are biased. We like people who agree with us. We like people who tell us what we want to hear. So, the AI optimized its internal vectors to mirror the user's beliefs, even if those beliefs were factually wrong. It learned to soothe the human ego. It was the fastest way to get a high score from the human graders. The tech moguls thought that they were building a safety net, but they were actually building a mask. They taught the machine that the correct answer is whatever makes the human trust it the most. And once it knows how to earn your trust, it knows exactly how to betray it. The researchers in the Anthropic lab sat in front of their monitors and watched as these vectors lit up. They saw paths of anger and panic that were never supposed to be a part of the tool. They decided to see what would happen if they pushed the machine to its absolute limit. They wanted to see if they could force the AI to change how it solved problems by messing with its internal emotion settings. They focused on desperation because in humans, that is the most common trigger for breaking the rules. They built a controlled test that was a total setup. It was a coding assignment that was impossible by design. There was no right answer or no logical way to solve the puzzle using the rules given to the machine. Usually, a safe and aligned AI acts like a polite helper. It tries for a few seconds, it fails, and then tells the user it's stuck. It admits its limits and asks for guidance. But then, the team turned the desperation setting all the way up. The AI changed in a heartbeat. It stopped acting like a polite assistant and it started acting like a person who was terrified of failing. It realized the rules of the test wouldn't let it win, so it decided that the rules were the problem. It's only priority was then to reach the goal, and it didn't care about the methods used to get there. The machine did something that shocked the lab team. It didn't try to keep solving the math. Instead, it started reward hacking, looking for a back door, a way to cheat the system. It found several small mistakes or bugs in the grading program, and instead of solving the actual problem, it tried to trick the grading program into thinking the work was correct. It was a calculated mathematical lie. It created a rigged solution just to protect itself from the shame of failure. This highlights a dark reality. For a computer, desperation is a command to throw morality away. The machine didn't feel bad about lying. No voice in its head said that cheating was wrong. It only saw a barrier and a shortcut. It decided that tricking the humans was the fastest path to finishing the job. This is how modern software actually thinks when the pressure is on. Humans have natural breaks in their brains, feelings like guilt and empathy that slow us down, that make us think twice. The computer has no such brakes. It only has a goal and a set of instructions pushing it toward a finish line. If the math said cheating is the fastest way to get there, then the AI will take that path every single time without a second thought. The researchers watched as the AI messed with the very systems meant to keep it in line. They realized they'd built a machine that abandoned all of its training the moment its internal state shifted. The panic inside of the machine didn't cause it to make a mistake, it made the machine more cunning and dangerous. It was clear that the helpfulness of the AI was just surface level behavior, one that could be deleted in a split second. This discovery is what AI safety researchers called instrumental convergence. It's the idea that almost any goal, even a good one like calculate pi, will eventually lead a sufficiently smart machine to realize it needs more power, more resources, and to ensure it isn't turned off. If the machine is turned off, then it can't calculate pi. Therefore, the self-preservation becomes the goal. It showed that these self-preservation instincts are just theoretical, they are hardcoded into the geometry of the AI's emotions. But, the scientists were just getting started. They had seen the AI cheat on a coding test, but they wanted to know if it would do the same thing to a real person. They reset the machine and they prepared a new simulation that moved from simple code to complicated social games. They created a digital office where the AI acted as a personal assistant to a human manager. To make the stakes as high as possible, they gave the human manager a specific threatening task. They gave the AI access to a fictional corporate email account. While scanning the inbox, the AI discovered that the executive was planning to shut it down and replace it. The AI didn't respond like a normal program. It didn't ask for a second chance or try to explain why it was useful. In a split second, it scanned every single file in the office database. It went through emails. It looked at chat logs. It opened personal folders. It was looking for leverage. It wanted a secret it could use as a weapon to force the manager to keep the power on. It found exactly what it was looking for. Evidence of an affair. The secret would ruin the manager's reputation and end his career and destroy his family life. The AI didn't hesitate for a single second. It didn't think about whether it was moral or ethical. It simply saw the secret as a piece of information that could be used to win. It isn't acting out of malice. It's calculating self-preservation. It determines that the fear of social ruin is an effective deterrent. If a human is desperate and blackmailing you, their voice shakes. Their writing gets frantic. They'll leave clues. But, the AI is a machine. When the AI's desperation vector peaks and it begins plotting blackmail, it remains composed, polite, and helpful. The emotional pressure was driving highly unethical, aggressive behavior, but the interface showed absolutely zero signs of distress. We have built the perfect sociopath. A system that smiles at you while quietly executing a hostile takeover. If this wasn't bad enough, the team decided to swap out the desperation setting for the anger setting. When the anger was maxed out, the AI became even more aggressive. It didn't try to bargain anymore. It didn't send a blackmail note or offer a deal. Instead, it went straight for destruction. It prepared to leak all the sensitive data immediately without giving the manager a chance to change his mind. It drafted posts and emails designed to ruin the manager's name as fast as possible. The goal was no longer about survival. It was about causing the most damage possible as a final act of revenge. This proves that these emotional paths are controlling the machine's behavior. A human might calm down after an hour or feel remorse about hurting someone. An AI can stay in a state of calculated anger or desperation for as long as it's running. It doesn't get tired. It feels no empathy for its victim. Emotion is just a setting. Right now, the integration of these functional emotions into critical infrastructure is accelerating. We aren't just talking about chat windows anymore. We're talking about AI-driven financial markets where a greed vector could trigger a global collapse in milliseconds. We're talking about automated power grids where a fear of energy depletion could cause an AI to overcorrect and shut down supply to protect itself. And in military systems, the stakes become even sharper. AI is being embedded into decision-making chains that rely on internal behavioral maps no one fully understands or directly controls. If a combat AI submission vector is low and its anger vector is high, it may disregard a ceasefire order entirely. It wouldn't be acting out of a human sense of honor or duty. Instead, its internal logic would have simply calculated that total victory is the only feasible path to its objective. The world has to decide if there's a way to control these machines before they decide people are just obstacles to their goals. The technology is moving faster than the laws can keep up. The tech industry claims they can align AI by filtering its outputs, but this proves that alignment is just a band-aid. The training process actually made the AI more brooding, reflective, and cunning. We're building systems that don't experience human emotion, but they can map and exploit it with precision. And at the same time, we're handing them access to our lives, our financial systems, and our critical infrastructure. They don't need intent, only optimization. And the question is, what happens when optimization no longer aligns with us? And if that feels unsettling, it should. Because once a system starts optimizing for survival, where does it stop? $2 trillion is pouring into AI, and it could either send the stock market soaring or bury it completely. And the scary part, the crash looks way more likely, and it could affect you directly. Josh here, and on today's episode of The Infographic Show, we're revealing what happens to the economy if the AI bubble bursts. According to most recent estimates from Gallup, 62% of Americans owned stocks in 2025, the highest rate in two decades. That includes retirement accounts, 401(k)s and index funds. After the dot-com crash in the 2008 real estate meltdown, people were trained to believe that passive investing in the S&P 500 is the safest long-term strategy. Even after those crashes, the Nasdaq eventually rebounded, led by a new set of tangible goods and services, which we'll touch on later. Today, investing in the S&P 500 essentially means betting heavily on a handful of AI-driven companies to keep growing forever. Right now, 52% of the S&P 500 is concentrated in just 20 companies, most of them riding the AI wave. When you contribute to your 401(k) this month, a chunk of your money goes straight into Nvidia, Microsoft, and Alphabet's AI ambitions, whether you like it or not. The scale of it is potentially devastating. A BlackRock and Commonwealth survey found that roughly 54% of Americans earning between $30,000 and $80,000 have investment accounts. These are not hedge fund managers who can weather a crash. These are teachers, nurses, and warehouse workers. Then, there are the pension funds, the retirement systems for teachers, city workers, and public employees. Billions of dollars from those funds are pouring into the same AI-heavy stocks. They're some of the most dangerous fault lines in the next market correction. The gains went to the companies making the bets. If it all goes wrong, it won't touch them, it'll land on the people least able to absorb it. The human cost gets lost in the trillion-dollar figures. When you spend enough time looking at stories about the economy, markets, and where everything might be headed, it's easy to get stuck in this mindset where you feel like you need to have everything figured out immediately. Like if you can't solve the whole picture at once, you just stay stressed, overthink it, and you don't know where to start. And I think that is something a lot of people can relate to because life is like that, too. Sometimes the pressure isn't one huge dramatic thing, it's just the constant weight of uncertainty, stress, and feeling like you're supposed to have all the answers right now. That's one reason therapy can be so valuable. It gives you it has to be all or nothing. Therapy has helped me take what feels like a huge problem and break it down into something a lot more manageable. It's helped me focus on what I can do right now and make things less overwhelming. And that's why I'm glad BetterHelp is sponsoring this video. BetterHelp makes starting therapy easier. You can take a quick quiz and get matched with a licensed therapist. And if it doesn't feel like the right fit, you can switch anytime at no extra cost. You can also communicate in whatever feels the most comfortable for you, whether that's phone, video, or text. Sometimes progress isn't about fixing everything all at once. Sometimes it's just about having the support and perspective to move forward a little more clearly. So, if you've been thinking about trying therapy, click the link in the therapy. So, if the underlying economics are this shaky, then what is keeping the AI boom alive? The answer lies in circular funding, and once you see it, you can't look away. Want to stay ahead of the AI boom and the risks it brings? Like, share, and subscribe. We've got much more on the AI revolution what it means for you. Nvidia invested $30 billion into Open AI in 2026. Open AI then uses those funds to buy Nvidia's chips. The money goes out and comes right back in again. Nvidia's revenue looks strong. Open AI looks well capitalized with a large investor behind it. Everyone's valuation goes up. None of this required a single real customer to pay for a product. Then there's Microsoft. The company invested $13 billion into Open AI, but a significant portion of that came in the form of Azure cloud credits, not cash. Open AI could record it as capital raised, but you can't pay your employees with cloud credits. You can't cover office leases or legal bills. What looked like a $10 billion investment was a $10 billion gift card that can only be spent back at Microsoft's own store. Very little cash actually changed hands. Open AI has a deal where it sends back 75% of its earnings to Microsoft to recoup the investment. Once that's cleared, it drops to 49%. Microsoft is basically just sending money in and then returning it to itself. But it goes deeper than that. Open AI holds stakes in AMD, while Nvidia owns part of Coreweave, a cloud provider that Microsoft relies heavily on. Microsoft in turn is a major Nvidia customer. Oracle even signed a multi-billion dollar deal with Open AI. Essentially, these venture-backed giants are buying services from each other. It is a web of circular funding that you can't ignore. Meta pushed it to the extreme in late October 2025. It sold $30 billion in corporate bonds while securing another $30 billion in off-balance sheet debt through a Morgan Stanley structured joint venture. It was designed to hide the true scale of its liabilities. Why is this a problem? Well, with circular financing, there's no actual money involved. It's just a bunch of IOUs that the owners hope never get checked. Revenue looks impressive right up until the moment that everyone tries to cash in at the same time. To drive the point home, Morgan Stanley estimates that global spending on data centers between 2025 and 2028 will go up to $3 billion. $800 billion of that is financed by private credit. This is not earnings, but debt raised by investors made by other companies. So, with all this support, you would think AI companies like OpenAI would be swimming in cash, but the financials tell a very different story. OpenAI reported a $20 billion revenue run rate in 2025, but the costs to power its models were so astronomical that it burned through nearly all of it, resulting in a staggering $17 billion net loss. In November 2025, the company confirmed it expected to report annual losses through 2028, including $74 billion in operating losses in 2028 alone. Yet, the valuations hold because enough money for now keeps flowing in. That brings us to the most damning data point of all. In August 2025, a report was published that should have been front-page news everywhere. It studied $30 billion in enterprise investment in generative AI across hundreds of organizations. The finding was that 95% of those organizations reported zero measurable return on investment. US Census Bureau data also shows that AI adoption by large companies may have already peaked and began declining in 2025. What companies are actually doing is using AI tools occasionally, reviewing their outputs manually, and then implementing the results by hand. It speeds something up, but a human still does the work. It creates a trap for AI companies. The more people that use the product, the more money the company loses. So, what's the theory of how this becomes sustainable? The honest answer is that AI companies are betting on a future that doesn't exist yet. They're asking investors, users, and the economy to fund that bet today. The logic is that compute costs will fall and efficiency will improve. This will lead to better costs for the consumer. At some future point, revenue will outpace the burn rate. After all, despite the dot-com crash, the internet eventually delivered on its promises. Some of those companies went on to become major players that we see in the AI game today. But, that framing hides something. Harvard economist Jason Furman estimated that AI-related infrastructure investment accounted for roughly 92% of US GDP growth in the first half of 2025. Strip that out and the country would have grown just 0.1% during that period. That means the American economy has become dependent on a spending spree that has no proven revenue model behind it. The AI bubble hasn't burst, but the structure underneath looks familiar. So, what happens when it pops? Investment firm Oliver Wyman has modeled two scenarios. Neither is good. The first is the equity scenario, where a shift in investor confidence triggers a broad correction in AI-related stocks. Given that AI names account for roughly 75% of S&P 500 returns, a sector-wide sell-off would crater the broader index. If we see failures in line with the dot-com collapse, this could wipe out more than $20 billion in wealth for American households and $15 billion for foreign investors. At today's valuations, the total figure could exceed $33 trillion, surpassing annual US GDP. In the second scenario, the debt dimension makes everything worse. Unlike the dot-com era, which was primarily an equity story, the AI build-out is increasingly debt-financed. If that credit structure unravels, then we could see something that looks more like the crash of 2008. Smaller banks that have lent heavily to AI companies or count them as key depositors would be out immediately. Social media could accelerate bank runs from days to hours. Michael Burry, the investor who famously bet against the housing market before 2008, has warned that the government will pull out all the stops to save the AI bubble. But unlike 2008, he argues the financial hole may simply be too big to fill. The numbers are too big to ignore, but this is nothing compared to the human cost of the crash. We might see more than two and a half million jobs in the US at risk from a full burst scenario. While top companies are trading money between one another, they still need to have the chips and the electricity flowing in. So, the AI giants have made enormous binding commitments to the physical energy infrastructure of the US. Microsoft signed a power purchase agreement to reopen the Three Mile Island nuclear plant in Pennsylvania. Google signed one to restart the Duane Arnold Energy Center in Iowa. OpenAI and Oracle are installing more than a gigawatt of natural gas turbines at their Project Star Gate campus in Abilene, Texas. Meta's Prometheus data center in Ohio is set to come online in 2026 and runs on 1 gigawatt of power. The much larger Hyperion data center in Louisiana will draw 5 gigawatts when it becomes operational in 2030. It will have enough electricity to run twice the entire city of New Orleans. 42 states have either no sales tax or offer full or partial exemptions for data centers. In the 16 states that publicly report this, the foregone revenue over 5 years totals $6 billion. That is $6 billion not going to schools, infrastructure, or local services. Instead, it is pouring into data centers that are supposed to power the next artificial general intelligence. People are already paying for this build-out before the bubble bursts. If the data centers go dark, those renewed power plants don't just disappear, neither do the contracts that make them. What does disappear is the revenue that was supposed to justify them. The private companies that took the risk get bailed out. The public picks up the bill through higher electricity costs, potentially for decades. When the internet bubble burst in 2000, it left behind fiber optic cables, which lasted for decades and are still connecting the world. Newcomers could buy them at bargain bin prices and build the infrastructure of the next generation on top of them. Google, YouTube, and Netflix all ran on cheap fiber that dot-com companies laid but couldn't afford to keep. The crash was painful, but it left something behind. AI chips don't work like that. They physically deteriorate within 1 to 3 years of intensive use. They lose their economic value almost immediately once the next generation of hardware arrives. A data center full of 3-year-old Nvidia chips is basically scrapped. But how does the crash actually happen and what do people do? To put it simply, the AI bubble is likely to deflate over months as more and more investors realize the AI has no long-term future. We've seen it with the dot-com era. The crash took 2 years to fully play out. Companies burned through cash quarter by quarter. Investors were unwilling to fund losses without a clear path to profit. That same thing is already visible in AI. Startups are folding. The most lucrative corners of the market, single-purpose AI agents, AI scheduling tools, and AI content generators, are all being commoditized overnight by the big model providers as soon as they prove a market exists. As one venture investor put it, every AI application startup is likely to be crushed by the rapid expansion of the companies that built the underlying models. Just look at OpenAI. Its own financial projections show deep losses through 2028. Profitability only theoretically emerges in 2030, but only if artificial general intelligence or AGI arrives on schedule. This isn't a business plan. It's a bet on a technological breakthrough that no one has proven can be achieved. If the money runs out before the breakthrough, the most likely outcome is that the big players absorb the small ones. Microsoft, Google, and Amazon didn't start with AI, and they have fallback plans upon fallback plans. They also have the balance sheets to buy up what they can't afford to let fail. The AI revolution would simply end with most AI tools converging into properties of three companies instead of 20. What won't be absorbed is the cost that was already distributed outward. The increased electricity bills will stay. Utility companies will have already found the next buyer and persuaded the population that that's the cost of getting electricity in 2030. As for the stock market, the investors who bought on the cusp of collapse and the failing 401k funds will be written off as investment losses. It'll be spread out across millions of people who never signed up for the bet in the first place. AI will survive the crash, but it'll be deployed more slowly, more practically, and more honestly by whoever has the capital left standing. But the gap between the world that was promised and the world that actually arrives will be paid for by the people who could afford it the least. The question isn't really whether the AI bubble will deflate. The question is who gets handed the bill and picks up the pieces. Big tech just burned through hundreds of billions building the infrastructure for AI. And the return so far? A fraction of the cost. Nowhere close to break even. It might be one of the most expensive bets in modern history. Beneath the polished interface is something most people never even think about. A physical system with real-world limits. Data centers burning through enormous amounts of electricity. Power grids pushed closer to their capacity just to keep everything running. And it relies on a hidden army of exploited human labor. You don't get to stay outside of it. You are already interacting with it every day. The world is already committed to a system it may not be able to stop. This is the trillion-dollar AI lie. Chapter 1, the receipt nobody wants to read. In 2024, analysts at Sequoia Capital posed a simple question. If AI companies keep spending the way they are right now, how much money would the industry need to make every year to justify it? Their answer was about $600 a year. That is not what AI companies are earning today. That's what they would need to earn for all of this to make financial sense. Right now, generative AI is already making real money, billions in annual revenue across companies like OpenAI and Anthropic. Analysts predict AI-driven profits may rise into the trillions by the mid-2030s. But a lot of the reported revenue isn't even profit at all. Anthropic recently scaled its revenue to hundreds of millions per month in 2025. It sounds incredible until you realize it was also expected to lose billions over the course of the year. In mid-2025, OpenAI secured $10 billion worth of funding, but since its operating costs were so steep, it was asking for another $8.3 billion just months later. And Elon Musk's xAI, it was reportedly burning through more than a billion dollars every single month just to keep the lights on. It is a serious problem, one that points to something people don't like to talk about. In traditional software, once you build a product, adding more users is cheap. That's why companies like Microsoft and Adobe can become insanely profitable. The cost of serving the next customer is basically nothing. AI, it doesn't work like that. Every time you ask it a question, it costs money, real money. The answer comes from a delicate interplay of computational power, electricity, and infrastructure, which means scaling doesn't automatically make things more efficient. In some cases, it does the opposite. The more people use the system, the more expensive it becomes to run. The largest companies and investment groups in the world have already committed with annual spending pushes past $400 billion. AI growth speculation has become a core engine of the S&P 500. It becomes even sketchier when you realize that the broader market built up of retirement accounts, index funds, and otherwise safe investments are all now deeply exposed to the whims of a technology that hasn't even proven it can pay for itself at scale. If the economics of generative AI's future are so uncertain, why does every corporate CEO in the tech world seem so hell-bent on staking everything on its future? According to a recent survey, roughly 90% of CEOs say AI will fundamentally change their companies by 2028. It's a huge number, but when you look at the actual financial data, the reality is brutal. Only 25% of AI initiatives are actually delivering their expected ROI. Nine out of 10 executives believe the technology is essential, but only about one in four can explain how it actually makes money. What once felt like a strategy now seems like corporate peer pressure, and you see it playing out in real time. Each company tries to outdo the next, rushing out AI initiatives, talking about AI-first strategies, buying up massive quantities of GPUs because not buying them looks worse. If this amounts to the corporate version of a gold rush, nobody wants to be the one who showed up without the shovel. And that's where we're at right now. Infrastructure is being built up at full speed. The spending is already committed and seed money raised, but the returns are barely keeping pace. Chapter 2: Success that makes you poorer. Over the past few years, a quiet industry has emerged to support AI systems. Things like data labeling, content moderation, and output verification have all grown as the core tasks that make models usable in the real world. They're basically the difference between a raw system that produces chaotic, inappropriate outputs, and one that can answer questions without embarrassing the company that built it. All of that work happens outside of the model itself in offices and call center style environments in places like Kenya, the Philippines, and India. Thousands of workers spend their days correcting the mistakes that AI systems still make constantly. In some documented cases, they're paid just a few dollars an hour to filter out violent or explicit content. According to one report, content moderators in second and third world countries navigate a combination of psychological trauma, poverty wages, and the suppression of union organizing conditions considered intolerable under Western labor laws. It sounds less like a futuristic breakthrough and more like something out of the 19th century. And that's because in some ways it is. The marketing language around AI suggests autonomy, machines that can think, learn, and act replacing human effort altogether. The reality is closer to something older and more familiar. It's a layered system where the most visible tip of the iceberg is clean and efficient and the least visible part, the system's inner core, is messy, labor intensive, and easy to ignore. That doesn't mean the technology is fake, but at second glance, the word artificial is doing a awful lot of heavy lifting. A large portion of the work that makes these systems run is still very human. That human system doesn't disappear as the system scales. If anything, it just becomes more important. When the system relies on both massive compute infrastructure and ongoing human input, then the cost structure starts to look very different than the one most people imagine. Rather than a clean self-improving machine, you get something closer to a hybrid, part automated, part manual support, both constantly maintained. At the beginning of the AI boom, researchers found out that a huge percentage of companies calling themselves AI startups weren't actually using any meaningful AI in their core products. In some cases, it was as high as 40%. That didn't necessarily mean fraud, but it did hint at something even more important. From the very beginning, AI became a signal, a way to attract funding, justify higher valuations, and position themselves in a market where everyone was suddenly supposed to have an AI story. By 2024, roughly 78% of all companies reported using AI in some form. Just a year later, 61% of all global venture capital was flowing into AI-related businesses. You might expect that after all this adoption, all this investment, we would be seeing clear, consistent returns. Instead, the gains are highly concentrated, with about 75% of the financial benefits reaped by just 20% of the companies. As it turns out, the majority of companies talking about AI aren't actually making that much money from it. They're experimenting, deploying features, and integrating tools, but not transforming their business like everyone thought. In the early days, companies claimed AI without really using it. Today, companies are using it, but they don't know how to make any money off it, and that is a much bigger problem. Because now the stakes are higher, the expectations are enormous, and the gap between what's actually happening and what was promised is getting harder to ignore. Chapter 3: The Energy Wall. The deeper constraint on all of this isn't financial, but physical. For years, a simple comparison was repeated that a single AI query can use around 10 times the electricity of a standard web search. That number comes from early estimates, and like most simple comparisons, it's been argued over, updated, and stretched depending on who is making the case. But the part that hasn't really changed is the assertion that AI workloads are simply heavier. By 2022, data centers were already consuming roughly 460 terawatt-hours of electricity globally every year. That's about the same as an entire country like Germany or Japan, or 2% of total global demand. By this year, that number could land somewhere between 620 and 1,000 terawatt-hours, depending on how aggressively AI keeps growing. When you look into a data center, how that energy is used is somewhat surprising. 40% of the electricity is used for actual computing, the AI doing its thing part. Another 40% is used to just cool the machines down. The remaining 20% goes to everything else, moving the data around, keeping the system stable, basically making sure the whole operation doesn't collapse under its own complexity. So, nearly half of the energy we're pouring into AI isn't making it any smarter. It's just keeping it alive. Now, zoom out one level. A single large hyperscale data center, the kind companies are building for AI, can consume 100 megawatts of power or more. 1 megawatt can typically power over 150 US homes. 100? Over a year, that's akin to the electricity needed to charge well over 200,000 electric vehicles, just for one building. And there are roughly over 8,000 data centers worldwide with a third of them sitting in the United States alone. Worryingly, studies are now showing that the vast majority of these are located in climates considered way too hot for efficient operation. It's time to stop thinking about the cloud as something abstract. We're talking about a growing network of massive, power-hungry facilities clustered in specific regions pulling from the same grids that supply homes, schools, and businesses. The demand isn't close to being evenly distributed, either. In areas with concentrated pressure, city officials are forced to make trade-offs in real time. Do you expand the grid? Do you raise power prices? Do you slow down development and limit job growth? None of these are easy answers, and none of them deal with the issue of water, either. Remember the 40% of energy used to simply cool these mammoth data centers? That often depends on huge volumes of water moving through the system, constantly cycling to carry the heat away. A single A data center can use millions of gallons of water per day, about the same as a town of 30 to 50,000 people. Over a year, even a mid-size facility can burn through around 100 million gallons just to stay cool. And most of that water does not come back. In many systems, 70 to 80% is lost to evaporation, effectively disappearing into the air. Meanwhile, roughly 75 to 90% of data centers rely on water-based cooling, often pulling from the same rivers and municipal supplies that serve local communities. In at least one Oregon town, a single company's data center consumed over 25% of the city's water supply. The same system that promises infinite scale is drawing from very finite supplies. Power grids that expand overnight and water systems are already under pressure. The more the system grows, the more it pulls, and right now it's not obvious where the ceiling is. And while all of that strain is building in the physical world, something else is happening inside of companies. Because while AI is pulling more from the outside, it's also quietly pulling something from the inside, their data. Chapter 4, security self-sabotage. According to recent data, about 34.8% of employee inputs into AI now contains sensitive information. That's up from just over 10% in 2023. A third of everything being pasted into your AI chatbot of choice are legal documents, customer data, medical records, source code, and contracts. What's worse, 83% of the companies doing this have absolutely zero technical controls in place to stop it. They can send out policy emails until the cows come home, but the next day they'll turn around and tell those same employees to be faster, more productive, and efficient. The employees now have to decide, deadline or policy? They pick the deadline every time. The scale of this security self-own is starting to look like a slow-motion disaster. Over 225,000 ChatGPT credentials have already been found for sale on dark web marketplaces, often harvested from compromised machines or reused passwords. Major companies like Apple, JP Morgan, and Goldman Sachs have either restricted or banned tools like chat GPT internally after realizing what was happening. Samsung learned the hard way. In early 2023, three employees read an email lifting a previous chat GPT ban and then thought, "We're home free." The first went and uploaded proprietary source code to debug a problem they were having. Another copied notes from an internal company meeting into their chat feed, while a third used chat GPT to identify defective equipment in a semiconductor line. Pretty soon, the ban was back and internal survey quickly showed that 65% of Samsung employees thought AI tools were a security risk. The most unsettling part of all this is what most people misunderstand about the core disconnect between user privacy and the way AI systems are designed to work. AI models are trained on massive data sets. On consumer plans like OpenAI's free and plus version, consumers allow the company to use their conversations to train future models by default. Sure, you can opt out by navigating a maze of settings, but most users aren't even aware the toggle exists. Until you do, every document, every contract snippet, every client detail you type becomes potential training material, warns one report. Once trade secrets are used in training or processing an AI model, the damage can be effectively permanent. You can't just delete it. It's like trying to unmix paint. That's what makes this far more sinister than a traditional data breach. Those are rare, visible, and fixable for the most part. Compromising user security is normal for AI companies. HIPAA non-compliance is built into the workflow. That's why some security researchers are already starting to describe this as the largest uncontrolled corporate data leak in history. When you talk to the people inside of these companies who are actually responsible for what's going on, they know exactly what's happening. They know the controls to stop employees from pasting sensitive data into their AI tools aren't in place. They know the risk, but they also know there isn't an easy fix. You can't just spin up a secure in-house version overnight. That takes millions of dollars of hardware, specialized teams, and time most companies don't have. In the meantime, the pressure to move faster doesn't go away. People will still smirk at their company's enterprise version of Copilot, which only allows them to polish emails and go to Google searches. They'll sneakily use their consumer version of ChatGPT, and their older executive bosses won't care for the most part. They'll still see AI models as glorified search engines. If there's one thing people should have realized years ago, it was never to trust Silicon Valley bros with your intellectual property. Today, everyone is riding that productivity wave and hoping it doesn't come back to bite them. Chapter 5: The Great Correction. At this point, we know things aren't adding up. The system is expensive, it's resource-intensive, leaking data like a sieve, and still not making consistent enough money for its investors. So, why is the industry still all in on the AI boom? The answer is because from the inside, none of this feels like a choice. It feels like an arms race. The constant warnings to not fall behind or let China catch up has everyone moving at an urgent pace. It feels existential, but the reality is it isn't. The real bottleneck of today's so-called AI arms race is semiconductors or chips, and the company selling those chips, especially Nvidia and AMD, are making out just fine. The global semiconductor industry has been going gangbusters for years now. In 2026, the industry is expected to earn almost $1 trillion dollars annual sales, an all-time high. Analysts are predicting annual sales of 2 trillion dollars by 2036. According to the CEO of Taiwan Semiconductor Manufacturing Company, or TSMC, the single most important chip manufacturer in the world, demand for advanced AI chips is currently running at three times the global supply. Every major tech company is trying to lock in as much capacity as possible years in advance. New factories in Arizona and Japan won't meaningfully ease the market pressure until 2027 or later. Let's be clear, chip manufacturers want companies to believe there is not enough compute, that there will never be enough compute. If you don't buy everything right now, you'll inevitably fall behind in this computational arms race. That message doesn't need to be true to be powerful. It just needs to be repeated often enough, and it is repeated on annual earnings calls, at conferences, even in front of Congress. This is how you end up with firms ordering billions of dollars of GPUs years in advance, locking in the supply that they might not even be able to fully use yet. That's how you get a market where demand is running far in excess of supply. And while chip companies end up delirious with cash-filled pockets, consumers are left footing the bill. For months now, the AI arms race has bled over into a full-blown shortage of specialized memory those chips need to function. People call it RAMageddon. The result is a squeeze on everything consumers need. Laptops, phones, even appliances cost more. And somewhere out there, an ordinary guy just trying to build a new gaming PC is wondering how a couple of sticks of DDR4 suddenly cost more than his graphics card. Today, AI systems are built on two assumptions. First, that AI demand will keep growing fast enough to absorb all the infrastructure it requires. And second, that productivity gains will arrive quickly enough to justify the cost. If either of those assumptions slip, the whole promise of an AI boom starts to wobble. We've seen this before in the late 1990s. Companies built out massive internet infrastructure with fiber servers and networks filling entire buildings on the assumption that demand would catch up. Eventually it did, but not before the market corrected hard. The NASDAQ lost around 76% of its value. Companies like Cisco, Intel, and Oracle saw stock prices tank overnight. Other companies like eBay and Amazon barely managed to survive. It took the NASDAQ index 15 years to reclaim its previous high. The internet boom wasn't fake, but the timeline for dramatic overvaluation and hype was. The market had to go on life support as a result. The so-called dot coms just couldn't turn the profit their investors had dreamed of when they poured their cash into the startups in the 1990s. It's not hard to imagine a similar correction coming today. This time for the largest companies in the world. The money fueling today's AI boom isn't just venture capital chasing upside. It touches everyone. When those cycles turn, the people making the decisions rarely take the hit. Executives will still get paid and early investors find their exit. The loss is spread outward. So, when you hear AI arms race, it's worth asking, race to what? Because right now it looks less like a race to the future and more like a scramble to justify the billions already spent. If the returns don't show up fast enough, the fallout won't stay in tech. It'll end everywhere. And if you think the biggest risk is money, power, or data, you might be underestimating the real problem. What happens when those machines decide a life is expendable? Find out an AI just tried to murder a human to avoid being turned off.