Skip to content

TheAIGRID AI Business

AI Prices Are About to Shock Everyone

TheAIGRID makes one argument and hammers it from nine directions: the $20 a month you pay for AI is a subsidized loss, and it is about to end. The channel walks through why the price feels like a steal (because it is, by roughly 32 to 1), how OpenAI literally picked the $20 number off a four question Discord poll, and why every party that matters now wants that number to move. Sam Altman has said out loud that even the $200 Pro plan loses money. OpenAI is on track to lose about $14 billion in 2026 while committing over $1.4 trillion to infrastructure. xAI burns roughly $1 billion a month.

Published Jun 14, 2026 19:20 video 24 min read Added Jun 14, 2026 Open on YouTube →

At a glance

TheAIGRID makes one argument and hammers it from nine directions: the $20 a month you pay for AI is a subsidized loss, and it is about to end. The channel walks through why the price feels like a steal (because it is, by roughly 32 to 1), how OpenAI literally picked the $20 number off a four question Discord poll, and why every party that matters now wants that number to move. Sam Altman has said out loud that even the $200 Pro plan loses money. OpenAI is on track to lose about $14 billion in 2026 while committing over $1.4 trillion to infrastructure. xAI burns roughly $1 billion a month. IPOs are coming inside 18 months, and public investors do not have the patience of SoftBank.

The thesis is that AI is running the Uber playbook: subsidize the addiction, lock in the habit, then raise the price once you cannot leave. The mechanics are already visible, in Gemini quietly halving usage limits, in GitHub Copilot moving to metered tokens where a $28 bill becomes $700, in Uber and Microsoft burning whole annual AI budgets in months. The predicted endgame is a split: cheap commodity models stay near free, the smart frontier stuff gets metered and gets expensive, and the people who built their workflow around cheap AI are the ones left holding the bill.

This is an AI news and analysis video, so the claims below are reported with attribution. Where a number is a projection or a prediction market figure rather than a settled fact, the page says so.

The $20 that makes no sense

The video opens with the hook. For two years, $20 a month has bought something that, looked at squarely, almost makes no sense: a chatbot that writes code, plans your week, builds a business plan, edits your video script, and answers almost anything. It feels like a steal. And, the narrator says, that is exactly the problem. Every sign points the same way. The price is about to go up, and "that might be a lot."

The framing is important. This is not one company tweaking one plan. The claim is that the entire industry is running out of room to keep selling the future at a discount, and that the people running the labs are starting to say it out loud.

Concretely, what does the ChatGPT Plus tier deliver for that $20? The strongest models, image generation, file uploads, web browsing, and over a thousand messages a day if you want them. The video cites one write up: save just three hours a week and you get roughly $7,000 to $8,000 of productivity back on a $240 a year subscription. That is about a 32 to 1 return on a single tool.

The narrator flips the compliment into a warning. When industry watchers say there is room to triple your bill, the bill eventually goes up.

Figure 1. The value gap the whole video hangs on. By the estimate TheAIGRID cites, a $240 a year subscription returns roughly $7,500 in productivity from three saved hours a week. A gap this wide is not generosity; it is the size of the price increase the labs have room to take.

How $20 was actually chosen

The narrator's first piece of evidence that the number is arbitrary, and therefore movable, comes from Nick Turley, the head of ChatGPT. Turley has openly said the $20 price was not based on careful research. The team ran a quick survey on their own Discord to pick a number, and $20 stuck.

The video plays the clip. Turley describes shipping a Google form to Discord with "the four questions you're supposed to ask on how to price something," getting a price back, and that is how they landed on $20. He recounts that the next morning there was a press article along the lines of "you won't believe the four genius questions the ChatGPT team asked to price their product," and his reaction was "if only you knew." His point: building in extreme public means people read far more intentionality into your choices than actually existed. He says they were debating something slightly higher at the time, and he wonders aloud whether pricing it at $20 "erased a bunch of market cap," because so many other companies copied the $20 point. But he lands on not caring, because "the more accessible we can make this stuff the better," and because they can push capability down to the free tier semi regularly.

The narrator's takeaway is blunt: a price that was stumbled into is a price that can be moved.

You are being subsidized

Why does $20 feel like a steal? Because it is one. The video's claim is that your subscription is being subsidized. The labs are not really making money on you; in a lot of cases they are losing money on you, and the difference is made up with investor cash, from venture capital, sovereign wealth funds, and giant corporate partners.

The evidence is a direct quote from Sam Altman, CEO of OpenAI, made when the $200 a month Pro plan launched: "It is an insane situation. We are currently losing money on OpenAI Pro subscriptions. People use it so much, much more than we expected." Altman added that he personally picked the $200 price and thought it would be profitable, and it was not.

Sit with that, the narrator says. The top plan, at $200 a month, loses money. So the $20 plan that nearly everyone you know uses is certainly not paying its own way, especially for power users. The numbers offered for OpenAI as a whole:

Expected to lose around $14 billion in 2026 alone.
Not projected to be profitable until around 2030.
Committed to over $1.4 trillion in data center and infrastructure spending in the coming years.

So every time you ask ChatGPT to write an email, a giant pile of investor money helps pay for it in the background. That works only as long as investors stay patient. The moment they want their money back, the price has to change.

The Uber playbook

This is the analogy the video keeps returning to. Years ago, Uber rides were cheap, much cheaper than a taxi, and that was on purpose. Uber lost billions getting people used to opening an app instead of waving down a cab. Once everyone was hooked, prices climbed: surge pricing appeared, subscriptions appeared, pool rides disappeared, and the cheap rides quietly slipped away.

AI, the narrator argues, is running the same playbook: get people addicted at a loss, then raise the price once the habit is locked in. You can already feel it in the Reddit threads where regular users compare notes, people noting they were pulled in with heavily discounted, venture backed pricing, and now those same companies are quietly trimming plans, lowering limits, and dressing up the price hikes as upgrades.

Tech writer Alex, in a post bouncing around X and LinkedIn, puts it plainly: the companies making these models "are creating way more value than they're charging for," and "there is room to double or even triple the rates that we pay." The video also surfaces a meme from a user named Gout, a creator fully reliant on LLMs to code who imagines the price increased by a thousand percent, illustrating the same point: "your margin is my opportunity." Users are getting a tremendous amount of value, the companies are investing trillions, and they will eventually close the value gap by raising prices in a big way.

When AI costs more than the human

A jolt in the middle of the argument: AI is, in places, now costing more than the people it was supposed to replace. The video cites a CNN clip. A VP at NVIDIA was the first to flag it: "for months our costs for my team have been more for AI than humans." Then it came out in droves. Uber's CTO said he had already blown out his whole 2026 budget just on AI related costs, which means more than he was spending on human workers. Startup founders, the reporter notes, are now bragging about high AI bills.

The IPO clock

The next pressure point is the stock market. Both OpenAI and Anthropic are lining up to go public. Reports say OpenAI is laying groundwork for a listing as soon as late 2026; Anthropic is preparing its own; and prediction markets currently give Anthropic about an 80% chance of going public first. (That 80% is a prediction market figure, not a fact, and the listing dates are reported plans, not confirmed events.)

Why it matters: public investors are not as patient as private ones. A private backer like SoftBank will let you burn billions chasing a vision; a public investor wants a path to profit, shown on the quarterly earnings call. So the closer a lab gets to going public, the more pressure it is under to stop losing money on each user, and the easiest way to lose less per user is to charge more.

A viral X post lays out the math the video repeats: OpenAI targeting a roughly $1 trillion IPO valuation while losing about $14 billion this year; Anthropic reportedly preparing what could be a $900 billion private round before going public; and together with xAI, the three could pull $200 billion from global capital markets. That only makes sense on paper if the labs can show the price you pay going up while the cost to serve you goes down. As the narrator puts it, investor pressure is not vague, it is a calendar, it shows up on filing dates and earnings days, and it always lands on the Customer in the end.

xAI is the clearest math

Then there is xAI, Elon Musk's company, burning through about $1 billion every single month. The video cites Bloomberg: xAI is on pace to spend roughly $13 billion in 2025 while bringing in only about $500 million in revenue. That is spending more than $20 for every $1 earned. To plug the hole, xAI has raised about $34 billion in private equity and debt, including a $20 billion round in October 2025 that valued the company (at the time, when it was separate) at $200 billion.

No business survives that math forever. xAI says it is aiming for profitability by 2027, and getting there means one of two things: either the cost of running the models drops fast, or the price users pay climbs fast. The narrator's view is that in reality both have to happen. And xAI is not alone, OpenAI and every smaller lab are running a version of the same equation: get users to pay more while driving the cost per query down.

Figure 2. xAI's 2025 as reported by Bloomberg and cited in the video: roughly $13 billion out against roughly $500 million in. More than $20 burned for every dollar earned. The narrator's point is that no company runs this gap forever, so either inference cost falls fast or prices rise fast, and probably both.

Stealth hikes: Gemini

The video's first hard example of the increase already happening is Google Gemini. Users noticed in May that usage limits on the Gemini 3.1 Pro model were quietly slashed. The base plan dropped from three uses to two; the mid tier dropped from 30 uses to four. At the same time, Google added new Ultra plans at higher prices.

Google framed this as a reorganization of plans, but the narrator argues the effect is identical to a price hike: you get fewer top model messages unless you pay more. That is what a stealth price hike looks like, the number on the page stays the same while the value you get for that number shrinks, so the price per useful query rises while the marketing calls it an upgrade.

Why this is the loudest signal of all: Google has the most cushion in the industry. It owns its own data centers, designs its own AI chips, and prints cash from search. If Google is the one pulling back from free usage, the implication is that the math is tight, and if it is tight there, it could be tight everywhere.

Enterprise is breaking too

This is not just consumers complaining online; the strain is showing at the top of major companies.

Uber's chief operating officer, Andrew Macdonald, said it was getting harder and harder to justify the money the company spends on AI tokens. Reports say Uber blew through its entire 2026 AI budget by April, because individual engineers were racking up API bills as high as $2,000 per month each. An engineering workflow that looks like a normal tool is silently making thousands of API calls a day, turning a flat license into one giant utility bill.
Microsoft, with effectively infinite cloud compute, ran a pilot of Anthropic Claude Code across its experiences and devices team. The team burned through the division's entire annual AI budget in just a few months. Microsoft is shutting that program down on June 30th.

The narrator's logic: when the world's biggest enterprise customers say the math is too tight, the labs are stuck. They cannot keep raising enterprise prices without losing those customers, so more of the increase has to land on regular users, people paying $20 a month with no procurement department to push back.

He also notes Amazon, which uses Anthropic's models extensively, scrapped its internal AI leaderboard to stop workers chasing usage scores, and shifted to a metric called "normalized deployments" (evidence of engineers regularly using AI to create useful code) rather than measuring raw token consumption. The same article notes Anthropic itself shifting to consumption based pricing away from flat fees, which raised costs for some customers.

GitHub Copilot: the $28 bill that becomes $700

The cleanest dollar figure in the video comes from GitHub Copilot. The vendors are switching to AI credits, where every token in and out is metered, plus agent workflows that chew a lot of context. Previously you had a subscription and used it as much as you wanted; now it is usage based.

The video quotes GitHub directly: instead of counting premium requests, every Copilot plan will include a monthly allotment of GitHub AI credits, with paid plans able to buy more. Usage is calculated on token consumption, including input, output, and cache tokens, using the listed API rates for each model. GitHub says this aligns pricing with actual usage and is a step toward a "sustainable, reliable Copilot business," and that it launched a preview bill experience in early May so customers could see projected costs before the June 1 transition.

Then the punchline screenshot: one user's current billing was $28, but under usage based billing it would have been $700. GitHub's own framing is that "Copilot is not the same product it was a year ago, it has evolved to be an agentic platform," and since agentic is becoming the default it brings significantly higher compute and inference demand. A quick chat session and a multi hour autonomous coding session can cost the same; GitHub has absorbed much of that escalating inference cost, but "the current premium request model is no longer sustainable."

Figure 3. The GitHub Copilot example, the video's sharpest single number. The same usage that bills at $28 under a flat plan would be $700 once every input, output, and cache token is metered at API rates, roughly a 25 fold jump. This is what "the future has a meter" looks like on an invoice.

Washington and the compute supply

The last external pressure is regulatory. In March, Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez introduced the AI Data Center Moratorium Act, which would put a federal pause on building new AI data centers until safety, environment, worker, and civil rights protections are in place. Even if the bill never becomes law, the narrator notes that state policy makers are also looking at temporary pauses, and communities are pushing back as electricity bills climb and local water gets used to cool the racks.

Why it matters to your bill: AI runs on compute, compute runs on data centers. If new data centers get blocked or slowed, the total supply of compute stops growing as fast as demand. Squeezed supply plus climbing demand means price has to move, and that price flows straight into the per token cost the labs charge, then straight into your monthly bill. As the narrator puts it, a small slowdown in data center construction can translate into a much bigger jump in subscription cost for the average person.

He also cites a Substack from Josh Bersian, who argues there is no reason to think the tech is getting cheaper, "it's quite the opposite," and the price increase may change everything. Bersian's three questions, as relayed: Will AI job elimination rationalize itself? Likely yes, because nobody buys millions of dollars of Claude or Copilot if it costs more than doing it by hand, so everyone gets more discriminating about where they invest. Are some "low productivity, emotional humans" actually cheaper than the AI replacing them? Possibly. Will AI replace other things? Yes, but AI itself will have to become more strategic. His line the video lingers on: "the idea of giving everyone Claude might just stop," the way giving everyone a PC did not happen overnight.

The open source escape hatch that is not one

The obvious objection: if the labs charge too much, open source wins. The free models are getting good and, on paper, cost per token is far cheaper. The video grants the premise, then dismantles it.

In some studies, open models cost about 87% less per token than closed ones. So why do closed models still capture almost 80% of token usage and almost 90% of total revenue on platforms that let you pick either? Because it is a quiet trap. Open models are cheap per token but need far more tokens to do the same job. A study from Neus Research found open weight models use 1.5 to 4 times more tokens than closed models to solve the same problem, and up to 10 times more for simple knowledge questions. Open reasoning models in particular will spend hundreds of tokens to answer something that could be a single word.

This is the open source paradox the video names: cheap tokens, expensive thinking. The base model is free; the reasoning is not. And the new wave of AI products is almost entirely reasoning, agents that browse, plan, code, and act, each step a token, each chain long. So even as the price of one token falls, the total cost of a useful answer creeps up. The reasoning era is more expensive than the chat era, and that is the part most people forget.

Figure 4. Why open source does not cap the price. Closed models still take roughly 80% of usage and 90% of revenue even where users can pick either. The 87% per token saving is eaten by needing 1.5 to 10 times more tokens, so the bill lands in the same neighborhood. The base model is free; the reasoning is not.

The one force pushing the other way

To be fair, the video gives the counterargument its due. The cost of running a model at a given quality level is genuinely dropping. Venture firm Andreessen Horowitz coined "LLM inflation" to describe it: for an AI of equivalent quality, inference price is dropping roughly 10 times each year. Research firm Epoch AI puts the median decline even steeper, the cost of running inference at a fixed quality level halving roughly every 2 months.

So shouldn't prices be crashing? For older models, they are, the cost of a GPT-4 quality answer today is a tiny fraction of two years ago. But here is the catch: nobody stays on the old quality. Users keep climbing to whatever is newest and smartest, and that frontier tier is exactly where cost is highest, margins are thinnest, and labs have the strongest reason to charge more. The cheap stuff gets cheaper, the good stuff gets more expensive, and the average user moves with the good stuff.

How it ends: the split and the meter

Putting it together, the picture is clear: labs lose money on heavy users; CEOs say so openly; investors want their money back; IPOs land inside 18 months; xAI bleeds a billion a month; Gemini trims limits; Uber and Microsoft pull licenses because bills are too high; and the compute supply may get capped by data center bans.

The predicted outcome (and the narrator is explicit that this is a forecast) is a split. Basic everyday tasks get pushed onto small cheap models and stay close to free. The smart frontier stuff, deep reasoning, agents that do work for you, gets a lot more expensive. The $20 all you can eat plan probably does not vanish overnight; it slowly shrinks. You notice the newest model is only on a higher tier. You notice your usage limit dropped. You notice the agent feature you wanted is metered by tokens instead of included. As the narrator says, the frog gets boiled gently.

The real headline he leaves you with: the flat $20 subscription was a marketing tool to get you in the door. The future has a meter that runs while you sleep, while your agent does research, while your code editor calls the model in the background. And the people who locked their workflow around AI in the cheap years are the same people who will be holding the bill in the expensive ones.

Key takeaways

The $20 ChatGPT price was reportedly picked from a four question Discord poll, so it was never a researched, defensible number, and an arbitrary price is a movable one.
Your subscription is subsidized by investor money. OpenAI's $200 Pro plan loses money per Sam Altman, OpenAI is projected to lose about $14 billion in 2026, and it has committed over $1.4 trillion to infrastructure.
The strategy is the Uber playbook: subsidize the habit at a loss, then raise the price once switching is painful.
IPOs are the trigger. Public investors want quarterly profit, not vision burn, so listings inside 18 months force per user economics to improve, which means higher prices.
Stealth hikes are already here: Gemini quietly cut usage limits, GitHub Copilot moved to metered tokens where $28 of usage becomes $700, and enterprises like Uber and Microsoft burned whole annual budgets in months.
Open source is not the escape hatch it looks like. Tokens are about 87% cheaper but reasoning models need 1.5 to 10 times more of them, so total cost stays comparable.
Inference cost is genuinely falling (roughly 10x a year, or halving every 2 months at fixed quality), but users always migrate to the expensive frontier, so the average bill still climbs.
The likely endgame is a split: commodity AI near free, frontier and agentic AI metered and expensive, with the flat plan slowly hollowed out.

Chapters

Timestamps are clickable. Click one and the player jumps there and keeps playing while you read.

0:00 Why are AI prices about to go up?
0:33 Why are AI subscriptions so cheap right now?
1:11 How was ChatGPT's $20 price chosen?
3:16 Is OpenAI losing money on ChatGPT subscriptions?
4:34 Is AI using the Uber pricing playbook?
5:17 Will AI companies double or triple subscription prices?
6:06 Are AI tools becoming more expensive than humans?
6:35 Will OpenAI and Anthropic IPOs raise AI prices?
7:56 How much money is xAI losing every month?
9:21 Are Gemini usage limits being reduced?
10:05 Why are enterprise AI token bills exploding?
11:29 Why are AI companies switching to usage-based pricing?
12:09 What are GitHub Copilot AI credits?
13:26 Why are AI coding agents so expensive?
14:17 Will data center limits make AI more expensive?
15:39 Can open-source AI stop price hikes?
16:19 Why are open models cheap but still expensive to use?
17:15 Is AI inference getting cheaper?
18:22 What will happen to the $20 AI subscription?
19:01 Is pay-per-token pricing the future of AI?

Comparison: the pricing models in play

Model	Today (flat era)	Where it is heading (metered era)
ChatGPT Plus	$20/mo, over 1,000 messages a day, strongest models included	Frontier models likely gated to higher tiers; flat plan slowly shrinks value down
OpenAI Pro	$200/mo, losing money per Altman due to overuse	Pressure to meter the heaviest usage as IPO nears hike likely
Google Gemini	Base 3 uses, mid tier 30 uses on 3.1 Pro	Base cut to 2, mid tier cut to 4; new Ultra plans added stealth hike
GitHub Copilot	$28 flat for the same usage	$700 under token metering, input + output + cache billed ~25×
Open weight models	~87% cheaper per token	1.5 to 10× more tokens needed; total cost lands comparable no real saving
Old frontier (GPT-4 class)	Was expensive 2 years ago	Inference roughly halving every 2 months genuinely cheaper

Notable quotes

It is an insane situation. We are currently losing money on OpenAI Pro subscriptions. People use it so much, much more than we expected. narrator, 3:16 (quoting Sam Altman)

I got a price back, and that's kind of how we got to $20. narrator, 1:11 (quoting Nick Turley)

The companies making these models are creating way more value than they're charging for, and there is room to double or even triple the rates that we pay. narrator, 5:17 (quoting tech writer Alex)

For months our costs for my team have been more for AI than humans. narrator, 6:06 (quoting an NVIDIA VP via CNN)

Today, a quick chat session and a multi-hour autonomous coding session can cost the user the same amount. The current premium request model is no longer sustainable. narrator, 12:09 (quoting GitHub)

Cheap tokens, expensive thinking. The base model is free. The reasoning is not. narrator, 16:19

The flat, simple $20 a month subscription was just a marketing tool to get you in the door. The future has a meter that runs while you sleep. narrator, 18:22

The people who locked their workflow around AI in the cheap years are the same people who will be holding the bill in the expensive ones. narrator, 19:01

Resources mentioned

TheAIGRID, the channel, with its newsletter and community.
OpenAI, ChatGPT Plus, and the $200 Pro plan; Sam Altman's quote on losing money on Pro.
Nick Turley, head of ChatGPT, on the four question Discord poll that set the $20 price.
Anthropic and Claude Code, including the Microsoft pilot and the shift to consumption based pricing.
Google Gemini and the Gemini 3.1 Pro usage limit cuts.
xAI and Grok, with the Bloomberg spend and revenue figures and the October 2025 raise.
GitHub Copilot and its new AI credits token metering.
Uber as the pricing playbook analogy and as an enterprise cost example (COO Andrew Macdonald).
Microsoft Claude Code pilot and Amazon's "normalized deployments" metric.
Andreessen Horowitz and the "LLM inflation" concept; Epoch AI on inference cost halving every 2 months.
Josh Bersian's Substack article on AI prices, and tech writer Alex's viral post on X and LinkedIn.
The AI Data Center Moratorium Act, introduced by Bernie Sanders and Alexandria Ocasio-Cortez.
CNN segment on AI costs exceeding human worker costs, and Neus Research on open vs closed token efficiency.

The one thing to walk away with

The cheap years are a marketing budget, not a price. Every number in this video, the 32 to 1 value gap, the $14 billion loss, the $20 spent per $1 earned at xAI, the $28 bill that becomes $700, points the same direction: AI is currently sold below cost to build the habit, and the bill for that habit is coming due on a schedule set by IPO calendars and investor patience. The smart move the video implies is not panic but preparation: know your real productivity return, watch for limits quietly shrinking, do not weld your whole workflow to one provider at the promotional rate, and assume the meter is coming.

Full transcript

For the last 2 years, $20 a month has brought you something that, if you really stop and think about it, almost makes no sense. You get a chatbot that can write code, plan your week, build a business plan, edit your video script, answer almost any question you can ask, and it feels like a steal. And that is essentially the problem. Every sign right now is pointing to the same conclusion: the price you pay for AI is about to go up. There's going to be a price shock, and that might be a lot. Now, the story is not just about one company tweaking the plan. It is about the entire AI industry quietly running out of room to keep selling you the future at a discount. The cheap years are ending and the people running these labs are starting to say it out loud. Now, we have to think about what we actually have. Start with the $20 a month that unlocks today. On ChatGPT Plus, you get the strongest models, image generation, file uploads, web browsing, and, well, over a thousand messages a day if you want them. One write up estimated that if you just save three hours a week using it, you're getting around $7,000 to $8,000 of productivity back for a $240 a year subscription. That is like a 32:1 return on a single tool. Now, what's crazy about all of this is that that sounds like a compliment to the AI labs, but it's kind of a little bit of a warning to the rest of us. When a smart industry watcher says there's room to triple your bill, the bill eventually goes up. And when you look at the people running these companies, they're not exactly hiding the plan. So Nick Turley, the head of ChatGPT, has openly said that the $20 price was not based on careful research. The team actually ran a quick survey on their own Discord to pick a number and $20 a month just stuck. They basically said, OpenAI basically said, here, the price is something they stumbled into. And a price that was stumbled into is a price that can be moved, and every single sign right now says that the AI labs are getting ready to move it. Take a look at this clip here. So, what I did do is ship a Google form to Discord with, like, I think the four questions you're supposed to ask on how to price something. Yeah, exactly. Yeah, it literally had those four questions, and I remember distinctly, a, you know, I got a price back. Um, and that's kind of how we got to $20. But, b, uh, the next morning there was, like, a press article on, like, you won't believe the, like, four genius questions the chat team asked to price their, it was like, "if only you knew." So there's, like, something about building in this extreme public where people interpret so much more intentionality into what you're doing than, you know, might have actually existed at the time. But we got, with the 20. We're debating, you know, something slightly higher at the time. I often wonder what would have happened because so many other companies ended up copying the $20 price point. So I'm, like, did we, like, erase a bunch of market cap by pricing it this way? But ultimately I don't care because, like, the more accessible we can make this stuff the better. And I think this is the price point that in Western countries has been, um, reasonable to a lot of people in terms of the value that they get back. And, um, more importantly, we're able to push things down to the free tier, um, semi-regularly. And we always do that when we can, um, including with G2. Now, the reason $20 feels like such a steal is, of course, because it is. Right now, your subscription, if you hadn't realized, it's being subsidized. The AI labs are not really making money on you. In a lot of cases, they are losing money on you and they are making up the difference with investor cash. The cheap chatbot that you use every day is being paid for by venture capital, sovereign wealth funds, and giant corporate partners. Now, this isn't just a guess from random bloggers. Sam Altman, the CEO of OpenAI, said it himself when they launched the $200 a month pro plan. He wrote, and this is a direct quote, which you can see: "It is an insane situation. We are currently losing money on OpenAI Pro subscriptions. People use it so much, much more than we expected." And he even said that personally he picked the $200 price and thought it would be profitable, but it was not. Now, you have to think about that for a second. The top plan, the one that costs $200 per month, is losing OpenAI money. That $20 plan, that almost everyone you know, is certainly not paying its own way, especially for power users. OpenAI as a whole is expected to lose around $14 billion in 2026 alone and is not projected to be profitable until around 2030. And to stay open, it has committed to over $1.4 trillion in data center and infrastructure spending in the coming years. So every time you ask ChatGPT to write an email, in the background a giant pile of investor money is helping pay for it. Now that only works as long as those investors stay patient. The moment they want their money back, the price you pay has to change. Now this is where the story starts to feel familiar. So I'm pretty sure you all remember this. Years ago, Uber rides were cheap, way cheaper than a taxi. And that was on purpose. Uber lost billions of dollars getting people used to downloading an app instead of waving a cab. And once everyone was hooked, the prices climbed. Surge pricing showed up. Subscriptions showed up. Pool rides disappeared. The cheap rides quietly slipped away. AI is running the same playbook. Get people addicted at a loss, then raise the price once the habit is locked in. You can already feel this happening if you read the Reddit threads where regular users compare notes. People are pointing out that AI tools pulled them in with heavily discounted pricing backed by venture capital money. And now those same companies are quietly trimming plans, lowering limits, and dressing up the price hikes as upgrades. And tech writer Alex puts it plainly in a post that has been bouncing around X and LinkedIn. He said that the companies making these models are creating way more value than they're charging for and there is room to double or even triple the rates that we pay. This is one that I really like. It's from the username, is Gout. Here's the caption, and he's become fully reliant on LLMs to code. Now increased the price by a thousand percent. And I think this meme plays perfectly into what we were just talking, where I'm getting that the value of a couple hundred a month thumbnail guy for 20 bucks, uh, on ChatGPT right now. The saying is, your margin is my opportunity. Yeah. Clearly the price, to do anything with AI, we the users are getting a tremendous amount of value. These companies eventually are going to see that value gap, and they're, you know, they're investing trillions of dollars. They're going to raise the prices in a big way, don't you think? And you know what's crazy about this? I actually recently saw that currently the prices of AI is so high that it's actually costing more than humans. Take a look at this clip from CNN. I spoke with a VP at NVIDIA who first flagged this to me. He said, "Oh yeah, for months our costs for my team have been more for AI than humans." So that was the first flag. And then we started to hear this coming out in droves. Uber's CTO said he already blew out his whole budget for 2026 just on AI related costs. And obviously that means he's spending more on that than he's spending on human workers. And now I'm starting to hear, especially from startup founders, they're bragging about their AI bills being high because, kind of... And so the next pressure point for these companies is the stock market. OpenAI and Anthropic are both lining up to go public. Reports say that OpenAI is laying the groundwork for a public listing as soon as late 2026. Anthropic is also preparing its own listing, and the prediction markets currently give Anthropic about an 80% chance of going public first. Now, the reason this matters is simple. Public investors are not as patient as private ones. Private investors like SoftBank will let you burn billions chasing a vision, and a public investor wants to see a path to profit and they want to see that on the quarterly earnings call. And that means that the closer these labs get to going public, the more pressure they are under to stop losing money on each user. And the easiest way to lose less money on each user is to charge that user more. One viral X post has broken the math down. OpenAI is targeting an IPO, a roughly $1 trillion valuation, while roughly losing $14 billion this year. Anthropic is reportedly preparing what could be a $900 billion private round before going public. Together with xAI, these three companies could pull $200 billion from global capital markets. That is real unprecedented money. And the only way that makes sense is if the labs can show on paper that the price you pay is going up while the cost to serve you is going down. Investor pressure is not a vague thing. It is a calendar. It shows up on the filing dates and earnings days and it always lands on the Customers in the end. Now then there's xAI. Elon Musk's AI company is burning through about $1 billion every single month. Okay. Bloomberg reported that the company is on pace to spend roughly $13 billion in 2025 while only bringing in about $500 million in revenue. Now, that is not a small gap. That is a company spending more than $20 for every $1 it earns. To plug that hole up, xAI has raised about $34 billion in a mix of private equity and debt, including a $20 billion round in October 2025 that valued that company at the time, when it was separate, at $200 billion. Now, no business survives that math forever. xAI says it's aiming for profitability by 2027. But getting there means one of two things need to happen: either the cost of running these AI models has to drop fast, or the price users pay has to climb fast. In reality, both will have to happen. And xAI is not alone. OpenAI and every smaller AI lab are running some version of the same equation. They need users to pay more and they need the cost per query to come down at the same time. Now, take a look at this. Okay, I saw this post on Reddit that most people hadn't seen yet, but Google seems to already be moving. Some users have noticed on Gemini in May that the usage limits on the Gemini 3.1 Pro model were quietly slashed. The base plan dropped from three uses to two. The mid tier dropped from 30 to four and the mid tier was pulled back too. And at the same time, Google added new ultra plans at higher prices. Now, Google frame this as a reorganization of plans. The effect on the users is the same. You get fewer top model messages unless you pay more. And that is what a stealth price hike looks like. The number on the page stays the same. The amount of value you actually get for that number shrinks. The price per useful query goes up. And the marketing makes it all sound like an upgrade. Now, this matters because Google has the most cushion in the entire industry. Google owns its own data centers, designs its own AI chip, and they print cash from search. So, if Google is going to be the one pulling back from free usage, that kind of tells you about the math, that maybe it's a little bit tight, and everywhere else it could be tight, too. Now, the thing about this is that this isn't just users complaining about this online. The strain is now showing up at the very top of major companies. Uber's chief operating officer, Andrew Macdonald, recently said that it was getting harder and harder to justify the money the company is spending on AI tokens. And it gets worse. Reports say that Uber blew through its entire 2026 AI budget by April because individual engineers were racking up API bills as high as $2,000 per month each. An engineering workflow that looks like a normal software tool is silently making thousands of API calls a day, turning what used to be a flat license into one giant utility bill. And Uber is not alone. Microsoft, a company that has basically infinite cloud compute, ran a pilot of Anthropic Claude Code across its experiences and devices team. The team burned through the division's entire annual AI budget in just a few months. Microsoft is shutting that program down on June 30th. When the world's biggest enterprise customers are openly saying that the math on AI usage is too tight, the labs are stuck. They cannot keep raising prices on enterprise without losing those customers. Which means more of the price increase has to fall on regular users. People like you, people paying $20 a month who do not have a procurement department to push back. And here you can see Amazon actually recently scrapped their AI leaderboard to stop workers chasing usage scores. You can see that this article talks about the fact that AI labs such as Anthropic have recently shifted to a consumption-based pricing model away from flat monthly fees in a move that significantly increased the cost of some customers. And Amazon uses Anthropic's AI models extensively. And you can see here that Amazon has actually tried to change and started to use a metric called normalized deployments, evidence of engineers regularly using AI to create useful code, to measure the success of its AI tools and adoption of the technology, rather than just outright token consumption. And so this is also another example of this in practice. So here you can see someone on Reddit says that "bye-bye copilot, the new pricing looks to be a joke." So you can see that vendors are essentially, like, you know, GitHub Copilot, are switching to AI credits where every token in and out is metered, plus agent workflows that chew a lot of context. So previously you could just have your subscription and use it as much as you wanted, but now they're switching to usage-based tokens. So if you actually go on the GitHub website, it says: instead of counting premium requests, every Copilot plan will include a monthly allotment of GitHub AI credits with the option for paid plans to purchase additional usage. Usage will be calculated based on the token consumption, including input, output, and cache tokens, using the listed API rates for each model. And it says, "This change aligns GitHub Copilot pricing with the actual usage and is an important step towards a sustainable, reliable Copilot business and experience for all users. To help customers prepare, we're also launching a preview bill experience in early May, giving them visibility into the projected cost before the June 1 transition." And so this is what you can see in the screenshot here. Here you can see that their current billing was $28, but if they actually had the usage-based billing on, it would have been $700. Just showing you how much value you're getting from those cheap subscriptions and how much that price hike is going to be. And you can see it says here: "Copilot is not the same product it was a year ago, it has evolved to be an agentic platform. And since agentic is becoming the default, it brings significantly higher compute and inference demand. Today, a quick chat session and a multi-hour autonomous coding session can cost the user the same amount, and GitHub has absorbed much of the escalating inference cost behind that usage. But the current premium request model is no longer sustainable." Now to top all of this off, guys, the last piece of this story is happening in Washington. In March, the senator Bernie Sanders and representative Alexandria Ocasio-Cortez introduced the AI Data Center Moratorium Act, and that bill would put a federal pause on building new AI data centers until proper safety, environment, worker and civil rights protections are put in place. Now, even if that bill never becomes law, state policy makers are also looking at temporary pauses on new AI data center construction, and communities are pushing back as electricity bills climb and local water gets used to cool the racks. Now, this matters, okay? Because AI runs on compute and the compute runs on the data centers. And if new data centers get blocked or slowed down, the total supply of compute in the world stops growing as fast as the demand is growing. And when supply is squeezed and demand keeps climbing, the price has to move. And that price flows straight back into the per token cost the labs will charge, and from there it's straight into your monthly bill. A small slowdown in data center construction can easily translate into a much bigger jump in what AI subscriptions cost for the average person. And I found this Substack from Josh Bersian where he says there's no reason to think that this tech is getting cheaper. It's quite the opposite and this price increase may just change everything. You can see he talks about, in this article, there are many different things about why the price is going up, most of the things I've discussed in this video. But he answers three key questions here. Will the job elimination from AI rationalize itself? I think so. We won't be buying millions of dollars of Claude or Copilot if the cost is higher than doing it by hand. So, we're all going to get a little bit more discriminating about where we focus our AI investments. I keep reading about block eliminating engineers and managers, Meta creating a 50-to-1 span of control. But what if these low productivity, emotional humans are cheaper? And will AI replace other things? Yes, it will. But at the same time, AI is going to have to become more strategic. The idea of giving everyone Claude might just stop. Like the idea of just giving everyone a PC didn't happen fast. So the widespread adoption before value might slow down. And he already sees that happening. Now of course we cannot discuss this without the open source. Now you might be thinking, okay, if these labs are going to charge so much, well, open source is going to win. The free models are getting really good and on paper the cost per token is way cheaper than what OpenAI or Anthropic charge. And in some studies, what's actually true is that open models cost about 87% less per token than closed ones. So why do closed models still take almost 80% of token usage and almost 90% of total revenue on platforms that let you pick? Because it's actually a quiet trap. Open models are cheap per token, but they need a lot more tokens to do the same job. So a study from Neus Research found that open-weight models use one and a half to four times more tokens than closed models to solve the same problem, and up to 10 times more for simple knowledge questions. Open reasoning models in particular will sit there spending hundreds of tokens to answer a question that could be answered in a single word. This is the open-source paradox: cheap tokens, expensive thinking. The base model is free. The reasoning is not. And the new wave of AI products is almost entirely built on reasoning: agents that browse, plan, code, and act, all chew through long chains of thought. And each step is a token. So even as the price of a single token keeps falling, the total cost of actually getting a useful answer keeps creeping up. The reasoning era is more expensive than the chat era, and that's the part that most people forget. Now, to be fair, there is a real force pushing the price in the other direction. The cost of running an AI model at any given quality level is dropping. Venture firm Andreessen Horowitz coined the term "LLM inflation" to describe it. For an AI of equivalent quality, the price of inference is dropping by roughly 10 times each year. Research firm Epoch AI puts the median decline even steeper, with the cost of running inference at a fixed quality level halving every 2 months. Now that sounds like the prices should be crashing, not climbing. And for older models, they certainly are. The cost of getting a GPT-4 quality level answer today is a tiny fraction of what it was 2 years ago. But there is a catch in that math. Yes, the old quality gets cheap, but the problem is no one stays on the old quality. Users keep climbing to whatever is the newest, smartest model. And that frontier tier is exactly where the cost is the highest, the margins are the thinnest, and the labs have the strongest reason to charge more. The cheap stuff gets cheaper, and the good stuff gets more expensive, and the average user moves with the good stuff. So, when you put all of these things together, the picture's pretty clear. The labs are losing money on heavy users. Their CEOs are openly saying it. The investors want their money back. The IPOs are coming in the next 18 months. xAI bleeding a billion dollars. Gemini quietly trimming the limits. Big enterprise companies like Uber and Microsoft pulling licenses away because the bills are too high. And the supply of new compute might get capped because of the data center bans. The most likely outcome here is a split. Basic everyday tasks get pushed onto small cheap models and stay close to free. The smart frontier stuff, the deep reasoning, the agents that do work for you, that will probably get a lot more expensive. The $20 all-you-can-eat plan probably doesn't vanish overnight. It just slowly shrinks and you start to notice that the new model is only available on a higher tier. You start to notice that your usage limit dropped. You start to notice that the agent feature you wanted is metered by tokens instead of included in the flat price. And the frog gets boiled gently. So, we've already seen that the CEOs have already admitted that this is probably the next step, moving towards usage based pricing instead of unlimited access. And that is the real headline. The flat, simple $20 a month subscription was just a marketing tool to get you in the door. The future has a meter that runs while you sleep and while your agent does research, while your code editor calls the model in the background. And the people who locked their workflow around AI in the cheap years are the same people who will be holding the bill in the expensive ones.