The AI industry has spent $1.4 trillion. Only Nvidia is profitable.
Amazon, Google, Microsoft, and Meta have spent $1.4 trillion building AI infrastructure since 2022. They have earned back roughly half. The only major AI company actually making money is Nvidia.
AI bubble burst incoming?
The artificial intelligence industry has spent roughly $1.4 trillion building infrastructure since the launch of ChatGPT in late 2022.
It has earned back somewhere between $613 billion and $718 billion in revenue, depending on which week you check the running tracker at isaiprofitable.com.
The gap is the biggest spend-versus-revenue mismatch in modern tech history.
The only major AI-related company that is currently turning a real profit is Nvidia, which sells the graphics processing units (the specialized computer chips that all AI models run on) that everybody else is buying. The most recent Nvidia earnings showed $57 billion in quarterly revenue, $31.9 billion in net income, and a market cap above $4.5 trillion.
Every other major AI player is losing money. A lot of it. And the part of the story that affects regular people the most is now starting to land. Your cheap AI subscriptions are about to stop being cheap.
What the tracker actually shows
The tracker at isaiprofitable.com pulls capital expenditure and AI revenue numbers from each company’s own SEC filings and earnings reports. The headline numbers as of late May 2026:
Amazon: $273 billion spent on AI infrastructure, $40 billion in AI revenue
Alphabet (Google): $287 billion spent, $60 billion in revenue
Meta: $230 billion spent, $3 billion in revenue
Microsoft: $205 billion in the red despite the OpenAI partnership and the Copilot rollout
OpenAI: $1.7 billion in annual revenue, still unprofitable when compute costs and the revenue-sharing deal with Microsoft are factored in
Nvidia: profitable at every level, with the data center business alone hitting $51.2 billion in a single quarter
The pattern is clean. The companies building AI services for consumers and enterprises are losing massive amounts of money. The company selling them the chips to do it is printing cash.
The economics work the way an old gold rush worked. The miners spent themselves bankrupt. The companies selling shovels got rich.
Why this is going to hit normal consumers
For most users, AI subscriptions have been one of the best deals in modern technology. Twenty dollars a month for unlimited ChatGPT access. Free Claude. Free Gemini. Free Copilot. Free Perplexity. Free image generation. Free coding assistance. Free everything, basically.
That era is ending. The signs are everywhere.
In January 2025, OpenAI CEO Sam Altman publicly admitted that his company is losing money on its $200-per-month ChatGPT Pro subscriptions. “People use it much more than we expected,” Altman wrote on X. “I personally chose the price and thought we would make some money.”
He was wrong. The Pro plan, with its access to advanced models, is so expensive to run that the $200 monthly fee does not cover the compute and energy costs.
In March 2026, the head of ChatGPT at OpenAI called the current pricing model “accidental“ and said it would “significantly evolve.” A new $100-per-month “Pro Lite” tier was discovered in the ChatGPT web app’s code, slotting between the $20 Plus plan and the $200 Pro plan.
In April 2026, The Information reported that OpenAI projects an 80 percent decline in $20-per-month ChatGPT Plus subscribers, from 44 million in 2025 to just 9 million in 2026. The company plans to make up the gap by growing its cheaper, ad-supported ChatGPT Go tier ($8 a month in the U.S., $5 in some other markets) from 3 million subscribers to 112 million.
In other words, OpenAI’s own forecasts say the affordable, ad-free middle tier is dying. The future is either a much cheaper plan with ads or a much more expensive plan without them.
In April and May 2026, GitHub announced it was moving Copilot, the popular AI coding assistant, from flat subscription pricing to usage-based pricing, effective June 1, 2026. That change took effect this week.
Companies are reporting their own AI cost surprises. Uber’s chief technology officer told staff in an internal memo that the company burned through its entire 2026 AI budget in just four months.
This is what subsidized pricing collapsing looks like in real time.
Anthropic is the exception
Among the major AI companies, Anthropic, the maker of Claude, has been positioning itself differently.
Anthropic is expected to generate roughly $30 billion in revenue in 2026, more than OpenAI. It spends substantially less on data centers. It derives most of its revenue from corporate and government clients rather than $20-per-month consumer subscriptions. It has publicly committed to not putting advertising in its products, with co-founder Dario Amodei repeatedly stating that the company will not pursue ad revenue.
That positioning is paying off. Anthropic’s consumer pricing has remained more stable. Its enterprise contracts are larger and more predictable. The company is one of the few in the AI race that has a credible business model that does not depend on either subsidized consumer pricing or ad-supported scale.
For users of Claude specifically, this means the path forward is probably more predictable than it is for users of ChatGPT, Gemini, or Copilot. The trade-off is that Claude does not have the same free tier ambition that OpenAI is currently running.
What this means for casual users
If you have been using free or cheap AI tools for the past two years, the next 12 months are going to feel different.
Free tiers will get worse. OpenAI is reportedly cutting free-tier capacity to push more users toward paid plans. Google has begun rate-limiting Gemini’s free tier. Meta has not significantly invested in consumer subscription products and may exit the consumer chatbot business entirely.
Paid tiers will get more expensive. OpenAI’s Plus tier is probably going to either rise in price, lose features, or be replaced by tiered pricing. ChatGPT Pro at $200 is probably going up, not down. Other services are likely to follow.
Usage-based pricing is coming for everyone. GitHub Copilot’s switch on June 1 is the first major AI service to make the move. Other services will follow within the next year. Power users will pay more. Light users may pay less. Predictability will decrease for everyone.
Ads will appear in more AI products. OpenAI started inserting ads in the free version of ChatGPT earlier this year. The ad-supported ChatGPT Go tier is the company’s main growth strategy for 2026. Axios has reported that AI advertising could be worth as much as $100 billion a year by 2030.
Premium “no ads” tiers will become a differentiator. Anthropic’s no-ad commitment positions Claude as the option for users who value a clean product. Other companies may follow.
For the average person who has been getting accustomed to using ChatGPT, Claude, or Gemini multiple times a day for free or for $20 a month, the math is changing. Within the next 18 months, expect to either pay more, see ads, accept tighter usage limits, or some combination of all three.
The bigger picture
The AI industry’s current pricing was always going to end. The question was when.
The companies building AI infrastructure were betting on three things. First, that consumer demand would grow faster than compute costs. Second, that ad revenue or enterprise contracts could subsidize cheap consumer pricing. Third, that they could capture enough market share at low prices to lock in users before raising prices later.
The first bet did not pan out as planned. Compute costs have stayed high. Energy costs are rising. Users are consuming more compute per query than was expected, partly because new “reasoning” models like OpenAI’s o1 and Anthropic’s Claude Opus use far more compute per response than earlier models.
The second bet is partially working. Anthropic and Microsoft have built credible enterprise businesses. OpenAI is rapidly scaling its enterprise and government contracts. But the consumer side has not scaled enough to subsidize the infrastructure costs.
The third bet, the market-share land grab, is now under pressure as the companies start raising prices to close the cost gap. Users who joined for cheap unlimited access are now being asked to pay more or accept ads. Many of them will simply stop using AI products as often.
Nvidia CFO Colette Kress said on a recent earnings call that the company has “visibility into $500 billion in spending on its most advanced chips over the next 14 months” and projects $3 to $4 trillion in annual industry spending on AI infrastructure by the end of the decade. The chips need to be paid for somehow. Either subscribers pay, advertisers pay, or someone defaults.
The current AI boom looks structurally similar to the late-1990s telecom buildout that ended in the dotcom crash and the long-tail collapse of telecom companies, even though the underlying technology (the internet) turned out to be transformational. The companies that built the wires went bankrupt. The infrastructure they built powered everything that came after.
That pattern is the cautionary tale every AI investor is watching. The technology may be real and transformative, and the companies funding the current buildout may still go bankrupt.
For most users, none of that matters yet. What matters is that the $20 ChatGPT subscription is going to cost more. The free Gemini queries are going to be rationed. The Copilot bill is going to surprise people. And the era of artificially cheap AI access is ending.
Enjoy the cheap subscriptions while they last. They are not coming back.
Article compiled and edited by Derek Gibbs (entertainment editor) and the Clownfish TV newsroom.
Clownfish TV is your source for news, views, and rants on gaming, tech, and pop culture. Watch the show on YouTube at @ClownfishTV where new episodes drop daily. Subscribe to the Clownfish TV podcast on Apple Podcasts, Spotify, iHeart, and wherever else you get your podcasts. Sign up for the free newsletter at more.clownfishtv.com.
Hat Tips:
isaiprofitable.com, primary aggregated AI industry capital expenditure and revenue tracker
Yahoo Finance and Gadget Review (May 2026), verified breakdown of the AI industry $1.4 trillion spend including the company-by-company specifics for Amazon, Alphabet, Meta, Microsoft, and OpenAI
Peq42 (May 2026), additional verified analysis of the tracker including the $613 billion revenue figure
Andrew Baker (May 23, 2026), updated $718 billion revenue figure and the “shovel sellers vs gold diggers” framing
Nvidia Q3 FY2026 earnings (November 2025), verified $57 billion quarterly revenue, $51.2 billion data center revenue, $31.9 billion net income, and $4.5 trillion market cap
AOL / Nvidia investor materials, verified Colette Kress quotes on the $500 billion 14-month visibility and the $3 to $4 trillion annual industry projection by 2030
Futurism (January 2025), Sam Altman’s verified ChatGPT Pro losing money admission including the “people use it much more than we expected” X post
WinBuzzer (March 18, 2026), verified head-of-ChatGPT “accidental” pricing model quote and the discovered $100 “Pro Lite” tier
The Information (April 28, 2026), OpenAI’s verified projection of 80% ChatGPT Plus subscriber decline from 44 million to 9 million in 2026
wheresyoured.at (April 28, 2026), Ed Zitron analysis of the verified ChatGPT Plus projection and the ChatGPT Go projected 112 million subscriber growth
Boston Globe (April 30, 2026), verified ChatGPT Go pricing data including the $8 U.S. and $5 international tiers and the projected $100 billion 2030 AI advertising market
The Information / The Outpost (May 2026), Uber CTO Praveen Neppalli Naga’s verified internal memo on burning through the 2026 AI budget in four months
GitHub corporate announcement (April 27, 2026), verified Copilot transition to usage-based pricing effective June 1, 2026
Stanford AI Index 2026 (2026), verified industry analysis on AI progress, costs, and public trust trends
Bloomberg, Variety, CreditSights (2025-2026), additional verified industry capex and revenue context