Is AI a Bubble? Here Is What the Data Actually Shows

What Is a Bubble, Technically?

A bubble is when an asset's price detaches from what its underlying cash flows can justify, driven by expectations of ever-rising prices rather than fundamentals. Being expensive or risky is not enough to qualify. In a bubble the price and the math no longer agree, and the gap closes through a crash, a slow deflation, or fundamentals catching up.

Economists don't have one clean formula for a bubble, but the pattern repeats: prices rise faster than earnings can plausibly justify, new money floods in chasing the trend instead of fundamentals, and skeptics get dismissed as people who "don't get it." The dot-com bubble, the 2008 housing bubble, and the 1990s telecom buildout all fit this shape, and none of them meant the underlying idea was fake. The internet, houses, and fiber optic cable were all real, and in each case the bubble was in the price rather than the product.

That distinction matters for AI because "is AI a bubble" is really two separate questions that get mixed together all the time: is the technology overhyped, and are the stocks overpriced? You can answer no to the first and yes to the second, and most of the credible voices we found in 2025 and 2026 do exactly that.

What Is the Bull Case for AI Right Now?

The bull case is that enterprise AI adoption is real and accelerating. Over 80% of organizations have piloted tools like ChatGPT or Copilot and roughly 40% report full deployment. Anthropic's revenue went from 9 billion to 30 billion dollars annualized in four months in early 2026, and OpenAI is running near 2 billion dollars a month.

We start with usage because it is the hardest number to fake. According to MIT's 2025 study on enterprise AI, over 80% of organizations have piloted generative AI tools and nearly 40% report they've moved past piloting into deployment. Behind those percentages are millions of people opening ChatGPT, Copilot, or Claude at work every day.

Revenue points the same way. According to SaaStr's reporting, Anthropic's annualized run-rate went from about 1 billion dollars at the end of 2024 to 9 billion at the end of 2025 to 30 billion by April 2026, with over 1,000 enterprise customers each spending more than 1 million dollars a year. OpenAI crossed 2 billion dollars in monthly revenue around the same time, near a 24 billion dollar annualized run-rate. Companies pay invoices for this software every month, and the growth curve on both companies looks closer to 3x-to-10x per year than the flat line you'd expect from a fad.

Nvidia CEO Jensen Huang makes the strongest version of the bull case. In his view the spending is an infrastructure buildout meeting compute demand that is currently constrained, rather than speculative capital chasing a trend. "There's been a lot of talk about an AI bubble," Huang said in November 2025. "From our vantage point, we see something very different." He argues that GPU-driven infrastructure is displacing older computing across search, ads, and data processing before agentic AI applications have fully arrived, so the demand curve is still climbing.

What Is the Bear Case Against AI Right Now?

The bear case rests on three problems: a revenue gap Sequoia Capital sized at roughly 600 billion dollars a year in new revenue just to justify current infrastructure spending, circular deals where the same companies fund each other's purchases, and stock gains so concentrated in a handful of names that the S&P 500 has effectively become an AI bet.

Sequoia partner David Cahn has run this math twice now. In 2023 he called it "AI's $200B Question," the gap between what GPU spending implied in future revenue and what AI companies were actually generating. By mid-2024, as capex accelerated, he updated it to "AI's $600B Question": even generous assumptions about new AI revenue from Google, Microsoft, Meta, and others still left a shortfall in the hundreds of billions. That gap has only gotten harder to close since, because capex kept climbing faster than most AI product revenue.

The second problem is money that circles back on itself. Nvidia has committed up to 100 billion dollars to help finance OpenAI's data center buildout, and OpenAI has committed to spending that money on Nvidia chips. OpenAI separately signed a roughly 300 billion dollar cloud contract with Oracle, which is spending tens of billions of that on Nvidia GPUs to fulfill it. Microsoft has put roughly 11 billion dollars into OpenAI while also selling it cloud capacity. None of this proves the underlying demand is fake, but a meaningful share of the industry's headline revenue and investment figures is the same dollars changing hands between a small group of companies rather than new money from independent customers. The chart below shows the scale.

The third problem is where the stock market's gains came from. As of mid-2026, the Magnificent Seven make up roughly a third of the entire S&P 500's value, and in 2023 those seven stocks alone accounted for 62% of the index's total return. At that level of concentration, a large share of most people's retirement accounts, index funds, and 401(k)s amounts to a bet that seven companies will execute an unprecedented, unproven spending plan correctly.

What Do Respected Voices on Each Side Actually Say?

Most prominent voices hedge, and the people closest to the money hedge hardest. Sam Altman, Jeff Bezos, and Jamie Dimon have all said some version of "yes, there's a bubble, and also this is real and important," while Jensen Huang and most AI lab leadership reject the bubble framing outright. Few serious voices take a simple position either way.

What Can the Dot-Com Bubble Teach Us About AI?

The dot-com bubble proves that a real technology can still have a stock bubble. The internet reshaped the economy exactly as the optimists promised, and the Nasdaq still fell almost 80% between 2000 and 2002. Telecom firms overbuilt so badly that most fiber laid in the 1990s sat dark for years, and AI can follow the same script.

We think this is the most useful lesson in the whole debate, and it gets misused in both directions. "The internet was real, so the dot-com crash didn't matter" is wrong, and so is "the dot-com crash happened, so the internet was overhyped." The internet went on to justify every serious prediction made about it, while Pets.com, Webvan, and hundreds of other companies went to zero and trillions of dollars of market value evaporated.

Telecom capex during that era is a closer analogy to today's AI buildout than the dot-com stocks themselves. Telecom firms spent heavily on fiber optic infrastructure betting internet traffic would keep doubling every few months. It didn't grow nearly that fast, and by some estimates 85% to 95% of the fiber laid in the late 1990s sat dark for years afterward. Today's hyperscalers are making a similar bet on data centers and GPUs. Unlike a stock price, those physical assets don't disappear when sentiment turns; they sit and depreciate until demand catches up, if it does.

Did the AI Bubble Already Start Popping in 2026?

Not exactly, though a real correction hit. In June 2026 the Nasdaq dropped over 4% in a single session, semiconductor stocks lost over 1.3 trillion dollars in market value, and Microsoft and Meta briefly entered bear market territory. Most analysts called it a valuation correction driven by rate uncertainty and profit-taking rather than a collapse of AI demand or revenue.

The selloff deserves a close look because it's the most concrete data point available. It hit hardest in semiconductors and the biggest AI capex spenders, which is what you'd expect if investors were pricing in more risk around the AI buildout specifically. The coverage at the time was consistent, though: it looked like a repricing of stretched valuations after nine straight weeks of gains rather than evidence that AI usage or revenue was falling. By most accounts fundamental demand stayed intact, and whether that holds is unknown. A correction doesn't prove a bubble popped, and it doesn't prove one didn't, so we won't pretend to know what happens next.

Should You Buy AI Stocks or an AI ETF?

We won't answer that for you, because we are an AI lab and not a financial advisor, and this article is not financial advice. If you're researching AI ETFs, know that most of them concentrate heavily in the same handful of names discussed above, so buying an "AI ETF" often means buying the same concentration risk under a different label.

We're flagging this because "best artificial intelligence ETF" is one of the most searched phrases connected to this topic, and we'd rather explain why we won't answer it than pretend we can. Recommending specific stocks or funds requires knowing your personal financial situation, risk tolerance, and time horizon, none of which we have. What we can say, because it's a fact about fund construction rather than a prediction, is that many funds marketed as "AI ETFs" hold a heavily overlapping set of the same large-cap names covered in the concentration numbers above. A fund with that name is not automatically diversified, and whether to buy one is a question for a licensed financial advisor.

What Does This Mean If You Run a Small Business, Not a Portfolio?

Whatever happens to Nvidia's stock price, the demand we see for AI implementation among small businesses is real and immediate. A restaurant that automates its phone reservations doesn't care about hyperscaler capex ratios. The task-level economics of adopting AI and the stock market's pricing of AI companies are separate questions with separate answers.

Our own experience is more useful here than any market forecast. We build AI systems, chatbots, and automation for small and mid-size businesses, and the work we get hired for has nothing to do with whether Nvidia beats earnings next quarter. A local business that saves 15 hours a week by automating scheduling, or a service company that stops losing leads because its intake now runs on AI instead of a missed voicemail, gets that value whether the Magnificent Seven are up or down 20% that month. The debate over AI stock valuations has almost nothing to do with whether one automation saves one business real money.

If the stock market side of this is a bubble and it pops, the most likely outcome based on the dot-com precedent is that a lot of overfunded, undifferentiated AI startups disappear and capital gets more disciplined, while AI itself keeps working and the tools that already deliver real task-level value keep getting used and keep getting cheaper. That is what happened with the internet, which kept working through 2001 and 2002 while the froth cleared out.

So, Is AI a Bubble?

Partly. The technology and enterprise adoption are real and growing, but the valuations on a handful of AI-exposed stocks, the circular financing, and the concentration of market gains show classic bubble behavior. Both are true at once, as they were of the internet in 2000. This article is not financial advice, and we are not investment advisors.

To repeat what we said above: nothing in this article is financial advice, and Actual Intelligence Labs is not a registered investment advisor. We don't know whether AI stocks go up or down from here, and we wouldn't trust anyone who claims to know with certainty. What we do know, from building this technology for real businesses every week, is that the underlying capability is real. What happens to the stocks that price it is a separate question, and one for a financial professional rather than a blog post.