The AI bubble already popped

Everyone keeps asking when the AI bubble will burst. They're scanning the horizon for a dramatic crash, a Lehman Brothers moment, a single day when the market wakes up and realizes it was all a fantasy. Here's the thing: the bubble already popped. You just didn't hear it. That's because most bubbles don't end with a bang. They end with a slow, boring deflation that nobody writes headlines about. And that's exactly what's happening right now.

The pop nobody noticed

The AI hype cycle peaked somewhere in mid-2024 to early 2025. That was the era of "put AI in everything," when slapping a chatbot on your SaaS product could double your valuation overnight. Every pitch deck led with "AI-powered." Every earnings call mentioned large language models. The bubble wasn't in the technology itself. It was in the narrative. And the narrative has been quietly deflating ever since. GIC's group chief investment officer, Bryan Yeo, said it plainly at the Milken Institute Asia Summit in 2025: "Any company startup with an AI label will be valued right up there at huge multiples of whatever the small revenue is. That might be fair for some companies and probably not for others." That kind of candid skepticism from a sovereign wealth fund is a signal. The smart money already knows.

The signals are everywhere

If you look at the data, the deflation is unmistakable. Startup shutdowns are maturing, not collapsing. SimpleClosure's 2025 report on startup wind-downs found that this isn't a year of collapse, it's a year of maturation. The companies shutting down are older, have raised more capital, and are further along in their lifecycle. Series A shutdowns jumped from roughly 6% to 14% of all closures, a 2.5x increase year-over-year. The correction has moved from failed ideas to failed business models. The tourists are leaving. High-profile AI startups have folded: Builder.ai ($445M raised), Adept AI ($415M), Jasper AI ($143M). MIT's State of AI in Business 2025 study found that most enterprise AI projects couldn't scale and failed to create tangible impact. The "put AI in everything" companies that had no real product-market fit are quietly dying. AI-washing is fading. In 2025, AI was cited as a reason for more than 54,000 layoffs, according to Challenger, Gray & Christmas. Even Sam Altman called it out, suggesting that companies were using AI as a scapegoat for cuts that had nothing to do with the technology. When the CEO of OpenAI is telling you the hype is overblown, you should listen. Valuations are compressing. Venture capital deal flow dropped 23% in 2025 even as total funding rose. That means fewer, larger checks going to fewer companies. The spray-and-pray era of AI investing is ending. Capital is concentrating in the handful of companies that have actual revenue and actual users.

The dot-com parallel everyone misquotes

People love comparing the AI boom to the dot-com bubble, but they always get the lesson wrong. They point to Pets.com and Webvan and say, "See? It was all nonsense." But the dot-com era also produced Amazon, Google, and eBay. The real lesson isn't that the dot-com bubble was irrational. It's that the bubble's deflation was the best time to build. Amazon didn't turn a profit until Q4 2001, right in the middle of the wreckage. It survived because it had a real business model built around genuine customer value, not hype. Google was founded in 1998 and grew through the crash because it solved a real problem that got more valuable as the internet expanded. The telecommunications companies of the late '90s laid over 80 million miles of fiber optic cable, most of which sat unused for years. That looked like waste. But that "dark fiber" became the backbone of the modern internet. The infrastructure outlasted the bubble. The same thing is happening with AI. The billions being poured into data centers, GPU clusters, and model training look excessive right now. But infrastructure built during a bubble doesn't disappear when the bubble deflates. It gets cheaper, more accessible, and more useful.

The spending is real, but spending isn't a bubble

Here's where the "AI bubble" discourse gets lazy. Critics point to the enormous spending numbers and declare it a bubble by default. The numbers are indeed staggering. Gartner forecasts worldwide AI spending to hit $2.52 trillion in 2026, a 44% year-over-year increase. The four major hyperscalers, Microsoft, Alphabet, Amazon, and Meta, are on track to spend upward of $650 billion on AI investments this year alone. Enterprise spending on generative AI jumped from $11.5 billion in 2024 to $37 billion in 2025. But spending is not the same thing as speculation. The question is whether the spend produces real infrastructure and real products, or whether it's just capital chasing a narrative. And here, the evidence is mixed but increasingly encouraging. Anthropic hit $14 billion in annual recurring revenue, up from $1 billion just 14 months earlier. Nvidia's 2025 revenue was $130.5 billion, a 114% increase from the prior year. Microsoft reported that AI contributed 16 percentage points to Azure's 33% year-over-year cloud growth. These aren't speculative bets. These are revenue numbers.

What to actually watch

The metric that matters isn't total AI spending. It's the ratio of AI spend to AI revenue. When that ratio starts converging, when the money going in starts producing proportional money coming out, you know the bubble is done deflating and what's left is an industry. We're not there yet. Free cash flow for big tech could drop significantly in 2026 as capital expenditure outpaces revenue growth. But the trajectory matters more than the snapshot. Revenue is growing, and it's growing fast. The other thing to watch is who's still building. During the dot-com crash, the companies that survived weren't the ones with the most funding or the flashiest products. They were the ones solving real problems for real users. The same filter is being applied to AI companies right now, and the ones passing it are the ones you'll still be using in 2030.

The human cost is real

None of this is meant to wave away the genuine damage. People have lost jobs, both from AI displacement and from AI-washing layoffs that had little to do with technology. Startups that raised hundreds of millions of dollars have shut down, leaving employees and investors with nothing. The deflation of a bubble always has casualties, and dismissing them as "creative destruction" is callous. The point isn't that everything is fine. The point is that the correction is already happening, and happening in the boring, gradual way that corrections usually do. No single dramatic crash. Just a slow separation of signal from noise.

The opportunity hiding in plain sight

Right now, the discourse is dominated by two camps: AI maximalists who insist the technology will change everything overnight, and AI skeptics publishing "when will the bubble pop?" thinkpieces on a weekly basis. Both are wrong, and both are creating an opportunity. While everyone is arguing about whether AI is overhyped or underhyped, the builders who survived the deflation are shipping products without competition from tourists. The companies with real revenue, real users, and real product-market fit are growing in the quiet space that bubbles leave behind. The best time to build during a technology wave has never been at the peak of hype. It's always been in the boring middle, the part where funding is harder to get, valuations are more rational, and the only people left are the ones who actually believe in what they're making. That's where we are now. The AI bubble already popped. The hype is fading. And the real work is just getting started.

References