The first AI bubble didn’t pop with a dramatic crash heard around the world. It deflated quietly, in the shares of companies that sold Wall Street a vision of artificial intelligence riches and then failed to deliver the receipts. While Nvidia soared and the Magnificent Seven dominated headlines, a graveyard of AI-adjacent stocks was forming just beneath the surface — a warning that the market’s enthusiasm for anything stamped with an AI label has real limits.
The evidence is now impossible to ignore. According to Yahoo Finance, one AI bubble has already burst. Not the big one — not the infrastructure buildout anchored by Nvidia, Microsoft, and Alphabet — but the speculative fringe where companies with thin AI credentials rode the hype to absurd valuations. These stocks, many of them small- and mid-cap names, surged in early 2023 and into 2024 on little more than press releases and investor credulity. Now they’ve given back most or all of those gains.
Think of it as the dot-com era’s pets.com moment, except distributed across dozens of lesser-known tickers. Companies like BigBear.ai, C3.ai, and SoundHound AI became retail darlings, their share prices multiplying on the promise that AI would transform their businesses. BigBear.ai, a data analytics firm that went public via SPAC in 2021, saw its stock spike dramatically during the AI frenzy. It has since collapsed. C3.ai, founded by tech veteran Tom Siemens, watched its valuation swell past $4 billion at its peak hype — only to see persistent questions about revenue growth and customer acquisition drag shares back down. SoundHound AI, which makes voice-recognition technology, followed a similar arc: meteoric rise, then a slow bleed as investors demanded proof of sustainable business models.
The pattern is consistent. And familiar.
What separates this moment from a full-blown AI market crash is that the companies at the center of the AI infrastructure buildout — the ones actually generating massive revenue from the technology — continue to perform. Nvidia reported $26 billion in quarterly revenue in its most recent earnings, a figure that would have seemed hallucinatory two years ago. Microsoft’s Azure cloud business, supercharged by its OpenAI partnership, keeps growing at rates that justify its premium valuation. Amazon Web Services and Google Cloud are posting similar trajectories. The money being spent on AI data centers, chips, and cloud capacity is real, and it’s accelerating.
But the periphery tells a different story entirely.
The divergence between AI’s core beneficiaries and its hangers-on has become one of the most striking features of the current market. According to recent reporting by Reuters, institutional investors have been quietly rotating out of speculative AI plays and concentrating their bets on the handful of companies with proven AI revenue streams. This consolidation mirrors what happened in the late 1990s, when the internet’s real winners — Amazon, eBay, a few others — separated from the hundreds of companies that claimed the internet would change everything about their business but couldn’t show it on an income statement.
The comparison isn’t perfect. Today’s AI infrastructure spending dwarfs anything seen during the dot-com buildout. Capital expenditure commitments from the hyperscalers alone are expected to exceed $200 billion in 2025, according to estimates compiled by Bloomberg. That money flows directly to chipmakers, server manufacturers, power companies, and cooling system providers — creating a real economic multiplier that speculative dot-com spending never achieved. The question isn’t whether AI spending is real. It is. The question is how far down the food chain the benefits actually reach.
For many companies, the answer appears to be: not far enough.
Consider the enterprise AI software market, where dozens of startups and public companies promised that their platforms would help businesses deploy AI at scale. The reality has been far messier. Corporate adoption of AI tools has been slower than the hype suggested, hampered by data quality issues, regulatory uncertainty, integration challenges, and — perhaps most importantly — a lack of clear return on investment for many use cases. A recent survey by McKinsey found that while AI adoption rates have increased, most companies are still in pilot or experimental phases, not full-scale deployment. The revenue bonanza that enterprise AI vendors projected hasn’t materialized on schedule.
This gap between expectation and reality is where bubbles form and burst. The AI trade, in its broadest form, was never a single bet. It was a collection of bets at different levels of the technology stack, each with different risk profiles. At the bottom — the chip and infrastructure layer — the bet has paid off spectacularly. One layer up, in cloud services and foundation models, it’s paying off handsomely but with more uncertainty about long-term margins. At the application layer, where software companies promise to deliver AI’s benefits directly to end users, the returns have been deeply uneven. And at the very top — the pure speculation layer, where companies slapped “AI” on their investor decks without meaningful technology — the reckoning is already complete.
The numbers tell the story with brutal clarity. An index of AI-themed stocks tracked by various financial data providers shows that while the top five AI names are up dramatically over the past two years, the median AI-labeled stock has underperformed the S&P 500. Strip out Nvidia and Microsoft, and the AI trade looks far less impressive. This is the dirty secret of the AI boom: it has been extraordinarily narrow in its beneficiaries, even as its narrative captured the imagination of the entire market.
Wall Street analysts are starting to acknowledge this bifurcation openly. Dan Ives of Wedbush Securities, one of the most vocal AI bulls, has repeatedly emphasized the distinction between “AI winners” and “AI pretenders” in his research notes. His thesis is straightforward: the companies building and operating AI infrastructure will generate enormous value, but companies merely claiming AI exposure without differentiated technology or sticky customer relationships will get punished. That punishment, for many stocks, has already arrived.
So where does this leave investors?
The uncomfortable truth is that the AI investment thesis now requires far more discrimination than it did eighteen months ago. The era of buying anything with “AI” in the description and watching it go up is over. What remains is a more traditional — and more difficult — exercise in fundamental analysis. Which companies have real AI revenue? Which ones have durable competitive advantages? Which are burning cash on AI projects that may never generate returns? These questions matter now in a way they didn’t when euphoria was doing the heavy lifting.
The infrastructure buildout itself isn’t immune to scrutiny. Some analysts have begun questioning whether the pace of data center construction can be sustained, particularly given constraints on power supply and permitting. The Wall Street Journal has reported extensively on the growing strain that AI data centers are placing on the electrical grid, with some projects facing multi-year delays to secure adequate power. If the buildout slows, even the core AI beneficiaries could see their growth trajectories moderate.
There’s also the question of AI model economics. Training frontier AI models costs hundreds of millions of dollars, and the price isn’t declining as quickly as some technologists predicted. OpenAI, despite its enormous revenue growth, continues to burn cash at a staggering rate. Anthropic, its closest competitor, requires billions in outside funding to stay competitive. The foundation model business may prove to be a natural monopoly or oligopoly — great for the two or three winners, catastrophic for everyone else. And if the winners end up being the same hyperscalers that already dominate cloud computing, the investment implications are significant: the rich get richer, and the independent AI companies get squeezed.
None of this means AI is overhyped as a technology. The productivity gains from AI coding assistants, customer service automation, drug discovery, and dozens of other applications are real and growing. But there’s a long and well-documented history of transformative technologies generating enormous value for society while destroying value for most of the companies that try to commercialize them. The railroad boom of the 19th century transformed America and bankrupted most railroad companies. The internet reshaped the global economy and wiped out the vast majority of dot-com startups. AI may follow a similar path — transformative in impact, brutal in its selection of winners.
The burst that’s already happened should be taken seriously, not because it signals the end of the AI trade, but because it reveals its true structure. The market isn’t wrong about AI. It’s wrong about which companies will actually benefit. The first correction has been a useful sorting mechanism, separating substance from story. More sorting is likely ahead.
For institutional investors and portfolio managers, the lesson is one the market teaches over and over again: narrative is not a business model. The companies that survived the dot-com bust were the ones with real customers, real revenue, and real competitive moats. The same filter applies now. AI will create trillions in economic value over the coming decades. But the distribution of that value will be far more concentrated than the current stock market listings suggest. The bubble at the edges has already burst. Whether the center holds depends on whether the enormous capital being deployed today translates into the enormous returns being projected tomorrow.
That translation is the single most important question in markets right now. And the answer isn’t in yet.
WebProNews is an iEntry Publication