For two years, artificial intelligence was the unstoppable force in financial markets. Every earnings call mentioned it. Every fund manager wanted exposure. Nvidia became shorthand for the future itself. And then the air started leaking out.
Not all at once. Not dramatically. But steadily, in a way that’s forcing investors to confront a question they’d been happy to ignore: What happens when the AI story collides with the AI math?
The numbers tell a sobering story. As Yahoo Finance reported, the so-called Magnificent Seven stocks — Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla — have collectively shed trillions in market capitalization from their peaks. Nvidia alone has lost roughly 30% from its highs. The Bloomberg Magnificent 7 Total Return Index dropped over 16% through the early months of 2025, dramatically underperforming the broader S&P 500. For the first time since the AI rally began in earnest, the biggest beneficiaries of the trade are becoming its biggest losers.
The catalyst wasn’t a single event. It was a cascade.
Start with DeepSeek. The Chinese AI startup rattled markets in January when it demonstrated that high-performance AI models could be built for a fraction of what American companies were spending. The implications were immediate and uncomfortable: if AI doesn’t require hundreds of billions in capital expenditure, then the entire investment thesis underpinning the semiconductor and cloud infrastructure boom needs rethinking. Nvidia’s stock plunged nearly 17% in a single session on January 27 — the largest single-day market-cap destruction for any company in U.S. history at the time.
Then came the tariffs. President Trump’s escalating trade war with China introduced fresh uncertainty into global supply chains, with particular exposure for technology companies that depend on cross-border semiconductor manufacturing and sales. The combination of trade policy volatility and questions about AI spending created a toxic cocktail for mega-cap tech.
But here’s where it gets more interesting — and more nuanced — than a simple “AI is over” narrative.
The hype may be fading. The underlying buildout is not. Capital expenditure plans from the largest hyperscalers tell a story of acceleration, not retreat. Microsoft has committed to spending $80 billion on AI-capable data centers in fiscal year 2025. Meta has raised its capex guidance to between $60 billion and $65 billion. Alphabet is spending $75 billion. Amazon’s figure is roughly $100 billion. These aren’t speculative promises. They’re budget line items, already flowing through to contracts with construction firms, chip suppliers, and power utilities.
The disconnect between stock prices and actual spending has created what some analysts see as a generational opportunity — and what others view as a warning sign of corporate overreach.
Gil Luria, a senior analyst at D.A. Davidson, captured the tension in comments reported by Yahoo Finance: the market is struggling to price in both the massive near-term costs and the uncertain long-term revenue streams associated with AI infrastructure. Companies are spending as if AI adoption will be universal and rapid. Investors, after two years of running on faith, are starting to demand proof.
That proof has been slow in arriving. Not absent — slow.
Microsoft reported AI-related revenue growth exceeding expectations in its most recent quarter, with Azure AI services growing at triple-digit rates. But even Microsoft acknowledged that demand was outstripping its ability to deploy capacity, which means revenue recognition gets pushed further into the future. Meta’s AI-driven advertising improvements have been meaningful but incremental, not the sort of step-function revenue growth that justifies a $60 billion annual spend. Alphabet’s Google Cloud division is growing, but the company has been vague about how much of that growth is directly attributable to generative AI versus traditional cloud migration.
The skeptics aren’t wrong to ask hard questions. The history of technology investment cycles is littered with examples of massive infrastructure buildouts that took far longer to monetize than projected. The fiber optic boom of the late 1990s created enormous capacity that wasn’t fully utilized for over a decade. The parallel isn’t perfect, but it’s instructive.
And yet the bears may be underestimating something fundamental about this cycle: the speed of adoption at the enterprise level. According to recent McKinsey survey data, nearly 65% of companies report regularly using generative AI in at least one business function, up from 33% just ten months earlier. That’s not a technology looking for a problem to solve. That’s a technology being integrated into operations in real time, even as investors grow impatient with the financial returns.
The stock market’s AI correction has been uneven in revealing ways. Nvidia, the most obvious proxy for AI spending, has taken the hardest hit — partly because its valuation embedded the most optimistic assumptions, and partly because DeepSeek raised genuine questions about whether future AI models will require as much raw compute. Broadcom and other semiconductor names have also pulled back significantly. But companies selling AI applications and services, as opposed to infrastructure, have held up somewhat better.
This rotation matters. It suggests the market isn’t abandoning AI as a thesis; it’s repricing where the value accrues. The early phase of any technology cycle rewards the picks-and-shovels players — the Nvidias, the data center REITs, the power companies. The second phase rewards the companies that figure out how to turn the infrastructure into products people pay for. We may be witnessing that transition in real time.
Consider Palantir Technologies, which has seen its stock roughly triple over the past year as it positioned itself as a bridge between raw AI capability and enterprise decision-making. Or ServiceNow, which has embedded AI into its workflow automation platform and reported accelerating subscription revenue. These aren’t the flashiest names in AI. They’re the ones closest to demonstrating tangible return on investment for customers.
The valuation reset has been painful but arguably necessary. At their peaks, the Magnificent Seven traded at an aggregate forward price-to-earnings multiple that assumed years of flawless execution and AI monetization. That’s not a bet most institutional investors are comfortable making in an environment of rising trade uncertainty, persistent inflation concerns, and a Federal Reserve that remains cautious about rate cuts.
Some of the froth was absurd. Companies with minimal AI exposure were adding “AI” to their earnings commentary and watching their stocks pop. That reflexive buying has stopped. Good.
What hasn’t stopped is the talent war. OpenAI, Google DeepMind, Anthropic, and a growing roster of well-funded startups are competing ferociously for a relatively small pool of top-tier AI researchers and engineers. Compensation packages for senior machine learning engineers at frontier AI labs now routinely exceed $1 million annually. This isn’t the behavior of an industry that thinks the boom is over. It’s the behavior of an industry racing to establish durable competitive advantages before the window closes.
The geopolitical dimension adds another layer of complexity. U.S. export controls on advanced semiconductors to China have created a bifurcated global AI market, with American and Chinese companies developing increasingly divergent technology stacks. DeepSeek’s breakthrough demonstrated that constraints can drive innovation — the company achieved its results partly because it was forced to optimize for less powerful hardware. That’s a lesson the U.S. technology establishment is still absorbing.
Meanwhile, the energy question looms larger by the quarter. AI data centers are extraordinarily power-hungry, and the planned buildout threatens to strain electrical grids in key markets. Goldman Sachs has estimated that U.S. data center power consumption could more than double by 2030. Utilities like Constellation Energy and Vistra have become unlikely AI beneficiaries, with their stocks surging as investors price in long-term power purchase agreements with hyperscalers. If anything, the energy bottleneck could prove a more binding constraint on AI growth than semiconductor supply.
So where does this leave investors?
The honest answer: in a messier, more differentiated market than the one that existed six months ago. The days of buying any AI-adjacent stock and watching it go up are over. What replaces that indiscriminate rally is a market that rewards companies with clear paths to AI revenue — and punishes those still operating on promises.
Nvidia remains the bellwether, for better or worse. Its next earnings report will be parsed with unusual intensity for any signs that hyperscaler spending is plateauing or that competition from AMD, Intel, and custom silicon from Google and Amazon is eroding its pricing power. The company’s data center revenue grew 93% year-over-year in its most recent quarter, a staggering figure — but one that came in below the most optimistic Wall Street estimates, a subtle but meaningful shift in market expectations.
The broader lesson is one the market relearns every cycle: transformative technologies generate transformative returns, but not on a straight line. The internet fundamentally changed the global economy. It also produced a stock market bubble that wiped out trillions before the real winners emerged. AI will likely follow a similar arc — immense long-term value creation punctuated by periods of excess, correction, and recalibration.
We’re in the recalibration phase now. It won’t be the last one.
For institutional allocators, the strategy increasingly favored by the sharpest money is selectivity: overweighting companies with demonstrable AI revenue, underweighting pure infrastructure plays that depend on perpetual spending growth, and maintaining exposure to the energy and industrial companies building the physical backbone of AI. It’s less exciting than the momentum trade that defined 2023 and 2024. It’s also more likely to produce durable returns.
The AI story isn’t over. But the easy chapter — the one where enthusiasm alone drove prices higher — clearly is. What comes next will be determined not by hype cycles or conference demos, but by the unglamorous work of turning a powerful technology into profitable products at scale. That’s harder. It takes longer. And it’s where the real money gets made.
WebProNews is an iEntry Publication