The mood in tech boardrooms this week can be summarized in a single, uncomfortable number: somewhere north of $200 billion in combined AI infrastructure spending planned for 2027, and not a single executive willing to say when it will generate commensurate returns.
Microsoft, Alphabet, Meta, and now the latest hyperscaler earnings have all delivered the same dissonant message: revenue growth is fine, AI products are popular, and by the way, capital expenditure is accelerating at a pace that would make a sovereign wealth fund blush. Wall Street's response has been to sell first and ask questions later. The Nasdaq's tech-heavy contingent has shed hundreds of billions in market capitalization in days, and Tokyo's Nikkei 225 followed suit overnight, dragged down by chip suppliers and cloud infrastructure plays suddenly looking overextended.
The capex arms race has no referee
What began as a prudent response to the generative AI boom has metastasized into something closer to mutually assured construction. Each hyperscaler justifies its spending by pointing to the others: if we don't build the data centers, someone else will, and we'll lose the enterprise contracts that matter. The logic is sound in isolation and terrifying in aggregate. Collectively, the major cloud providers are on track to pour more into AI infrastructure over the next three years than the entire semiconductor industry generated in revenue a decade ago.
The problem is not that the spending is irrational—it may well prove visionary. The problem is that no one can articulate the unit economics. When pressed on earnings calls, executives retreat to gauzy language about "long-term positioning" and "generational platform shifts." These are not lies, but they are not answers either.
Investors are done waiting for the punchline
For two years, markets extended extraordinary patience to AI-related capex, treating it as a special category immune to normal return-on-investment scrutiny. That grace period appears to be ending. The sell-offs this week are not panics; they are repricing events. Analysts are beginning to model scenarios where AI spending continues at current rates but enterprise adoption plateaus, leaving hyperscalers with expensive, underutilized infrastructure and compressed margins.
The downstream effects are already visible. Chip suppliers who rode the AI wave to absurd valuations are suddenly being asked harder questions about order visibility. Data center REITs are wobbling. Even power utilities that signed long-term contracts with cloud providers are watching their counterparties' stock prices and wondering about credit risk.
Our take
The AI infrastructure buildout is probably necessary, possibly transformative, and almost certainly mispriced. Big Tech is making a collective wager that the applications justifying this spending will materialize before investor patience expires and before interest rates make the debt financing these projects untenable. History suggests that technology bets of this magnitude sometimes pay off spectacularly and sometimes produce expensive monuments to hubris. The honest answer is that nobody—not the CEOs, not the analysts, not the engineers—knows which outcome awaits. What we do know is that Wall Street has stopped pretending otherwise, and that repricing has only just begun.
