AI-related equities sold off on June 23, 2026, as investors revisited the foundational question that has hovered over the sector's two-year rally: whether the capital being committed to AI infrastructure will ever generate commensurate returns. The move extended a pattern that began earlier in the month, with tech and chip stocks declining again on June 23–24 as the rotation spread from U.S. markets into Asia, rattling semiconductor names across the region.

The numbers behind the anxiety are not abstract. Hyperscalers — the handful of cloud giants running the world's largest AI infrastructure — were on track to spend over $800 billion in capital expenditure in 2026, according to Reuters, a figure roughly equal to the combined capex of every non-technology company in the S&P 500. That comparison matters. It implies that the AI infrastructure buildout, funded largely by a narrow slice of the market, has become one of the largest coordinated capital allocation decisions in modern corporate history. The revenue base that must eventually justify it does not yet exist at scale.

The June 23 selloff was not the first crack. On June 8, tech stocks declined as Federal Reserve policy expectations rattled the AI rally, and Asian markets experienced turbulent trading in the week that followed, driven largely by the U.S. semiconductor selloff, per Reuters. The pattern across these episodes is consistent: sentiment turns on any signal — macro or monetary — that compresses the long-duration growth assumptions baked into AI valuations.

The Fed's Line of Sight

The Federal Reserve has been watching the AI buildout closely, and its posture is more cautious than celebratory. At his June 17 press conference, Fed Chair Warsh said the central bank would survey the pace, reach, and economic impact of general-purpose technologies including AI, according to the FOMC press conference transcript. That framing — general-purpose technology, pace, reach — is deliberately measured. It signals that the Fed is not prepared to treat AI-driven productivity as a fait accompli when calibrating monetary policy.

The Fed's own research found that just over 20% of U.S. firms expected to use AI in the first half of 2026, per a Federal Reserve note published in April. For context, that figure describes intent to use, not demonstrated productivity uplift. The gap between adoption and measurable output gains is precisely where the bull case for AI remains most vulnerable.

On the labor market side, Governor Cook noted in a May 27 speech that unemployment was running in line with estimates of the natural rate, suggesting balanced labor supply and demand. That assessment cuts both ways for AI: a tight-to-neutral labor market provides no urgent productivity imperative that would force rapid AI adoption, and it also limits the near-term deflationary tailwind that AI optimists frequently cite.

Governor Barr went further in February, stating plainly that AI may deeply disrupt labor markets and harm some workers in the short term, in a speech on February 17. That the Fed is already flagging distributional labor market risk — before AI adoption has even crossed the 25% threshold among firms — tells you something about how seriously policymakers are gaming out the downside scenarios.

What the Selloff Is Actually Pricing

The core tension is straightforward. Hyperscaler capex is a present-tense cash outflow of extraordinary magnitude. The productivity and revenue gains that justify it are, at best, a future-tense probability distribution. When rates are low and growth is scarce, markets assign that distribution a high present value. When rates stay elevated — or when policy signals suggest they will — the discount rate applied to those future cash flows rises, and valuations compress.

The selloff on June 23 reflected exactly that mechanism: investors recalibrating the probability that $800 billion in annual infrastructure spend will translate into earnings power on a timeline that justifies current multiples. Whether it does depends on variables — enterprise adoption curves, regulatory posture, competing model efficiency — that are genuinely unknown. What is known is the spend. It is committed. It is large. And the market is increasingly unwilling to treat its eventual payoff as a certainty.