SAN FRANCISCO, July 20 — For two years, semiconductor stocks were the market’s engine room. This month, they have become its biggest source of anxiety. Since peaking in late June, the Philadelphia Semiconductor Index has fallen more than 20% from its 52-week high, erasing over a trillion dollars in market value across the sector and dragging the Nasdaq lower in three of the past five sessions. The selloff deepened again late last week, with chip shares leading the market down on Thursday.
What makes this rout unusual is what did not cause it. There has been no collapse in orders, no cancelled data centers, no profit warnings from the industry’s giants. Instead, a series of individually modest announcements combined to attack the assumption underneath the entire AI trade: that demand for computing power would outrun supply indefinitely.
Three announcements that changed the mood
The first crack appeared in early June, when Broadcom guided third-quarter AI chip sales to roughly $16 billion — comfortably a record, but short of the $17.2 billion analysts had penciled in — and declined to raise its full-year AI revenue forecast. The stock fell 14% in a session. In a market priced for perpetual upside surprises, merely meeting expectations had become a disappointment.
The second blow came on July 1, when reports emerged that Meta plans to launch a cloud business — dubbed Meta Compute — that would sell surplus AI training and inference capacity to enterprise customers. The implication rattled investors: if one of the world’s largest buyers of AI hardware has spare capacity to rent out, the industry may have built ahead of real demand. Memory maker Micron dropped more than 10% on the news and AMD fell nearly 7%, as traders concluded that a new source of computing supply had just walked onto the market.
The third signal came from South Korea, where SK Hynix — a critical supplier of the high-bandwidth memory that AI accelerators depend on — said it would delay part of its next-generation HBM4 expansion in favor of higher-margin conventional memory production. Executives framed it as portfolio discipline. Investors read it differently: a company with the clearest view of AI memory demand had chosen not to bet everything on its growth.
An echo with history
Markets have seen this movie before, and the memory is uncomfortable. In 1999 and 2000, telecommunications companies laid vastly more fiber-optic cable than the internet of that era needed, on the theory that demand would always catch up. It eventually did — but not before the overbuild bankrupted operators and vaporized shareholder capital. The question now dividing Wall Street is whether AI infrastructure is following the same arc, or whether the comparison is lazy: unlike dark fiber, today’s data centers are running hot, and the companies doing the building are among the most profitable enterprises in history.
The bearish case adds two aggravating factors that 2000 did not have: a Federal Reserve that may raise interest rates as soon as next week, which compresses the valuations of long-duration growth stocks, and chip valuations that had reached levels leaving no room for doubt. The bullish case replies that the selloff has been about positioning, not fundamentals — an unwinding of crowded trades after an extraordinary run, with institutional money rotating into financials, industrials and energy rather than leaving equities.
Intel: the hardest hit
No large chipmaker has suffered more in the rout than Intel, whose shares have fallen by roughly a fifth this month. The company sits at the intersection of every current worry: heavy capital spending, a turnaround still in progress, and exposure to any slowdown in data-center construction. Its results on Thursday will be examined less for the quarter just ended than for what management says about bookings in the second half.
The verdict arrives this week
The timing of the panic gives it a short fuse. Alphabet reports second-quarter results on Wednesday evening, with Tesla alongside and Intel to follow. Alphabet’s earnings call in particular has become a referendum on the AI buildout: its capital-expenditure guidance and cloud growth numbers are the closest thing to ground truth about whether the demand for computing is real and paying. Strong numbers and steady spending plans would undercut the overcapacity narrative in a single evening. Cautious ones would confirm it.
Either way, the sector’s period of effortless ascent is over. The market is no longer asking whether AI will change the economy — it is asking when the money comes back, and to whom. That is a harder question, and for the first time in two years, chip investors are being forced to answer it with evidence rather than faith. The stakes for the broader market are laid out in our week-ahead briefing.