AI-Capex Doubts Trigger $1.3T Semiconductor Rout as Intel Slides

A one-week rout across the semis has stopped being a story about valuations and started being a story about whether the AI-capex trade itself is intact. Peter Cohan writes at Forbes that the Philadelphia Semiconductor Index is off 10.8%, the VanEck SMH ETF has fallen 13% over ten sessions, Intel has pushed down more than 20 percent, and Micron dropped 13% in a single session that erased roughly $138 billion in market cap. Reuters, in reporting cited alongside, estimates the aggregate sector wipeout at around $1.3 trillion.

The proximate triggers are more interesting than the tape. Samsung reported preliminary Q2 2026 operating profit of about $58 billion, a 19-fold jump from a year earlier, and its shares still fell as much as 10% intraday in Seoul on profit-taking after a 150% run this year. Days before that, CNBC covered on July 1 Meta standing up a cloud business to sell idle AI compute to outside customers, which the market read as a signal the biggest builders now have more capacity than they can consume in-house. Zuckerberg is reported to have told staff internally that Meta's AI agent development is behind schedule, and Anthropic has separately signaled plans to co-develop custom chips with Samsung. Each is a small tug at the merchant-GPU thesis.

Morgan Stanley is calling this a 'mid-cycle reset' rather than a top, and the reporting is careful to note that the root cause is not weakening end demand but an amalgam of dot-com-era valuations, a more hawkish Fed, and rising doubt about the return on AI infrastructure spend. Leveraged products like SOXL, which target three times the daily return of the PHLX Semiconductor Index, are amplifying the move as they mechanically sell into weakness.

The honest caveat is that most of these specifics are single-session or single-week prints, and the 'mid-cycle reset' framing is a house view that has yet to be tested by an actual earnings-season cut to hyperscaler capex guides. What the reporting does not give you is a clean read on enterprise AI budgets themselves, only the shape of the anxiety and Meta's own hint that it has compute to spare.

If Morgan Stanley is right, this is a moment to shop the SOX names on the pullback rather than run from them. If they are wrong, the more interesting question is which builder blinks first on 2026 capex, and whether the next chapter is Meta-style monetization of excess compute rather than another straight-line ramp on new orders.