The SOXX semiconductor ETF fell 10% in a single day on June 5, and the sector’s biggest names fell harder. AMD dropped 10.9%, Intel 11.3%, and the memory and equipment companies that had led the market’s AI rally followed them down. A week later, the sector is no longer moving together.
The divergence is sharp. Intel has rebounded strongly, helped by a large foundry order from Alphabet’s Google and by reports that Nvidia is evaluating Intel’s manufacturing for some products. Micron has led on the strength of AI memory demand. Tokyo Electron, the Japanese equipment maker, jumped 10% in a day. Meanwhile, some of the stocks that fell hardest are still below their pre-selloff levels.
The June 5 selloff began with Broadcom’s earnings, which disappointed a market that had priced in perfection for AI suppliers. The initial reaction was a blanket de-rating: every stock with an AI narrative sold off regardless of its own outlook. The recovery has been more discriminating, with investors separating the companies whose businesses are demonstrably strong from those whose valuations had run ahead of their results.
The memory trade has been the clearest winner of the rebound. Micron, Samsung and SK Hynix are selling high-bandwidth memory as fast as they can make it, and the pricing power that the June selloff briefly threatened has reasserted itself. Analysts expect Micron to report record free cash flow this year, and its June 24 earnings will be a test of whether the memory upcycle is as strong as the market now believes.
Intel’s rebound is the most surprising leg of the divergence. The company’s foundry business has been a years-long project with more setbacks than breakthroughs, and its stock has lagged the sector for most of the AI boom. The Alphabet order, reported in recent days, gives Intel’s foundry a marquee customer to pair with Microsoft, and the Nvidia evaluation reports have raised the possibility that Intel could win a second source role in the AI supply chain. Both stories are early, analysts caution, but they have changed the market’s view of Intel’s downside.
The equipment makers are benefiting from a different logic. Tokyo Electron’s 10% jump reflects expectations that the memory boom will drive a new round of fab construction, and that equipment orders will follow chip prices with a lag. The equipment trade is a bet on the cycle’s longevity: if memory makers believe the boom will last, they will build fabs, and the equipment suppliers will collect the bill.
The split also reflects different exposures to the same headline. Broadcom’s miss was concentrated in its custom-chip business, where a customer delayed orders, and the stocks that fell hardest were the ones most exposed to the same customers. Memory demand, by contrast, showed no weakness in the weeks after the report: the suppliers kept shipping, and their customers kept ordering. The market has effectively concluded that the Broadcom episode was a company-specific scheduling issue rather than the start of an AI slowdown. That conclusion is the foundation of the rebound, and every earnings report in the sector over the next month will test it.
The divergence has a complicating factor: inflation. May’s producer price index rose 6.5% from a year earlier, the fastest pace in nearly four years, and the reading has fed fears that the Federal Reserve will keep rates higher for longer. For technology valuations, which are sensitive to discount rates, that is a direct threat. The rebound in chip stocks has happened alongside the inflation scare, and analysts said the two forces are now in tension.
If inflation keeps running hot, the memory and equipment stocks that have led the rebound could face a second correction, this one driven by rates rather than earnings. The companies with strong cash flows — the memory makers, the foundries — would be relatively protected, while the highest-multiple names would take the brunt. That is the shape of the divergence investors are positioning for.
The other variable is the AI buildout itself. Cloud providers have committed to record capital spending, and their orders underpin the demand for memory, accelerators and equipment. The Broadcom report raised questions about the timing of some AI deployments, but the subsequent weeks have produced no evidence that the buildout is slowing. As long as the orders hold, the fundamental floor under the sector holds.
For investors, the lesson of the week is that semiconductor stocks are no longer a single trade. The sector has split into businesses with different cycles — memory at the peak of a supercycle, foundries fighting for share, equipment waiting on the next wave of fab construction — and the market is pricing each one on its own merits. The SOXX ETF’s 10% day showed how far the sector can fall together; the divergence since shows that it can recover separately.
