The AI chip selloff has dragged the world's most important semiconductor index into a bear market, unwinding a rally that once looked unstoppable. Wall Street built two years of gains on the belief that computing power would stay scarce forever. That belief is now cracking in real time.
By Blumefield | July 18, 2026
A Rally Meets Its Reckoning
The AI chip selloff that began quietly at the start of July has now delivered the number every trader dreads. The Philadelphia Semiconductor Index, the benchmark that tracks the world's most important chipmakers, has fallen more than 20 percent from the record high it set on June 22, dropping it into an official bear market. Three weeks ago these same stocks looked untouchable. The index had climbed roughly 105 percent over the previous year, minting trillions in paper wealth and convincing much of Wall Street that the artificial intelligence buildout would run without pause.
That conviction is gone. The Nasdaq semiconductor benchmark peaked near 14,655 before the reversal took hold, and the descent since has been brutal in its speed. Intel has shed more than 30 percent from its own high. Marvell and Arm have each dropped by a similar margin. Micron, the memory maker most exposed to the pricing cycle, has fallen roughly 17 percent. Even the names investors treated as bulletproof are no longer immune.
What makes this AI chip selloff so unsettling is not a single piece of bad news but the collapse of a story. For two years, chip stocks traded on the assumption that graphics processors would remain permanently scarce and that demand would always outrun supply. Every quarter seemed to confirm it. Now, in the space of a few weeks, that assumption has been challenged from three directions at once, and the market has decided it can no longer pay any price for exposure to the theme.
The Scarcity Trade Unravels
The first crack in the AI chip selloff appeared on July 1, when reports surfaced that Meta planned to launch a cloud unit selling its surplus computing capacity to outside customers. The detail sounds technical, but its meaning was immediate and profound. If a company that spent tens of billions building AI data centers now has enough spare capacity to rent it out, then the shortage that justified sky high valuations may be ending. Hyperscalers were supposed to be insatiable buyers. The moment one of them looked ready to become a seller, the entire supply and demand equation shifted.
Investors had priced these companies for a world without slack. Nvidia traded above 50 times earnings. Intel reached more than 12 times sales on hopes for its foundry turnaround. Memory makers watched their multiples balloon on the belief that every processor shipped would need expensive high bandwidth memory attached to it. When the scarcity premium starts to unwind, all of those multiples become vulnerable at the same time, and the selling feeds on itself.
The supply signals kept coming. SK Hynix said it would delay expansion of its most advanced memory in favor of higher margin products, a move investors read as quiet doubt about how fast AI demand would keep growing. Reports that Intel's next generation manufacturing process would not reach profitable yields until late 2026 or 2027 added a company specific wound to an already nervous sector. Each headline on its own might have been survivable. Together they formed the outline of a genuine AI chip selloff rather than a passing dip.
China Lights the Fuse Again
The sharpest blow landed on July 17, and it came from Hangzhou rather than Silicon Valley. Chinese startup Moonshot released Kimi K3, an open model that testers rated alongside the best systems from OpenAI and Anthropic while costing a fraction to run. Markets have seen this movie before. A cheaper, capable model from China raises the same uncomfortable question every time. If frontier performance no longer requires the most expensive hardware in enormous quantities, how much of the projected demand for that hardware is real?
The reaction was swift. Semiconductor shares that had been grinding lower gapped down, and the broader market followed. An open model that anyone can download and run locally undercuts the premise that intelligence must be rented from a handful of well capitalized labs, each of them buying chips by the warehouse. That premise has underwritten a large share of the sector's valuation, so any evidence against it moves prices hard.
None of this means the buildout stops. It means the market is no longer willing to assume the buildout is infinite. A cheaper Chinese model does not erase the need for computing power, but it does force investors to separate durable demand from the speculative kind that inflates during a mania. The current AI chip selloff is the sound of that separation happening in real time.
Record Earnings, Falling Shares
The cruelest part of this AI chip selloff is that the fundamentals still look strong. Taiwan Semiconductor Manufacturing, the foundry that builds the most advanced chips for Nvidia, Apple and nearly everyone else, reported record second quarter results on July 16. Revenue reached 40.2 billion dollars, up almost 34 percent from a year earlier, while net profit surged 77 percent to a fresh record. High performance computing, the category that includes AI accelerators, made up 66 percent of sales. The company even pledged another 100 billion dollars of investment in the United States.
The stock barely reacted, and the AI chip selloff rolled on regardless. That disconnect is the whole story. Samsung earlier reported a rise in operating profit of more than 1,800 percent and watched its shares slide anyway. When results this good cannot lift prices, it tells you the market has already priced in perfection and is now hunting for reasons to worry. Good news is being sold, not bought, which is the classic signature of a mood that has turned.
The gloom was not confined to chips. Netflix added to the unease when it posted second quarter numbers that met expectations on earnings but guided softly on future revenue, sending the stock down more than 10 percent and to a 52 week low. Rising oil prices on renewed Middle East tension completed an ugly backdrop. For a market that had leaned so heavily on a handful of technology winners, the timing could hardly have been worse.
What Breaks the Fall
The path out runs through earnings and policy. Intel reports around July 23, and its guidance on manufacturing yields will either steady the story or confirm the bears. More important are the capital spending plans of the largest cloud buyers. The big four are still guiding toward a combined figure well above last year, up roughly 77 percent, which suggests the underlying demand has moderated rather than vanished. Any cut to those plans would turn the current AI chip selloff into something deeper.
Monetary policy is not helping. Under new chair Kevin Warsh, the Federal Reserve has signaled that its next move could be a rate increase rather than a cut, with inflation still running above 4 percent. Higher rates punish exactly the kind of high multiple growth stocks that dominate the semiconductor index, raising the cost of the capital these firms need to keep building. That combination of a hawkish central bank and a nervous market leaves little cushion.
For now, the lesson is humility. The AI revolution is real, and the demand for computing power will keep growing for years. What has broken is the belief that it would grow in a straight line, and that any company touching the theme deserved to trade at any price. Investors who want to track where the sector goes next can follow the coverage at Blumefield, where the gap between narrative and numbers has been the story all along. Markets always overshoot in both directions. The hard part, as Blumefield readers know, is telling a correction from a turning point while you are still inside it.