Micron lost more than a tenth of its market value in a single session. SoftBank and ASML slid alongside it. By the time the closing bell rang, the selloff had stopped looking like a bad day for a few semiconductor names and started looking like a verdict on the entire trade that has powered the market for the better part of three years.
That trade has been remarkably simple to explain and remarkably hard to bet against: AI needs compute, compute needs chips, chips need memory, and everyone from hyperscalers to sovereign wealth funds will pay whatever it takes. It worked. Semiconductors are still up roughly sixty percent this year by some measures, a number that would be the envy of any sector in any decade. The problem with a run like that is what it does to expectations going forward. When the bar has already been set at the ceiling, merely clearing it no longer counts as success.
The current rout seems less about any single earnings miss or guidance cut than about a creeping suspicion that the spending curve, as steep as it has been, cannot stay vertical forever. Memory pricing, which drove so much of Micron’s recent run, is cyclical by nature. Every memory executive knows this. Every memory investor knows this too, which is why the stocks tend to move first and ask questions later. ASML sits in a different part of the chain, but it is arguably the most exposed of all, because its lithography machines are the capital expenditure that customers commit to years in advance. When those customers get cautious, the caution shows up at ASML’s door with a lag, and the market knows how to price a lag.
SoftBank’s slide is its own story. The conglomerate has positioned itself as a financier and cheerleader of the AI buildout, with stakes that rise and fall on sentiment as much as fundamentals. When the mood turns, the most leveraged believers feel it hardest. That is not necessarily a sign that SoftBank’s thesis is wrong, but it is a reminder that conviction and valuation are two different things.
What makes this moment interesting is that the rout is happening against a backdrop of genuine product momentum rather than visible weakness. Meta just launched a new AI model and is testing whether users will pay for it. Epic Games has confirmed a significant pivot toward AI in how it builds and operates Fortnite. Anthropic is out warning that models could soon propagate themselves across home devices, which is either a safety disclosure or a marketing flourish depending on your disposition. Google’s quantum team keeps publishing error-correction results that inch the field toward something useful. None of this reads like a sector running out of ideas.
It reads instead like a sector running into math. The companies buying all these chips have to eventually show returns that justify the depreciation schedules. Data center buildouts are expensive in ways that show up on income statements for years, and the revenue they generate is still, in many cases, theoretical or early. There is a real possibility that the demand is durable and the valuations still got ahead of themselves. Those two things can be true simultaneously, and markets are not always patient enough to distinguish between a correction and a collapse.
The Baltimore venture data offers an accidental mirror here. Deal counts in that market held steady, but capital concentrated at the top. That is the same dynamic playing out in AI hardware globally: money flowing toward a handful of obvious winners while everyone else waits to see whether the tide reaches them. Concentration feels efficient in a boom and fragile in a pullback.
None of this means the AI supercycle is over. Texas Instruments, notably absent from the worst of the selling, sits at the analog and embedded end of the market, where demand is diversified across industrial, automotive and consumer applications rather than dependent on a single narrative. That kind of quiet exposure has a way of looking smart exactly when the loud trades stumble.
For investors in Micron, ASML and the rest, the question is not whether AI will keep growing. It will. The question is whether the current prices already assume a future that even a healthy boom cannot deliver on schedule. Wall Street spent this week answering that question the only way it knows how, by selling first and analyzing later. Whether that proves to be wisdom or panic will depend on what the next round of earnings actually shows about orders, backlog and the willingness of customers to keep signing checks. Until then, the chip trade is not broken so much as newly uncertain, which may be the most honest thing that has happened to it in a long time.