Memory chip stocks have sold off hard over the past month despite the sector delivering some of the strongest quarterly results in its history, and the split in how investors are explaining that gap has become the central argument in semiconductors right now. One camp treats the memory market as a commodity business that has never once escaped its own history: shortages inflate prices, inflated prices fund new capacity, new capacity floods the market, and margins collapse. The other camp argues AI has changed what memory actually is to the industry that buys it, and that a demand base this large and this committed cannot unwind the way DRAM gluts have in the past. Both camps are looking at the same capacity expansion numbers and drawing opposite conclusions.
The oversupply case has the weight of precedent behind it. DRAM and NAND behave like commodities, not software: prices rise when supply is tight and fall once manufacturers expand production, and every memory cycle of the last two decades has gone from shortage to glut faster than the fabs built to chase it. Micron is trading near 849, down from the 1,255 all-time high it hit on June 25 and formally in a bear market as of July 8, even with data-center and enterprise SSD revenue still climbing. SK hynix’s Nasdaq-listed ADRs have drifted back to roughly 154 from a 168 debut close on July 10, inside a 52-week range of 145.57 to 194.80. Skeptics point to a widening set of building blocks for the glut case: hyperscalers are now directing more incremental AI capital toward areas beyond the HBM bottleneck that defined the first phase of the buildout, China’s CXMT is moving toward its own IPO as a new supply source, and every major producer is simultaneously ramping capacity and improving yields into the same demand window. As one fund manager put it to the financial press, the industry has historically average returns on capital that is currently priced to make very high returns in the future, and a leopard does not often change its spots.
The supercycle case rests on the scale of the demand numbers rather than on faith that this time is different by assertion. SK Group Chairman Chey Tae-won said this week that customers have asked SK hynix for 60 to 100 percent more AI memory in 2027 than this year, and with AI now accounting for more than half of overall semiconductor consumption, he put total demand growth at a minimum of 50 to 60 percent. Samsung has said some customers have already secured supply allocations through 2027. SK hynix’s own executives have said demand will likely stay above the company’s supply capability even beyond 2030, and Chey has put greenfield fab lead times at more than five years, meaning fresh capacity from the current wave of expansion lands near the tail end of the shortage window rather than closing it early. Under that view, the capex numbers that oversupply bears cite as the seeds of the next glut are simply catching up to a demand curve that was undersized from the start.
What makes this debate unusually hard to resolve cleanly is that Chey himself is arguing both sides at once. The same briefing in which he laid out demand running 50 to 100 percent above current supply is the one in which he called today’s prices abnormal and warned that sustained highs would invite new competitors and geopolitical retaliation. That is not the language of an executive confident the structural-shortage argument fully insulates the business from its own history. It reads more like an acknowledgment that the bear case has a real mechanism behind it, even from the person best positioned to talk his own book.
The cleanest way to track which narrative is winning is not the stock price, which is reacting to sentiment and positioning as much as fundamentals, but the gap between spot pricing and the multi-year contracted volume that Micron and SK hynix have already booked. A supercycle that survives its own capacity buildout should show contracted pricing holding even as spot softens. A conventional cycle rolling over on schedule should show both sliding together. That data point, not the next earnings beat or the next capex headline, is the one that will actually settle the argument.
