The first real correction in the AI chip boom just erased a year's worth of hype in two trading days.
The Summary
- SK Hynix ADRs fell as much as 9.3% in US trading after the stock plunged a record 15% in Seoul, triggering a market-wide trading halt when the Kospi dropped 9%
- The selloff spread across memory and storage companies, with Micron, SanDisk, and Western Digital all down more than 6%
- ADRs now trade just above their $149 debut price from last week, wiping out gains that took months to build
- Investor caution followed an unprecedented rally in global chipmaker stocks fueled by AI infrastructure spending
The Signal
SK Hynix's 15% single-day crash in Seoul wasn't just a bad day for one stock. It was the sound of air escaping from the AI infrastructure trade. The company makes high-bandwidth memory for AI training chips. When its stock drops that hard, that fast, it's a referendum on whether the buildout everyone's been betting on is real or overpriced.
The contagion was immediate. Memory and storage peers dropped in lockstep, with Micron down over 6%, SanDisk and Western Digital following. This wasn't stock-picking. This was a sector repricing. Investors stopped asking "which memory maker wins?" and started asking "did we overpay for all of them?"
"The decline brought ADRs to trade just a few dollars above the $149 level where they priced last week."
What makes this correction sharp is the timing. SK Hynix ADRs just started trading in the US last week. They debuted at $149, rode the hype a bit higher, then gave it all back in 48 hours. New listing, instant reality check. That's not normal profit-taking. That's a market saying it doesn't believe the price it just paid.
The unprecedented rally in global chipmaker stocks this year was built on a simple thesis: AI training requires memory, lots of it, and specialized chips to use it. SK Hynix sells both. But the thesis assumed linear growth in AI spending. What we're seeing now is the market testing that assumption. If hyperscalers slow their data center builds, even slightly, the memory makers get hit first and hardest.
Key factors driving the selloff:
- Valuation stretched after months of straight-line gains
- No new demand catalysts to justify current multiples
- Rising concern that AI infrastructure buildout is ahead of actual AI revenue
The Kospi's 9% drop and market-wide trading suspension added fuel. Circuit breakers are designed to prevent panic. They also tell everyone watching that panic was already happening. When a national index falls hard enough to trigger a halt, global investors don't wait around to see what happens next. They sell.
The Implication
If you're building anything in the agent economy, this is your reminder that the infrastructure layer is a bet, not a given. The companies selling picks and shovels for AI got repriced because the market stopped trusting the gold rush narrative. That doesn't mean the rush is over. It means the easy money phase is.
Watch what hyperscalers report in their next earnings. If capex guidance stays flat or drops, memory stocks have further to fall. If spending accelerates, this was a buying opportunity. Either way, the correction separates the companies building real products from the ones riding coattails. For anyone tokenizing compute, building agent platforms, or deploying AI at scale, cheaper memory is good news. Lower stock prices don't change the tech. They just change who profits from it.