AI spending squeezes software; sector differentiation intensifies

SK Hynix ADR surged 27% in a single day, while IBM was down more than 20% in pre-market trading, setting its biggest-ever intraday drop. In the same day and the same AI narrative, two extremes with completely opposite directions—this kind of differentiation is inevitable after AI investing enters a mature phase.
The logic of early AI trading was “shotgun”: buy everything related to AI, because you don’t know who the winner will be, so you bet on the whole track. That logic worked in 2023 and 2024—almost the entire AI sector benefited, and industry allocation mattered more than stock picking.
But today’s IBM plunge declares that this era is over.
Goldman Sachs’ report lays out the underlying logic clearly: enterprise IT budgets are a fixed pool, while AI compute hardware is fighting for market share, squeezing traditional software budgets out. IBM’s results aren’t because IBM did something wrong—it’s because customers shifted the portion of budgets that would have gone to IBM into buying GPUs and compute infrastructure. This is a zero-sum game in capital expenditures: someone wins and someone loses—Hynix is the winner, while traditional enterprise software companies are the losers.
The detail that the Hynix ADR premium over Korean stocks has expanded to more than 50% is even more interesting. Two trading days ago, Korean stocks just recorded the biggest-ever single-day decline, yet today the ADR is up 27%, and the premium has widened to 50%. This isn’t the same pool of funds making decisions—it’s the pricing of the same company rapidly diverging between the U.S. market and the Korea market. U.S. capital is chasing the AI narrative, while Korean capital is digesting de-leveraging pressure; for now, the pricing logic in the two markets has fallen out of sync. This kind of premium is not sustainable; it will ultimately converge. Whether Korean stocks catch up or the ADR gives back is an arbitrage signal worth watching next.
A U.S. Department of Commerce official hinted that chip and AI regulatory measures are about to be issued. The timing is very delicate. Hynix just finished rallying 27%, and then a regulatory signal comes out—does it cool down an overheated market in advance, or is it just routine statements? We’ll need to see the specifics, but the market will price it as risk first.
The story of total AI growth is still intact, but mean reversion has ended. From now on, buying the word “AI” is no longer enough; you need to determine who is a net beneficiary of AI spending and who is a net loser. The era of sector allocation is over, and the era of stock-level screening begins.
DYOR, not investment advice