The Quiet AI Trade That’s Raking in Billions

SK Group chairman Chey Tae Won said AI demand should remain strong over the next few years, reinforcing a constructive outlook for AI-related hardware demand such as HBM4 memory chips. The comments downplay concerns about an AI spending bubble and support sentiment for semiconductor and infrastructure suppliers. The article contains no hard financial figures, so the likely market impact is limited but mildly positive for the sector.

Analysis

This is less a sentiment headline than a read-through on the durability of the AI capex cycle: when the supply chain is publicly signaling confidence at the component level, it usually means order books are visible far enough out that buyers are still prioritizing capacity over near-term ROI scrutiny. The second-order effect is that memory remains one of the cleanest ways to express AI upside because it monetizes every incremental accelerator deployment, even if hyperscaler software monetization lags. That makes the beneficiary set broader than the obvious chip designers; packaging, substrates, test equipment, and upstream materials should continue to benefit as vendors race to secure multi-quarter supply.

The more interesting competitive dynamic is that optimism itself can tighten allocation. If HBM4 ramps remain constrained, the winners are the vendors with the strongest qualification, yield, and customer lock-in, while late movers may see gross margins pressured by forced discounting or underutilization. The risk is not a demand collapse in the next few months, but a digestion phase in 2026 if hyperscalers front-load orders now and then spend a quarter or two working through installed capacity; that would hurt the less differentiated suppliers first.

Consensus is probably underweighting how much this environment favors the picks-and-shovels names versus the model-layer beneficiaries. The market keeps treating AI as a binary bubble debate, but the more important variable is supply chain elasticity: when a constrained input enjoys multi-year visibility, earnings revisions can remain positive even if sentiment cools. The contrarian view is that the strongest stocks may already be discounting the headline growth narrative, so the better risk/reward is in laggards that still have structural operating leverage to the same capex wave.

Key Decisions for Investors

• Long the AI memory complex on pullbacks: favor exposure to HBM-linked suppliers and equipment makers for a 3-6 month horizon; best risk/reward is on names with visible backlog and operating leverage, not the most crowded hyperscaler proxies.
• Pair trade: long memory/packaging beneficiaries versus short a basket of software-heavy AI winners over the next 1-3 months, betting that near-term earnings revisions stay strongest where revenue is tied to physical buildout rather than monetization promises.
• Use any 5-8% selloff in semiconductor equipment or advanced packaging names to add risk; the setup is asymmetric because customer qualification cycles and supply constraints make cancellations less likely than deferrals.
• Avoid chasing the most extended AI mega-cap names after strength; if hyperscaler capex commentary softens for even one quarter, multiple compression could outweigh the incremental optimism from the supply chain signal.