SK Hynix is set to become the 2nd Korean company to hit $1 trln market cap

KB Securities raised SK Hynix’s target price to 3 million won, implying roughly a $1 trillion market cap and making it potentially the second South Korean company to reach that level. The broker lifted operating profit estimates on sharply higher DRAM/NAND price forecasts, projecting 2026 ASP gains of 194% and 244% as AI data-centre demand surges and AI operators account for 70% of memory shipments. It also said SK Hynix is entering a de facto zero-supply era, with long-term contracts through 2028-2030 supporting a re-rating.

Analysis

The second-order winner here is not just the memory suppliers; it is the entire AI capex stack that can convert rising token demand into persistent component scarcity. If memory ASPs keep compounding while new wafer starts stay constrained, GPU vendors like NVDA should see little direct unit risk but a more important indirect benefit: customers will be forced to prioritize highest-ROI inference and training projects, which supports premium accelerator utilization and keeps hyperscaler spending sticky.

The market may still be underestimating how much of this cycle is turning memory into a quasi-utility input rather than a cyclical semiconductor line item. That matters because once supply is locked via long-duration contracts, earnings visibility improves and the multiple can re-rate before the peak earnings inflects, especially for the name with the cleanest balance sheet and longest contract book. The key laggard risk is downstream cloud economics: if memory inflation outpaces model monetization, smaller AI platforms will delay deployments, concentrating spend further into the top few operators.

The near-term risk is not demand exhaustion but policy and capacity timing. A six- to twelve-month horizon still looks favorable, but the setup becomes vulnerable if a surprise fab restart, export restriction change, or macro slowdown resets pricing expectations before 2027. The contrarian view is that consensus may be extrapolating a straight-line scarcity regime; history says extreme margin phases attract capital, and the market can start discounting normalization well before actual supply arrives.

Key Decisions for Investors

• Long NVDA vs short SMH basket on a 3-6 month horizon: the trade isolates AI compute demand from any multiple compression in the broader semiconductor complex while benefiting if memory scarcity extends accelerator deployment budgets.
• Add to high-quality memory exposure on pullbacks, not breakouts: use 5-10% dips in SK Hynix/Samsung-linked proxies to build positions for a 12-18 month hold, with the thesis that earnings revisions and contract visibility can keep re-rating momentum intact.
• Buy call spreads on NVDA or broad AI capex proxies for 6-9 months out: use strikes ~10-15% above spot to express upside from sustained hyperscaler spending while limiting premium if the market starts discounting peak-cycle risk.
• Pair long NVDA / short lower-quality AI infrastructure names that rely on external funding: if memory inflation tightens AI economics, capital should continue concentrating in the few platforms that can self-fund capex.
• Set a risk trigger on any signal of new wafer capacity or contract price rollovers in 2026 pricing: if the market starts pricing 2027 normalization early, reduce AI semiconductor beta by one-third immediately.