SK Group chairman says wafer shortage to last until 2030, trying to stabilise memory prices

SK Group Chair Chey Tae-won warned the global chip wafer shortage could persist until 2030, forecasting more than a 20% shortfall. SK Hynix, the main HBM supplier with a 57% HBM share and a 32% global DRAM share, is reviewing a potential U.S. ADR listing and expects its CEO to unveil a plan to stabilise DRAM prices. Shares rose 3.5% in Seoul (KOSPI +3%) as the group also said it is seeking alternative energy sources amid higher energy costs from Middle East tensions.

Analysis

A sustained mismatch between AI-driven HBM demand and wafer capacity reorders pricing power across the stack: memory fabs and HBM-focused DRAM lines gain structural gross-margin tailwinds while commodity DRAM and legacy foundry customers face rationing and higher unit economics. That rationing amplifies pricing optionality for GPU OEMs and server integrators — vendors who can secure long-term wafer/packaging allocations (via prepayments, JV capacity, or inventory financing) will convert scarcity into lasting margin expansion over the next 12–36 months.

The primary reversal risks are technological and supply-side: a step-change in model efficiency or a rapid rollout of alternative memory/packaging (chiplet-HBM substitutes or on-package HBM capacity increases) could collapse HBM intensity per model and remove the scarcity premium. Policy/export shocks that further fragment supply would tighten the market faster but also raise execution risk for cross-border capex, making 2026–2028 a highly uncertain transition window as new fabs come online and capacity utilization rebalances.

Consensus is underweight the buyer-side structural responses: hyperscalers and large OEMs will accelerate vertical securing of wafers (long-term contracts, captive fabs, or exclusive allocations), meaning standalone suppliers without contractual sales may see order visibility fall while integrated partners capture outsized returns. That creates an asymmetric opportunity: companies with visible contracted HBM supply and scale (server OEMs, select memory names, and equipment vendors) can outperform even if headline GPU sell-through lags; beware crowded long positions in marginal memory names where a capex wave would flip pricing sharply negative.

Key Decisions for Investors

• Long SMCI (Super Micro Computer) — Buy shares or call options with 9–12 month tenor to capture server ASP and backlog premium as customers pay up for allocated HBM/GPU capacity; target asymmetry: 30–50% upside vs 20–30% downside if GPU supply normalizes (use covered-call or protective put to cap tail risk).
• Defined‑risk NVDA bullish spread — Buy a 12‑18 month call spread (long-dated near-OTM / short higher strike) to express continued GPU pricing power without full vega exposure; expect 2:1 gross reward:risk if scarcity sustains, and limited loss if capacity ramps faster than expected.
• Long SK Hynix (000660.KS or ADR if available) — Accumulate over 12–24 months on conviction that HBM monetization and contractual allocations improve memory ASPs; risk: 25–40% drawdown if new capacity materially oversupplies DRAM markets following aggressive capex by competitors.
• Hedge: buy 6–12 month puts on software/monetization-sensitive names (example: APP) sized to 10–15% of gross AI exposure — protects against demand-side drag if limited GPU supply slows SaaS/ads monetization cadence over the next 6–12 months.