Some Japanese shops start rationing GPUs — graphics cards with 16GB VRAM and up are becoming harder to find, says one store

Retailers in Japan have begun rationing high-VRAM graphics cards as global DRAM/GDDR shortages driven by surging AI data-center demand tighten supply chains; Tsukumo eX. is limiting purchases to one GeForce RTX 5060 Ti 16GB-or-higher or Radeon RX 9000-or-higher per buyer and other shops report difficulty restocking 8GB+ VRAM cards. Nvidia is reportedly withholding VRAM supplies to board partners and memory prices have surged—RAM module prices rose roughly 246% in 2025—while memory makers are deferring capacity expansion, making meaningful supply relief unlikely before 2026 and elevating risks to PC builders, consumer hardware sales, and GPU-dependent supply chains.

Analysis

Market structure: Memory suppliers (Micron MU, SK Hynix) and HBM/GDDR wafer owners are the clear near-term winners as DRAM/GDDR pricing surged (DRAM ASPs +246% YTD per reports) and buyers are rationed (Japanese shops restricting 16GB+ GPUs now). Consumer GPU OEMs and retail/DIY channels are losers: expect margin squeeze for consumer-focused SKUs and inventory-driven revenue misses for gaming segments in NVDA/AMD over the next 1–3 quarters.

Competitive dynamics & supply/demand: Wafer allocation has shifted to AI/AI-accelerator demand, crowding out GDDR supply for consumer cards; expect GDDR/GDDR6X tightness through 2026 with new fab capacity only online in 2026–2028. This increases pricing power for memory firms (higher gross margins) and gives large datacenter GPU buyers preferential fills, pressuring smaller board partners and second-tier competitors.

Cross-asset & risk assessment: Tight memory tightens tech capex cycles, pushing semiconductor equities higher while increasing input-cost-driven volatility in consumer discretionary and retail. Tail risks include a demand-side AI bust (90% price reversal scenario), geopolitically driven export curbs, or a fab outage — any could flip winners/losers in 1–4 months. Watch quarterly ASPs, wafer start cadence, and NVDA/AMD gaming vs datacenter revenue splits as near-term catalysts.

Trade/contrarian lens: The market may underprice sustained memory oligopoly rents but over-penalize NVDA/AMD because datacenter demand remains robust — a calibrated approach is warranted. If NVDA gaming rev misses >10% QoQ or MU guidance raises capex >20% year, re‑rate positions aggressively; otherwise use spread structures to express views without net directional gamma risk.

Key Decisions for Investors

• Establish a 2–3% long position in MU (Micron) over 6–12 months: thesis is sustained DRAM/GDDR pricing and margin expansion; target +20–35% upside, place a protective stop at -12% and scale out if MU raises 2026 guidance by >15%.
• Implement a hedged options pair: buy 6–9 month MU 20–30% OTM call calendar (to capture persistent price appreciation) and buy a 3–6 month NVDA 3–5% OTM put spread (sell lower strike) sized to 0.5–1% portfolio to hedge consumer GPU downside risk through next two earnings seasons.
• Short retail/DIY exposure (0.5–1% notional) and reduce weighting in consumer-facing hardware suppliers (e.g., specialty PC retailers or small-board makers) through Q1–Q4 2026; re‑allocate proceeds to semicap/memory names or SMH exposure.
• Pair-trade opportunity: long MU (or SMH) vs short AMD (0.5–1% notional) if AMD reports gaming revenue contraction >8% QoQ at next earnings — this captures memory-driven margin upside vs consumer GPU share pressure.
• Trigger-based action: if monthly DRAM/GDDR ASPs decline >20% sequentially or memory makers announce >25% immediate capacity expansion, close longs in memory names and flip to selective long NVDA/AMD exposure (reassess within 30 days).