NVIDIA to Use SK hynix and Samsung HBM4 for "Vera Rubin" Without Micron

NVIDIA's upcoming 'Vera Rubin' VR200 NVL72 rack systems (late-summer ship) will use HBM4 memory supplied roughly 70% by SK hynix and 30% by Samsung, with Micron reportedly receiving no HBM4 commitment; Micron will instead supply SOCAMM2 LPDDR5X for NVIDIA's Vera CPUs (up to 1.5 TB). NVIDIA has driven aggressive memory specification scaling — the VR200 NVL72 now operates at 22 TB/s (up from a 13 TB/s target), implying meaningful revenue and design-win consequences for HBM suppliers while shifting Micron toward CPU memory content. Investors should view this as positive for SK hynix/Samsung and for NVIDIA's product competitiveness, and mixed for Micron given the loss of HBM4 share but offset by LPDDR5X CPU opportunities.

Analysis

Market structure: NVIDIA’s VR200 win concentrates HBM4 demand into SK hynix (~70%) and Samsung (~30%), amplifying their near-term pricing power for HBM4 and likely lifting ASPs by a material margin (we estimate a 10–25% premium vs prior-gen HBM pricing into H2 2026). Micron loses HBM4 share but gains LPDDR5X supplier leverage for Vera CPUs (up to 1.5TB/module demand), which can offset revenue loss if NVIDIA ramps Vera as a Xeon/EPYC competitor; NVDA’s 22 TB/s spec (up ~70% from 13 TB/s) forces suppliers to prioritize high-margin advanced parts.

Risk assessment: Tail risks include supply disruption at SK hynix (factory outage) or a last-minute Micron re-qualification that would compress HBM4 spreads; regulatory/antitrust scrutiny of supplier concentration is a medium-probability event over 6–18 months that could force design diversification. Timewise, market reactions should show in memory suppliers’ H1 2026 results and NVDA’s earnings run-up into late summer shipments; downside triggers include missed system volume targets or US export controls affecting Korean/Taiwan supply chains.

Trade implications: Direct plays: overweight SK hynix and Samsung memory exposure and long NVDA exposure into late-summer VR200 shipments; underweight or hedge Micron relative to SK hynix given lost HBM4 role but monitor LPDDR5X bookings which could re-rate MU. Options: favor calendar or call-spread structures around NVDA to cap premium paid for the late-summer catalyst and buy 3–6 month call options on SK hynix/Samsung to capture memory-price upside.

Contrarian angles: Consensus treats Micron as a net loser; we think LPDDR5X volume for Vera could monetarily offset >50% of lost HBM4 revenue if Vera gains CPU share vs Xeon/EPYC within 12–18 months — a risk markets underprice. Conversely, supplier concentration may be overbought: a 30–40% rally in SK hynix/Samsung into H2 2026 would be vulnerable to re-rating if NVIDIA diversifies or internalizes memory tooling over multi-year horizon.

Key Decisions for Investors

• Establish a 2–3% net-long position in NVDA using a 3–6 month call spread into late-summer 2026 (buy 3–6 month 25-delta calls, sell 12.5-delta calls) to capture upside from VR200 shipments while capping premium; trim if NVDA rallies >20% pre-shipment or if implied vol rises >30% above 90-day average.
• Allocate 3% long to SK hynix (000660.KS) or Samsung Electronics (005930.KS) via shares or 6-month call options to capture HBM4 price/vol recovery; target take-profit at +30% and stop-loss at -15% from entry, reassess on H1 2026 memory pricing/earnings prints.
• Initiate a relative-value pair: 2% long SK hynix (000660.KS) / 2% short Micron (MU) for 3–6 months to express HBM4 concentration risk; unwind if MU announces >20% HBM4 supply commitment to VR200 or if Micron posts sequential LPDDR5X design-win revenue >$300m in a quarter.
• Reduce direct exposure to Intel (INTC) and AMD (AMD) server-CPU beta by 1–2% of portfolio weight over the next 3 months (sell covered calls or trim positions), because NVIDIA’s Vera CPU + SOCAMM2 LPDDR5X could exert pricing pressure in OEM deals within 4–12 months; reinstate if Vera uptake lags benchmarks by >50% versus NVIDIA guidance.