HBM4 allocation & roadmap — mass prod confirmed for Nvidia, HBM4e/custom 2027-28
Key Questions
What is the timeline for HBM4 mass production?
Mass production of HBM4 is ramping up in mid-March for NVIDIA's Vera Rubin platform. Full allocation for 2026 is locked in, with HBM4E and custom variants expected in 2027-28.
How is NVIDIA's Vera Rubin platform involved with HBM4?
Vera Rubin is sold out for 2026-27, featuring 48GB per 16 stacks of HBM4. NVIDIA has confirmed mass production allocations, driving strong demand.
What pricing changes are expected for HBM4?
HBM4 pricing is up 60-70% compared to previous generations. This reflects sustained AI shortages and a memory supercycle.
What is the market share outlook for HBM4 suppliers?
Suppliers are targeting 25-21% share versus SK Hynix. Hybrid bonding provides a competitive edge in production.
Is HBM3E supply also constrained?
Yes, full 2026 NVIDIA allocations for HBM3E are sold out. This contributes to ongoing AI memory shortages.
How significant is hyperscaler spending on memory in 2026?
Hyperscalers are projected to allocate 30% of their spend to memory in 2026. This sustains the AI-driven supercycle.
What role does Micron play in NVIDIA's AI platforms?
Micron is ramping production for NVIDIA Vera Rubin systems and emerging as a key AI chip memory player. AI demand has ignited a memory supercycle with prices soaring.
Why is the AI memory market in a supercycle?
AI shortages and high demand from NVIDIA platforms sustain the supercycle. DRAM prices have surged 90-95%, with suppliers booked out for two years.
Ramping mass prod mid-Mar for NVIDIA Vera Rubin ('26-'27 sold-out, 48GB/16H, +60-70% pricing/25-21% share vs SKH); full '26 Nvidia locked/HBM3E sold-out; hyperscaler 30% memory spend '26; AI shortages sustain supercycle; hybrid bonding edge.