Relationship
NVIDIA relies on High Bandwidth Memory
Advanced NVIDIA AI accelerators depend on high-bandwidth memory to support data-intensive artificial intelligence workloads. Blackwell B200 uses 8x HBM3E stacks (192GB total); Blackwell Ultra uses 12-Hi HBM3E (288GB). HBM3E-generation supplier share across SK hynix/Samsung/Micron is disputed across analyst sources (reported, not reconciled). HBM capacity is gated more by cumulative TSV/die-stacking/test yield losses across the stacking process than by raw wafer supply alone (reported, industry analysis).
- From
- NVIDIA
- Importance
- Critical
- Confidence
- 98/100
Sources (4)
- SK Group and NVIDIA expand strategic partnership across AI factories and next-generation memory — CNBC / NVIDIA Investor Relations, July 24, 2026
- Micron hikes capex to $20B with 2026 HBM supply fully booked, HBM4 ramps 2Q26 — TrendForce, December 18, 2025
- NVIDIA Blackwell/Blackwell Ultra HBM3E specs and disputed supplier-share estimates — SiliconAnalysts / Wing VC / TrendForce, September 1, 2026
- HBM capacity constrained more by TSV/die-stacking/test yield across the process than by raw wafer supply — SemiEngineering / SiliconAnalysts, April 1, 2026
History
- 2025-10HBM3E 12-Hi memory began shipping into NVIDIA's B300 systems, with SK hynix holding roughly 62% HBM revenue share, Micron around 21%, and Samsung around 17% as the three qualified suppliers.
- 2026-03SK hynix shipped the first HBM4 12-Hi samples to NVIDIA for the upcoming Rubin platform, reported as a 6-9 month lead over Samsung and Micron; NVIDIA separately requested next-generation 16-Hi HBM4 samples from all three suppliers by Q4 2026.
Cite this
InsightNodes, "NVIDIA relies on High Bandwidth Memory" — Critical importance, 4 sources. https://insightnodes.com/claims/r/nvidia~hbm~depends-on (accessed October 29, 2026).
This page has a permanent address — link to it from notes, articles or models. Not investment advice.