Nvidia cuts Rubin Ultra memory to 192GB as HBM costs soar to 40% of TCO
rwang07 · x · 2026-09-02
Nvidia has despecced Rubin Ultra's HBM from HBM4E 12-Hi (384GB) down to HBM4 8-Hi (192GB). The move comes after recent HBM and DRAM price hikes, which made memory quietly account for 40% of total capital cost of ownership for rack-scale systems.
More from Infra
- Is it worth switching to AMD for AI Video Generation? ROCm Inquiry — iridescentblob · 2026-09-02
- HybridInfer: Router auto-falls back to cloud when local model wedges — simrankoulsm · 2026-09-02
- Looped transformer rumors may be true, bearish for HBM demand — zephyr_z9 · 2026-09-02
- Buying GPUs to run local models is pointless, argue devs: cloud APIs win on cost — brandon_galang · 2026-09-02
- HybridInfer: open-source router runs local models first, auto-falls-back to cloud on stalls — simrankoulsm · 2026-09-02
- 4,000 GB200s arrive in Texas for Horizon TACC, billed as largest academic supercomputer — AlexGDimakis · 2026-09-02