HBM demand shifts: low-stack inference plans shelved, Rubin to go 12-high only
BenBajarin · x · 2026-10-26
Per Daishin Securities, CSPs and AI labs have shelved plans for 4-high/6-high HBM in low-end inference products, with 8-high again the minimum stack. NVIDIA's Rubin reportedly dropped the 8-high option from its planned 8/12-high mix amid base-die supply constraints and customers prioritizing bandwidth; multiple Rubin Ultra variants (2027) are in tape-out. Ben Bajarin adds that 8/12-high remains for training SKUs (a shrinking share) while 4-6-high suits inference.
More from Infra
- OpenAI's initial Vera CPU results: 51% more Python throughput vs leading x86 — firstadopter · 2026-10-27
- Cloudflare discloses cross-request KV cache leak in GLM-5.3 on Workers AI: 155 responses affected — michellechen · 2026-10-27
- Cloudflare discloses cross-request KV cache exposure in GLM-5.3 on Workers AI (155 responses) — wondersky · 2026-10-27
- One month fully local on a DGX Spark: daily tasks fine, code still needs Claude to check — michaelthatsit · 2026-10-27
- Cloudflare Workers adds spanContext, setStatus and updateName tracing APIs — irvinebroque · 2026-10-27
- M5 Max benchmark: Kapitsa hits 36.6 tok/s with KFlash, 28GB at 32K vs mlx-lm's 52GB — DotaMate · 2026-10-27