DeepSeek confirms gigawatt-scale ambitions, designing own 128-GPU supernodes instead of Huawei 950 pods
teortaxesTex · x · 2026-09-30
teortaxesTex notes the first confirmation that DeepSeek targets infrastructure from megawatt to gigawatt scale. DeepSeek specifically says it worked on 128-card 'supernodes' — about two cabinets' worth of Huawei Ascend 950 — but shows no sign of buying Huawei's prefab 950 pods, instead designing its own systems, consistent with other filings.
More from Infra
- DeepSeek open-sources Huawei Ascend toolkit with TileLang support, challenging Nvidia's CUDA — kimmonismus · 2026-09-30
- Orbital compute is coming: SpaceX and Tesla target 200GW solar per year, says AI KOL — tetsuoai · 2026-09-30
- On-prem LLM stack: the leaks you miss are OCR parsing and hosted eval judges — lucasbennett_1 · 2026-09-30
- Batching vLLM calls with shared prefixes: a warm-up-then-fan-out KV cache strategy — Theboyscampus · 2026-09-30
- Memory price surge threatens on-device AI hardware: BOM tops $300, device may cost $1,500 — stephen280up · 2026-09-30
- Is local hardware worth it for AI agents? Reddit math: 5-10 years to break even vs cloud — pixpuz · 2026-09-30