Huawei reportedly deploying 256K-chip 950 SuperCluster aimed at frontier-scale training
teortaxesTex · x · 2026-09-18
X user teortaxesTex claims Huawei is deploying a 256K-chip 950 SuperCluster. Citing Huawei's Wenfeng, he notes 200K 950DT chips is the floor for training an 800B-parameter active model at the absolute US frontier — arguing the cluster's purpose is frontier-scale training, not inference.
More from Infra
- OpenAI and Anthropic hunt for 20-30 MW compute deals in UK and Nordics to serve inference — demian_ai · 2026-09-18
- 54k-star proxy runs Claude Code, Codex and 8 more coding agents through 50 free providers — eyishazyer · 2026-09-18
- UK scholar warns: most AI-exposed economy has almost no data center plan — sudoraohacker · 2026-09-18
- Anthropic to scale compute to 5GW by year-end, 10GW by 2027, NYT reports — basedjensen · 2026-09-18
- Ternary Bonsai 2 WebGPU kernels space trending on Hugging Face — webml-community · 2026-09-18
- Polymarket rebuilds API gateway in Rust, pages now 4-8x faster — Polymarket · 2026-09-18