Rumor: DeepSeek pretraining 2T-param V5 on Huawei Ascend 950 chips
Kyrannio · x · 2026-11-12
An unverified rumor circulating on Chinese X claims DeepSeek has started pretraining V5, a 2 trillion-parameter ultra-sparse MoE model, natively on Huawei's new Ascend 950 / 950DT chips instead of NVIDIA GPUs — with Huawei kernel engineers reportedly working on-site to keep the cluster running. If true, a frontier-scale run on Ascend silicon would reshape the AI hardware story. Note: this remains a rumor with no official confirmation. Reposter Jason used it to argue "America needs NVIDIA to win," sharing a screenshot of buying 2,000 NVDA shares at $159.97 ($319,940).
Related event: Rumor: DeepSeek pre-training 2T-parameter V5 on Huawei Ascend 950(4 posts)→
More from Infra
- SGLang Q4 roadmap: 10+ accelerator families supported, plugin system makes new backends first-class — ying11231 · 2026-11-12
- SGLang ships Unified Memory: one GPU memory arena for every token lifetime, no partitioning knob — ying11231 · 2026-11-12
- Polymarket odds hit 72% for Jensen Huang's removal by tomorrow amid board investigation — Polymarket · 2026-11-12
- Cerebras explains wafer-scale yield: independent voltage tuning across 84 die regions plus redundant cores — Sethwinterroth · 2026-11-12
- gVisor joins CNCF with Modal as maintainer, running 40M+ containers daily — charles_irl · 2026-11-12
- Hugging Face Launches Open-Source Llama for Windows, Bringing No-Setup Local AI to the Masses — unixterminal · 2026-11-12