Huawei says China's AI training will shift to Ascend 950DT SuperPoDs starting 2027
pstAsiatech · x · 2026-09-18
Huawei rotating chairman Eric Xu Zhijun said at Huawei Connect 2026 in Shanghai that starting next year, much of China's AI model training will run on SuperPoD or SuperCluster infrastructure built on Ascend 950DT processors.
Xu claimed Huawei's proprietary Ascend chips have overtaken Nvidia in domestic market share, though he gave no specific figures. Context: Nvidia resumed H200 shipments to China last month under a new US licensing arrangement, but those sales accounted for less than 1% of its Q2 data-center revenue.
More from Infra
- AI ported a distro in 20 minutes — is NVIDIA sawing off its own CUDA moat? — BringTea_666 · 2026-09-18
- India Lands $12 Billion in Semiconductor Investment Pledges Within Months — pstAsiatech · 2026-09-18
- NanoGPT Speedrun hits new 68.0s record with 96-dim QK and packed FP8 attention — kellerjordan0 · 2026-09-18
- Baidu Cloud bets on 'capability mass production' for embodied AI, serving 50+ robotics firms — 量子位 · 2026-09-18
- Hedge fund CIO: Anthropic burns maybe 80% less per token than OpenAI — rohanpaul_ai · 2026-09-18
- Engineer: companies hide servers from management to dodge forced cloud migrations — irth7 · 2026-09-18