AI

DeepSeek V5 is planned to be pre-trained on Huawei Ascend chips.

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 03/11/2026, 21:47:25 UTC.

Briefings

Industry sources indicate DeepSeek V5 will be pre-trained on Huawei Ascend chips, marking the first time China's top-tier large models are trained on domestic hardware. This follows a 2025 Spring Festival deployment of over 3,000 Ascend chips for inference, with Huawei and DeepSeek completing model hosting at Huawei Cloud's Guian Data Center. The company anticipates a price reduction for the Pro version once the Ascend 950 super-node series enters mass-market availability in the second half of the year. DeepSeek has also released open-source infrastructure components tailored for the Ascend platform, including DeepGEMM and DeepEP libraries, though full training remains concentrated on inference currently.

Key points

• DeepSeek V5 is planned to be pre-trained on Huawei Ascend chips, marking the first time China's top-tier large models are trained on domestic hardware.
• Over 3,000 Ascend chips were deployed and verified between Lunar New Year's Eve and the second day of the Lunar New Year for inference tasks at Huawei Cloud's Guian Data Center.

Context

DeepSeek's next-generation large model, DeepSeek V5, is planned to be pre-trained using Ascend chips. Previously, Huawei's Ascend chips were primarily utilized for inference tasks for DeepSeek's large models. The first large-scale deployment of DeepSeek models on the Ascend platform took place around the 2025 Spring Festival.

Limitations

Evidence is limited to industry supply chain reports and Huawei's official announcements. The initial pre-training phase for DeepSeek V5 has not yet been completed; only inference deployments and partial post-training on domestic hardware are confirmed. Costs associated with migrating training ecosystems remain a critical variable not yet resolved.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 03/11/2026, 22:09:35 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeiglqecanuvqfqwifnf3mts4al5xdchr42lcoje442r3h5gdtgwfba