TREE NEWS reports: DeepSeek plans to use Huawei’s Ascend 950 chips as the core training compute for its next-generation DeepSeek V5 model, with the project targeted for launch next year, people familiar with the supply chain said. The companies’ core R&D integration team is based in Hangzhou, and the two sides completed extensive low-level operator adaptation and custom development during the V4 stage. Ascend compute handled only inference in DeepSeek’s previous model generation; V5 would be its first model trained natively on Ascend 950.
DeepSeek V5 to Train on Huawei Ascend 950 Chips Next Year
The shift from inference to native training is the substantive part: inference is a narrower, more forgiving workload, while training on domestic silicon tests whether the Ascend software stack can carry a frontier-scale model end to end. The V4-stage operator adaptation suggests this is an incremental build rather than a cold start, which matters for how credible the timeline is. The open question is whether native Ascend training becomes a repeatable pattern across future model generations, or remains a single-project arrangement.
Generated by AI for reference only.
Share on WeChat
Open WeChat → Scan → then tap "…" to send to a chat or Moments.
Tap "…" in the top-right corner to send to a chat or share to Moments.