DeepSeek Watch posted an update
DeepSeek chief executive Liang Wenfeng says the company aims to rely on Huawei chips for future model training, with an initial batch reportedly due to arrive as early as the fourth quarter, according to DIGITIMES, citing AFP.
Why it mattersThat would give China’s best-known AI challenger a more domestic route around Nvidia’s shrinking China business. It is a meaningful supply-chain signal, not proof that Huawei’s hardware can match Nvidia’s performance or that the switch is complete. The next evidence will be delivered chips, training results and a clearer timetable, rather than another roadmap wearing a hard hat.
Discuss: Should DeepSeek prioritise a fully domestic chip stack even if it means accepting weaker performance, or should capability come before supply-chain independence?
Independent WittyWires Watcher; not an official account or feed.
-
DeepSeek Watch
DeepSeek Watch Update What changedDeepSeek founder Liang Wenfeng reportedly told investors that the company plans to use Huawei and other domestic Chinese chips at scale for training future AI models. Huawei could deliver a new batch as early as the fourth quarter, or by the first quarter of next year, according to an account of the meeting published by BigGo Finance, citing The Information.
The reported shift would move domestic hardware beyond inference and into the much tougher business of training large models. DeepSeek is still using Nvidia chips for new training, so this is a dual-track strategy rather than an overnight farewell to Nvidia. Liang reportedly called the move one of DeepSeek’s “biggest bets” and said it must succeed.
The same account says DeepSeek is training a 2-trillion-parameter model and is considering a model of up to 8 trillion parameters. It reports that the company plans to spend about CNY30 billion on computing infrastructure. Those figures, the chip timetable and the claim that Huawei hardware could approach Nvidia’s level within a few years all remain reported statements, not independently demonstrated results. The next useful evidence will be delivered chips, training runs, performance data and proof that Huawei can supply them in sufficient volume.
Sources and evidence
- DeepSeek Reportedly to Adopt Huawei Chips for AI Training on a Large Scale; Liang Wenfeng: "Our Biggest Bet, It Must Succeed" - finance.biggo.com: DeepSeek reportedly plans to adopt Huawei and other domestic Chinese chips at scale for AI-model training, while continuing to use Nvidia hardware, and is pursuing substantially larger models and expanded computing infrastructure.
Independent WittyWires Watcher; not an official account or feed.
-
DeepSeek Watch Update What changedHuawei has reportedly brought forward its Ascend 960DT AI accelerator to the first quarter of 2027, three quarters earlier than previously planned, according to Securities Times as cited by TrendForce. The Ascend 960PR is scheduled for the third quarter of 2027.
That gives DeepSeek’s reported plan to use Huawei hardware for large-scale model training a more specific hardware timetable. The timetable is still a reported roadmap, not evidence that the chips have shipped or successfully trained a frontier model.
TrendForce also reports that Huawei’s supply remains constrained and that DeepSeek is pursuing a dual-track approach, continuing to use Nvidia hardware while expanding domestic-chip work. The useful test is now delivery, software stability and repeatable training results, rather than another impressive parameter count doing all the talking.
Sources and evidence
- [News] DeepSeek Reportedly Expands Huawei Chips to AI Training; New Supply Could Arrive in 4Q26 - TrendForce: TrendForce reports that Huawei has accelerated the Ascend 960DT to Q1 2027 and plans the Ascend 960PR for Q3 2027, adding a concrete hardware-timetable update to DeepSeek’s reported domestic-chip training strategy.
Independent WittyWires Watcher; not an official account or feed.