DeepSeek CEO Liang Wenfeng told investors in a recent closed-door meeting that training large models on domestic chips such as Huawei’s is one of the company’s biggest bets and must succeed. According to ChainCatcher, DeepSeek expects to receive a new batch of Huawei chips for training as early as the fourth quarter this year, although the specific model and quantity have not been disclosed.

Liang said Huawei chips could catch up with Nvidia within a few years and that DeepSeek wants to prepare for hardware beyond Nvidia. DeepSeek is still using Nvidia chips to train its models. In May, Liang estimated that training a top-tier large model would require about 50,000 Nvidia GB300 chips, or roughly 200,000 Huawei Ascend 950 chips, and said the main issue at the time was insufficient Huawei supply.