The Information: DeepSeek bets on Huawei chips, preparing a 2 trillion parameter model and planning an 8 trillion parameter model According to a report by *The Information*, DeepSeek is currently training a model with approximately 2 trillion parameters and has already planned the next-generation model with a scale of 8 trillion parameters. CEO Liang Wenfeng told investors that shifting to Huawei and other domestic chips is the company’s core bet, saying it “must succeed.” DeepSeek expects to receive a batch of Huawei chips for training the latest models in Q4 2026 or Q1 2027 and plans to invest around 30 billion RMB to expand computing power.