Alibaba Group CEO Wu Yongming said at an earnings conference call that in the first quarter of 2027, the next generation of T-Head domestic chips will be fully capable of supporting ultra-large-scale model training, Securities Times reported on November 19.
'We have already started training next-generation models on this year's T-Head chips,' Wu said. 'As the capacity of domestic chips continues to expand, it will be a long-term support for the development of Alibaba's models. In the long run, most models, whether for Alibaba or our partners and customers, will rely entirely on domestically produced chips for both pre-training and post-training.'
The remarks highlight a broader push by Chinese technology companies to reduce dependence on foreign semiconductors for AI. Domestic GPU and chipmakers have been expanding capacity as large-model developers seek cheaper and more secure computing options, and Alibaba's statement signals that in-house and locally designed silicon is moving from trials to core training infrastructure.





