Sources: DeepSeek and Huawei are collaborating on future AI chip designs; DeepSeek plans to start using Ascend 950DT for pre-training as soon as 2026's end

If you train large AI models, the cost of advanced training hardware in China will fall.

Chinese AI firm DeepSeek is reportedly collaborating with Huawei Technologies on the design of future AI chips. Sources familiar with the matter indicate that DeepSeek plans to utilize Huawei's Ascend 950DT chips for its large-scale AI model pre-training, potentially beginning as early as the end of 2026. This partnership aims to bolster China's capabilities in developing advanced AI models, leveraging the strengths of both companies in AI research and hardware manufacturing.

This collaboration signifies a significant push within China to achieve greater self-sufficiency in cutting-edge AI hardware and model development. By integrating DeepSeek's AI model expertise with Huawei's Ascend chip technology, the partnership aims to create more powerful and cost-effective solutions for training frontier AI. The move could accelerate the development and deployment of sophisticated AI applications originating from China, potentially challenging existing global leaders.

The decision by DeepSeek to adopt Huawei's Ascend 950DT for pre-training suggests growing confidence in the capabilities of China's domestic AI chip industry. This strategic alliance may set a precedent for future hardware-software co-design efforts, fostering an ecosystem where AI model requirements directly influence chip architecture. The timeline indicates a medium-term commitment, with tangible impacts expected to emerge after 2026 as these new designs and training protocols are implemented.