DeepSeek and Huawei team up on AI chip design to challenge Nvidia

DeepSeek plans to invest in Huawei’s AI chip design, with the companies collaborating on new chips and software optimization, The Information reported.

The companies plan to rework Huawei’s Ascend chip design to improve performance. DeepSeek will provide Huawei with computational data problems encountered when running large language models.

Huawei will focus on improving memory bandwidth efficiency and the interconnect framework of next-generation Ascend chips. The collaboration aims to reduce difficulties in training large models, which can require thousands of chips working simultaneously on lengthy computations.

The companies also plan to develop an open-source ecosystem that includes a local AI model supporting the full hardware architecture and software stack.

Huawei will strengthen its CANN computing platform to compete with Nvidia’s CUDA. Ascend chips faced substantial limitations after U.S. sanctions hit Huawei, though the company has improved them over time.

DeepSeek is expected to pursue an ultra-low-cost design for high-efficiency AI models.

Huawei has announced its Ascend chip roadmap for 2027. The company plans to unveil the Ascend 960DT in the first quarter, while the Ascend 960PR could be introduced in the third quarter.

Featured image credit