China seeks alternatives to NVIDIA GPUs

China is working to reduce its reliance on NVIDIA's graphics processing units (GPUs), which are critical for training artificial intelligence (AI) systems. This effort is part of a broader strategy to develop domestic technology and lessen dependence on foreign components, particularly in the high-stakes AI sector.

Huawei partners with Deepseek on AI chips

Huawei has formed a partnership with Deepseek, a prominent AI company, to co-design AI chips. This collaboration involves Huawei engineers working directly with Deepseek to integrate feedback and improve their chip designs, aiming to create a more competitive alternative to NVIDIA's offerings.

Open-source ecosystem for AI development

The partnership between Huawei and Deepseek includes the development of an open-source ecosystem that supports both software and hardware. This ecosystem is intended to provide a comprehensive platform for AI development, using local AI models and hardware architecture.

Complementary strengths in the partnership

Deepseek contributes expertise in identifying computational challenges during AI model execution, while Huawei focuses on enhancing the memory bandwidth efficiency of its chips. This combination of strengths is meant to produce more effective and efficient AI hardware.

Commercial plans and future goals

Deepseek plans to use Huawei's Ascend 950 DT chips by the end of the year and aims to prioritize the upcoming Ascend 960 DT chips, expected to launch in early 2027. This commercial commitment highlights the potential of the partnership, though it remains to be seen when these chips can match the performance of NVIDIA's GPUs.

Performance gap with NVIDIA's GPUs

Currently, NVIDIA's B300 GPU outperforms Huawei's Ascend 950 by a factor of seven in standard AI calculations. This performance gap underscores the challenge Huawei and Deepseek face in developing competitive alternatives to NVIDIA's technology.