DeepSeek starts work on its own chip for running AI models
Chinese AI company DeepSeek is developing a ed for inference, the process of running a trained AI model to produce answers. Three people familiar with the work said the project is still at an early stage.
DeepSeek has increased hiring for design and is discussing possible with manufacturers. A successful chip could reduce its reliance on Nvidia and Huawei hardware, which it currently uses to train and run its models.
US limits on advanced chip exports to China and Beijing's push for domestic technology add pressure to this effort. Designing the chip is only one step; limited manufacturing capacity, memory, and packaging could still prevent large-scale .
Key points
- DeepSeek is developing its own chip for inference rather than .
- The effort is at an early stage and relies on reports from three unnamed sources.
- DeepSeek is hiring more designers and speaking with manufacturers.
- Its goal could include reducing dependence on Nvidia and Huawei chips.
- No price, result, date, or customer availability has been announced.