Deepseek, a Chinese startup, is developing a custom AI chip tailored for inference tasks, according to Reuters. The chip is intended for the phase where a trained model generates responses for users, rather than for training new models. This move could help the company reduce its dependence on foreign chipmakers like Nvidia and Huawei. The project remains in its early stages, with Deepseek engaging with chip design, manufacturing, and memory companies while quietly hiring engineers for months without public job postings. Source: thedecoder
The chip development comes amid challenges posed by US export controls, which have limited Chinese companies' access to the most advanced semiconductor technologies. Deepseek is also raising outside capital for the first time, with reports indicating it is seeking $7 billion at a valuation between $52 and $59 billion. The company is not alone in this effort, as OpenAI and Anthropic are also working on their own custom chips. Source: thedecoder
According to three people familiar with the matter, the chip is specifically designed for inference, not for training new models. The project is still in its early stages, with Deepseek engaging with chip design, manufacturing, and memory companies while quietly hiring engineers for months without public job postings. Source: thedecoder
