DeepSeek and Huawei team up on AI chip design in a bid to take on Nvidia

Dataconomy reported on October 20 that DeepSeek is teaming up with Huawei on AI chip design, a collaboration framed as a direct challenge to Nvidia's position in AI compute.

According to the report, the arrangement goes beyond procurement. DeepSeek is described as taking part in shaping the design of future chips so that the silicon fits the computational profile of its own models, which in turn would let the company train advanced systems with less dependence on Nvidia hardware.

The sourcing deserves to be stated plainly. Dataconomy's account rests on unnamed sources rather than a joint announcement, and neither company has publicly confirmed a co-design programme on the record. The scope, timeline and product form of any collaboration therefore remain unknown.

Set against the 2026 landscape, the significance is easier to read. Chinese model developers have spent years working around constrained access to high-end GPUs, while Huawei is one of the few domestic suppliers with a full stack running from silicon to software tooling. When a model company opens up its training requirements to a chip company, what the two sides trade is a roadmap: chips designed around the model, and the model tuned to the chip.

The returns from hardware-software co-design run both ways. The chip side learns early which operators the model will lean on next and can avoid building for the wrong workload; the model side gains efficiency that is not available off the shelf, effectively converting engineering effort into performance that would otherwise have to be bought.

There is a cost to that intimacy. Co-design creates lock-in: if the chip roadmap and the model roadmap diverge, switching becomes expensive, and a chipmaker optimising for a single large customer has to balance bespoke tuning against the needs of everyone else.

If the collaboration holds, the shift is less about one company's procurement list than about who supplies the compute behind China's frontier models.

Three things are worth watching: whether either company puts its name to the arrangement; whether later Huawei parts show unmistakable signs of being designed around one model family; and whether DeepSeek's next generation can post a convincing training run on domestic silicon. Until then, this is a directional signal rather than a settled fact.