Reuters reports that Chinese startup Deepseek is building its own AI chip designed for inference rather than model training. This hardware project aims to reduce reliance on Nvidia and Huawei processors while the company quietly hires engineers and engages with design and manufacturing firms. The initiative occurs alongside a first-time fundraising round seeking $7 billion at a valuation between $52 and $59 billion. US export controls restrict access to advanced components, creating significant hurdles for Chinese firms. OpenAI and Anthropic are pursuing similar in-house silicon strategies. The chip development sits in early stages without public job listings or detailed technical specifications available yet.
Developing custom silicon allows Deepseek to optimise hardware for its specific models without paying premium prices for generic units. Controlling the inference stack reduces costs per query and improves response speeds for end users. The move signals a shift from renting compute power to owning the underlying infrastructure required to run large language models.
* Chips target inference, not model training
* Valuation ranges from $52 to $59 billion
* Seeking $7 billion in external capital
