DeepSeek, a Chinese AI company, is developing its own computer chip for artificial intelligence, according to reporting from Reuters citing sources with knowledge of the plans. The company is already talking to manufacturing partners and hiring engineers to build the team Reuters.

The chip is designed for a specific purpose: running AI models that have already been trained, rather than training new models from scratch. Think of it like the difference between having a chef and having a line cook — both important, but different jobs. Training AI is the expensive, demanding work upfront. Running a finished model is cheaper and less intensive, though it still needs serious computing power when you're doing it at scale.

Until now, DeepSeek has relied on chips made by Nvidia and Huawei. Those are the major suppliers of AI hardware, especially in China, where US export rules limit what companies can buy from American makers like Nvidia. Building its own chip would let DeepSeek depend less on outside suppliers.

Why Now?

This move reflects a broader pattern in DeepSeek's recent history. In June 2025, a US official said the company had used Nvidia chips obtained after the US banned their sale to China, and that DeepSeek had aided Chinese military applications while getting around export controls Reuters. A few months later, DeepSeek released an AI model that worked with homegrown Chinese chips. Then in February 2026, a Chinese official said DeepSeek's newest model had actually been trained on Nvidia's most advanced chips despite the ban Reuters. Within days, Reuters reported that DeepSeek was keeping that model away from Nvidia engineers, breaking from normal industry practice where hardware makers get early access to tune their chips Reuters.

This back-and-forth — using Nvidia when possible, switching to Huawei's chips when needed, and now building its own — shows a company hedging its bets across multiple options. Building a chip for inference (running existing models) is less complicated than building one for training, which may be why other big AI labs like Google and Amazon have done the same thing.

What Comes Next

The real question is where and how DeepSeek will actually manufacture this chip. US export controls on chip-making equipment mean that any chip built in China will likely use older, less advanced manufacturing than what Nvidia makes in Taiwan or South Korea. That matters for how much power the chip uses and how many can be produced without defects.

DeepSeek already operates under scrutiny from Washington following claims about military aid and export violations. A company designing its own AI chip while under this kind of attention will likely draw additional scrutiny from US policymakers. Building custom silicon changes the stakes of enforcement beyond what releasing new software would.

The broader industry context is worth considering. DeepSeek made headlines by building capable AI models at a fraction of what Western companies spent. Other major AI labs — Google, Amazon, even OpenAI through partnerships — have started making their own chips to control costs. The economics of running AI at large scale push companies in that direction, no matter where they operate. Whether DeepSeek can execute on chip design with the same efficiency it brought to making cheaper AI models is an open question. Chip manufacturing takes years, not months, so the answer will take time to emerge.