DeepSeek is reportedly moving toward a major shift in its AI strategy as the Chinese startup looks to develop its own custom chip for inference, reducing its dependence on hardware from NVIDIA and Huawei.
At the moment, DeepSeek’s AI models are being trained and deployed using chips from major suppliers, including NVIDIA and Huawei. However, the company is said to be exploring an in-house semiconductor designed specifically to run pre-trained AI models and generate responses for users. The chip is not expected to be used for training new models, which requires far more computing power, but instead for inference, the stage where an AI system delivers answers, summaries, code, or other outputs after training is complete.
This move could give DeepSeek greater control over its AI infrastructure at a time when demand for artificial intelligence computing power continues to surge. By building its own inference chip, the company could potentially reduce costs, improve efficiency for its own models, and become less exposed to supply limitations or pricing pressure from external chip providers.
DeepSeek’s reported plan also reflects a broader trend across China’s technology sector. Major Chinese companies such as Alibaba and Baidu have already been working on their own AI chips as they seek to reduce reliance on foreign hardware and strengthen their domestic technology ecosystems. Huawei has built a strong position in China’s AI chip market, especially as access to advanced NVIDIA hardware has become more restricted. DeepSeek is currently among the companies using Huawei’s chips, but its interest in internal chip development suggests it wants more independence over the long term.
According to people familiar with the matter, DeepSeek has already started recruiting chip design engineers in recent months. The hiring process has reportedly been handled quietly rather than through public job postings, suggesting the project is still in an early and sensitive stage.
Even so, building a competitive AI chip is an extremely difficult challenge. Semiconductor development can take years and often requires billions of dollars before a product is ready for mass production. The process includes chip architecture design, verification, tape-out, testing, manufacturing, and optimization, all of which demand deep expertise and significant financial resources.
DeepSeek’s biggest hurdle may be manufacturing. Due to U.S. trade restrictions, Chinese companies have limited access to the world’s most advanced chipmaking equipment. That means DeepSeek would likely need to rely on domestic semiconductor manufacturing options, such as SMIC’s older 7nm-class process. While capable, this technology is less advanced and less power-efficient than the cutting-edge manufacturing nodes used by global leaders in AI hardware.
The company also cannot easily access the most advanced EUV lithography machines produced by Dutch supplier ASML, as those systems are restricted for Chinese entities under export controls. Without access to the latest production tools, DeepSeek may face difficulty matching the performance, energy efficiency, and scale of top-tier AI chips built using more advanced manufacturing processes.
Still, the potential benefits explain why DeepSeek may be willing to take on the risk. Custom inference chips can be tailored for a company’s own AI models, allowing better performance per watt and lower operating costs over time. For a fast-growing AI company serving large numbers of users, even modest gains in efficiency can translate into major savings.
The move could also help DeepSeek protect itself from supply chain uncertainty. As AI adoption accelerates globally, demand for high-performance GPUs and AI accelerators has remained intense. Companies that depend entirely on outside suppliers may face shortages, rising costs, or geopolitical constraints. Developing internal silicon could give DeepSeek a stronger foundation for future expansion.
Funding may play a key role in whether the chip effort succeeds. DeepSeek was previously reported to have raised around $7 billion in a funding round that valued the company between $52 billion and $59 billion. That would mark a significant change from its earlier reluctance to accept outside investment and could provide the capital needed to pursue long-term hardware ambitions.
For now, DeepSeek’s custom AI chip project appears to be in the early stages, and there is no guarantee it will result in a competitive product. The company faces technical, financial, and manufacturing barriers that even experienced chipmakers struggle to overcome.
However, if DeepSeek manages to deliver a capable inference chip, it could reshape its position in the AI market. The company would gain more control over the hardware behind its models, reduce reliance on NVIDIA and Huawei, and strengthen its role in China’s push for greater AI and semiconductor independence.
