DeepSeek wants to develop its own AI chips to reduce dependence on Nvidia and Huawei

Chinese AI company DeepSeek, headquartered in Hangzhou and owned by hedge fund High-Flyer, is working on its own chips for AI inference. The goal is to reduce its dependence on both Nvidia and China's Huawei, following successive US export restrictions that have progressively curtailed Chinese companies' access to high-end semiconductors.

The plans are still at an early stage. The chips are specifically intended for inference, the process by which a trained model responds to user queries, and not for training new models. That distinction matters: inference places different, and in several respects less extreme, demands on hardware than training large language models.

DeepSeek previously gained wide recognition for the relatively low training costs of its R1 model. That model was run on Nvidia's H800 chips. The US government banned the export of those chips at the end of 2023, after which DeepSeek had to rely more heavily on Huawei's Ascend processors. With its own chips, the company aims to reduce that dependency as well.

Export controls as the direct trigger

US export regulations have been affecting Chinese technology companies on an increasingly broad front for several years. For DeepSeek, the situation became concrete when the H800, a restricted version of Nvidia's H100 designed for the Chinese market, also fell under export restrictions at the end of 2023. The company subsequently shifted largely to Huawei's Ascend chips, but those are technically less advanced than the top-tier Nvidia products that Western AI labs use freely.

Developing its own chips is a logical next step when both external suppliers are unreliable or limited. At the same time, it is a costly and technically demanding path: the semiconductor industry has high barriers in terms of design, manufacturing and software ecosystems. That DeepSeek is taking the step nonetheless is also linked to the company's significantly improved financial position.

Large funding round strengthens its position

In June 2026, DeepSeek closed a funding round equivalent to approximately $7.4 billion, or more than 50 billion yuan. This pushed the company's valuation above $50 billion. By comparison, earlier that year, in April, a round of $300 million was being discussed at a total valuation of $10 billion.

The largest contributions in the June round came from founder and CEO Liang Wenfeng himself, who contributed approximately $3 billion from his personal wealth, from technology conglomerate Tencent with roughly $1.48 billion, and from battery manufacturer CATL with $740 million. China's National AI Industry Investment Fund also participated and obtained voting rights in the company.

Parties such as IDG Capital, Monolith Capital, JD.com and NetEase are reported to still be in talks about participation. As of May 2024, Liang Wenfeng held an 84 percent stake in DeepSeek through two holding vehicles. High-Flyer, the quantitative hedge fund he co-founded, initially financed the company entirely from its own returns.

Western tech companies remain involved in model optimisation

Notably, DeepSeek collaborates with AWS, Nvidia and Intel on the optimisation of its AI models. This software-level collaboration is separate from the hardware plans and is not affected by the export restrictions, which specifically concern advanced chips. It illustrates that the relationship between Chinese and American technology companies has multiple layers even in the current climate.

Microsoft is also exploring whether a future DeepSeek V4 model could be deployed as a lower-cost option in Copilot Cowork. No decision on this has been announced. The training costs of DeepSeek-R1, estimated at around $6 million, had already compared favourably with the estimated $100 million that OpenAI is said to have spent on GPT-4 in 2023. That cost differential makes DeepSeek attractive to commercial partners, independent of the geopolitical complications.

What chip development means for the broader ecosystem

DeepSeek is not the only Chinese company investing in its own semiconductor technology in response to export restrictions. Alibaba, Baidu and Huawei each have their own chip programmes under way. What sets DeepSeek apart is the combination of relatively modest training budgets and still competitive model performance, an approach that has attracted the attention of investors and researchers outside China.

For European and Dutch AI companies and policymakers, this sketches a backdrop in which hardware independence is increasingly viewed as a strategic priority. The debate over access to computing power is also playing out in Europe, albeit from a different position: European parties are not subject to Chinese export restrictions, but are heavily dependent on American chipmakers and cloud infrastructure. The way DeepSeek builds efficient models with limited resources is, in that light, being watched on this continent as well by labs and policymakers thinking about sovereign AI capacity.