Chinese AI startup DeepSeek, known for its breakthrough models such as R1 and V4, has entered the race for its own hardware. According to my data, the company has been developing its own processor for about a year, designed for inference — the stage of executing neural network models after training. This is an important signal for the market: DeepSeek aims to reduce dependence on external suppliers, including Nvidia and Huawei.
Currently, the project is in its early stages. DeepSeek is actively negotiating with partners for chip design, contract manufacturing, and memory supply. Additionally, engineers for processor design are being hired in a closed mode. The company has not commented on these steps, which only heightens interest in their strategy.
Previously, DeepSeek used Nvidia H800 chips adapted for the Chinese market and actively collaborated with Huawei, applying their Ascend processors in the V4 and V4-Flash models. However, as I note, U.S. export restrictions and growing demand for inference solutions make a proprietary chip not just an ambition but a necessity. The startup's founder, Liang Wenfeng, already in 2024 called control over chip supply a key issue.
DeepSeek is entering a battle with giants like OpenAI, Broadcom, and Alibaba, which are already investing in their own infrastructure. However, success is not guaranteed: high competition, enormous capital expenditures, and limited access to advanced foreign factories create serious barriers. Nevertheless, this step demonstrates DeepSeek's drive for technological autonomy.
Analytical opinion: Amid geopolitical tensions and semiconductor export restrictions, creating a proprietary chip for inference is a logical but risky move. If DeepSeek manages to successfully integrate it into its ecosystem, this will not only reduce dependence on Nvidia and Huawei but also provide advantages in data processing speed and cost. However, this will require years of development and billions in investment.