Reuters: DeepSeek Developing Its Own AI Inference Chips to Reduce Reliance on Nvidia and Huawei

Chinese artificial intelligence company DeepSeek is developing its own AI inference chips, according to a Reuters report citing three people familiar with the company's plans. The move represents a significant strategic escalation from software-only innovation toward full-stack hardware integration.

Sources said DeepSeek has been exploring the concept of proprietary AI accelerators for about a year, with the initiative still in early stages. The company has contacted outside partners, held discussions with chip design firms, foundries, and memory companies, and is actively recruiting experienced chip designers.

The focus on inference chips is strategic: inference — where a trained model generates responses for users — is the stage that becomes the recurring cost center and revenue driver for AI companies. While DeepSeek shot to fame in January 2025 with its low-cost, high-performing R1 model trained on Nvidia's H800 chips, inference economics are where long-term competitive advantages are built.

DeepSeek's goal appears to be reducing reliance on both Nvidia and Huawei Technologies, its two primary AI chip suppliers. After the U.S. banned exports of advanced Nvidia chips to China, DeepSeek turned increasingly to Huawei's Ascend GPUs. But sources told Reuters the company does not want to become overly dependent on Huawei either.

The approach mirrors strategies by U.S. AI firms like OpenAI, which designed a custom inference chip called Jalapeño in collaboration with Broadcom, and Anthropic, which is also pursuing custom silicon. The shared motivation is gaining greater control over the technology stack and reducing dependence on a single supplier.

DeepSeek is believed to have sufficient funding for this capital-intensive endeavor, having recently raised money from a host of Chinese investors. Rivals including Alibaba and Baidu are also developing their own AI processors, indicating a broader industry trend in China.

Developing competitive AI chips requires world-class design teams, multiple tape-out cycles, and manufacturing capacity. Whether DeepSeek can successfully deliver a production-ready inference chip while maintaining its edge in foundational model research will be a key story to watch in the coming months.