DeepSeek, a prominent Chinese startup specializing in large language models, is planning to enter the silicon industry to develop its own hardware. According to Reuters, the company has been working on this initiative for approximately one year, meeting with potential hardware partners and hiring specialized engineers. The strategic focus is on designing data center chips specifically for inference rather than training. This move aims to decrease DeepSeek's dependency on Nvidia, which faces US export bans in China, and Huawei, which currently dominates half of the local data center chip market. This trend is not unique to China; US-based firms like OpenAI and Anthropic are also exploring custom chip designs to gain greater control over their technology stacks and reduce reliance on external providers like Nvidia amidst increasing competition for compute resources.