DeepSeek, the Chinese AI startup behind the open-source DeepSeek-V3 and R1 models, has initiated an in-house AI chip...

DeepSeek, the Chinese AI startup behind the open-source DeepSeek-V3 and R1 models, has initiated an in-house AI chip project focused on inference workloads, as reported by Reuters and corroborated by multiple outlets. The move aims to reduce DeepSeek's dependence on overseas suppliers like NVIDIA and domestic rival Huawei, amid US export controls that have limited access to advanced GPUs. According to Reuters (Source 1: Technode), the project is in early stages, with DeepSeek quietly recruiting chip engineers over the past year. Ars Technica (Source 3: Ars Technica) notes that the chip will target data center inference, not training, and aligns with a broader trend of AI companies (e.g., OpenAI, Anthropic) developing custom silicon. Semafor (Source 4: Semafor) highlights that this could reshape China's semiconductor race and pressure NVIDIA. DeepSeek's chip effort is part of its strategy to reduce compute costs, which account for over half of operating expenses, and is likely to be funded by its planned $7 billion fundraising round.

What the sources said

• Reuters via Technode: "DeepSeek has launched an in-house AI chip project focused on inference workloads." (Source 1)
• Ars Technica: "The focus is on data center chips for inference, not training, and the goal is likely to reduce reliance on both Huawei and Nvidia." (Source 3)
• Semafor: "The move could reshape China’s domestic semiconductor race while also ramping up pressure on US chip giant Nvidia." (Source 4)
• The report also notes DeepSeek is seeking its first external funding of about $7 billion, which would support chip development.