DEEPSEEK DEVELOPS ITS OWN AI CHIPS

DeepSeek, known for developing cost-effective AI models, is now pursuing the same efficiency goal in chip design. The company is in early stages of developing an inference chip—hardware optimized for running trained AI models rather than training them. Three sources familiar with the project confirmed the initiative to Reuters. The chip development targets DeepSeek's dependency on external vendors, particularly Nvidia and Huawei, for its computational infrastructure. The effort reflects a broader industry trend. As AI compute demands grow, major AI developers increasingly design proprietary silicon to optimize performance and reduce costs. Google, Meta, and others have pursued similar strategies with custom chips. For DeepSeek, custom silicon could enhance its competitive position. The startup has gained attention for deploying capable AI models at a fraction of typical development costs. Inference chips—which handle queries from deployed models—represent a logical extension of this cost-optimization strategy. China's chip sector faces significant constraints. U.S. export controls limit access to advanced semiconductors needed for AI development. Creating domestic alternatives addresses both technical needs and geopolitical vulnerabilities. DeepSeek's inference focus suggests a phased approach. Training chips require more processing power and face greater technical hurdles. Inference chips, while still complex, present a more achievable near-term target. No timeline for the chip's release or deployment was disclosed. The project remains in early development stages.