Report: China's DeepSeek follows OpenAI in developing its own custom inference chips

What happened

Chinese AI startup DeepSeek is developing its own in-house AI chip targeting inference workloads, according to a Reuters report from July 7, 2026. The company has contacted chip design, foundry, and memory partners and is actively recruiting experienced chip engineers.

Context and impact

DeepSeek, which has relied on Nvidia and Huawei hardware, is following the playbook of OpenAI (Jalapeño chip with Broadcom) and Anthropic (Samsung talks). Roughly 70% of AI compute demand now comes from inference rather than training, making a custom inference chip critical for model commercialization. U.S. export restrictions on Nvidia chips to China provide further motivation.

Details

• Source: Reuters (July 7, 2026)
• Project stage: early (~1 year of exploration, active hiring)
• Focus: inference (not training)
• Current suppliers: Nvidia and Huawei
• Context: OpenAI (Jalapeño), Anthropic (Samsung), Google (TPU) all have custom silicon
• ~70% of AI compute demand = inference