deepseek is joining top ai labs in developing its own ai chip

chinese startup @deepseek_ai is in the early stages of designing a custom chip for inference (not training). the goal – cut reliance on nvidia and huawei amid us export controls. it has been quietly hiring chip-design engineers and holding talks with foundries and memory suppliers

it is not alone. the pattern across ai-first labs:

• @OpenAI: unveiled "jalapeño," its first custom inference chip built with broadcom, on jun 24. still in testing

• @AnthropicAI: in early talks with samsung to build a custom chip. exploratory, no design locked yet

• @Meta: mtia already deployed at scale, with four new chip generations on the roadmap

• @Alibaba_Qwen: launched the zhenwu m890 via its t-head unit – training and inference on one chip, 560k+ units already shipped

• bytedance: developing its seedchip accelerator with tsmc, taping out this year

every major lab wants independence from nvidia, tighter control over inference economics, and, increasingly, a new revenue line in hardware itself. owning the chip is becoming as strategic as owning the model