Reuters broke on Monday that DeepSeek, the Hangzhou lab that spent 2025 accidentally clearing the Nasdaq every quarter, is working on its own AI chip. Early stage. Inference only, not training. The company has been quietly hiring chip-design engineers over the last several months, without public job postings, and is in talks with foundries, memory suppliers, and chip-design partners. No named partner. No process node. No timeline.
The strategic point is straightforward. DeepSeek currently runs on some combination of Nvidia hardware (whatever it can source through the export controls) and Huawei’s Ascend line (which is what the Chinese state would prefer everyone use). It wants to reduce reliance on both. Nvidia because of the geopolitical squeeze. Huawei because Huawei is a competitor first and a supplier second, and a lab that also runs its own model prefers not to owe its silicon to a lab that also runs its own model.
This is now the pattern. OpenAI has Jalapeno. Anthropic is publicly working on its own accelerator plans and has the Amazon Trainium partnership as a bridge. Meta, Google, and Microsoft all shipped internal accelerators years ago. Alibaba and Baidu are doing the same thing DeepSeek is doing, one province over. Huawei holds roughly half of China’s $50 billion AI chip market by itself. The lab-designs-its-own-chip announcement has become a rite of passage, somewhere between raising a Series C and announcing an office in Singapore.
The honest read comes from Richard Windsor at Radio Free Mobile, who told Reuters that DeepSeek “has almost no chance of selling silicon outside of China unless it gets access to leading-edge manufacturing.” Translation: this chip, if it ever ships, is a captive-use inference part for DeepSeek’s own workloads, not a Nvidia competitor. Which is fine. That is the actual bar. Not “beat Blackwell.” Just “run our own models cheaply enough on our own domestic supply chain that we do not have to sweat the next export-control update.” Nvidia investors reacted anyway. NVDA slipped about 1.5 percent in premarket the morning of the story. The market prices the vibe first and the shipping schedule later.