DeepSeek Plans Custom Inference Chips to Reduce Dependence on Nvidia and Huawei
Summary
- • DeepSeek is developing its own data center inference chips, a project active for roughly one year.
- • Move aims to cut reliance on both Nvidia (blocked by US export controls) and Huawei (~50% of China's AI chip market).
- • DeepSeek is hiring chip engineers and meeting hardware partners; focus is inference, not training.
- • OpenAI (Jalapeño chip) and Anthropic are pursuing parallel custom-silicon strategies for similar reasons.
Details
DeepSeek silicon program active ~1 year
Working on chip move for approximately one year; hiring engineers and meeting hardware/silicon partners, per Reuters citing 3 sources familiar with the matter
Focus on inference chips, not training
DeepSeek's target is data center inference chips — where operational cost and latency are highest in deployed AI services — not training accelerators
Huawei controls ~50% of China's AI chip market
US export controls have blocked Nvidia from China; Huawei filled the void with ~50% data center chip market share — DeepSeek wants to escape that single-vendor dependency
Alibaba and Baidu also developing custom AI chips
DeepSeek is not alone among Chinese tech companies; Alibaba and Baidu have been making parallel moves into custom silicon, signaling a broad industry push
OpenAI launched 'Jalapeño' inference chip with Broadcom
OpenAI and Broadcom jointly announced the Jalapeño chip for inference at scale just weeks ago — showing US and Chinese AI labs converging on the same vertical integration strategy
Anthropic also exploring custom chip design
No public milestones yet, but Anthropic's parallel exploration confirms custom silicon is now a standard strategic priority across frontier AI labs
Data center compute remains constrained globally
Custom silicon gives AI labs a structural advantage as GPU supply stays tight; controlling silicon and data center layers is increasingly seen as a moat in an era of scaled AI deployment
Tracks DeepSeek's entry into chip design: program scope, strategic rationale, geopolitical context (US export controls, Huawei), and parallel moves by Chinese tech giants and US AI labs.
What This Means
DeepSeek's move into chip design marks a significant escalation in AI's semiconductor supply chain tensions. Blocked from Nvidia's best hardware by US export controls and unwilling to remain dependent on Huawei, DeepSeek is now pursuing the same vertical integration strategy as OpenAI and Anthropic — only with the added urgency of geopolitical cut-off. If successful, it would give the lab direct control over its inference infrastructure, reducing both cost and supply-chain vulnerability. The convergence of US and Chinese AI labs all independently building custom silicon signals that the era of universal Nvidia dependence is ending, with profound implications for Nvidia's dominant market position and for the global AI compute landscape.