Editorial illustration for DeepSeek Plans Own Chips Amid US Export Controls, Following Huawei, Alibaba
DeepSeek Building Own Chips to Bypass US Export Ban
DeepSeek Plans Own Chips Amid US Export Controls, Following Huawei, Alibaba
DeepSeek wants to build its own chips. Reuters reported on July 7, citing three people familiar with the matter, that the Chinese AI startup has spent roughly a year quietly working toward entering the semiconductor business, meeting with hardware and silicon partners while hiring engineers for the effort.
The move puts DeepSeek in company with Huawei and Alibaba, both of which have pushed into chip design as Beijing works to reduce dependence on foreign silicon. For DeepSeek, the calculation is more specific: US export controls have kept Nvidia's chips largely out of reach in China, and Huawei's alternatives haven't closed the gap. Rather than wait on either, DeepSeek appears to be betting it can design around the bottleneck itself.
The chips in question are reportedly aimed at inference workloads in data centers, not training the massive models that made DeepSeek's name last year. That distinction matters. Inference is where AI companies burn compute at scale once a model is deployed, and where supply constraints could pinch hardest as DeepSeek and its rivals push their services to more users.
For example, OpenAI and Broadcom jointly announced Jalapeño , the former's first chip designed for inference at scale, just a couple of weeks ago.
Why this matters
DeepSeek building its own chips is a sign of how deep the US export controls have cut, not a side project. When a company that shook up the LLM market with efficient training methods decides it needs to design silicon too, that tells you the Nvidia workaround well is running dry. Huawei's grip on roughly half of China's data center chip market, plus Alibaba and Baidu chasing their own designs, means DeepSeek is entering a crowded, state-encouraged race rather than blazing a lone trail.
For developers and founders outside China, the near-term effect is probably invisible: model releases keep coming. But the medium-term picture is a bifurcating hardware stack, Chinese AI companies optimizing for domestic silicon with different constraints than CUDA-tuned workflows assume. Researchers benchmarking DeepSeek's models should watch whether performance or release cadence shifts as it reallocates engineering talent toward chip design.
And anyone betting on export controls slowing Chinese AI progress should note that pressure is now producing vertical integration, not retreat. That's a different outcome than the policy intended.
Related Reading
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
From Archives Trump’s Davos drama, AI‑fuelled midterms draw tens of millions early
From Archives Dynamic Power Boost Compensates for GPU Failures in LLM Training
Common Questions Answered
Why is DeepSeek planning to develop its own chips despite US export controls?
DeepSeek is building its own chips because US export controls have severely limited access to foreign semiconductor technology, making domestic chip design a strategic necessity. The Chinese AI startup has spent roughly a year working with hardware and silicon partners while hiring engineers, indicating this is a core business priority rather than a side project.
How does DeepSeek's chip development strategy compare to other Chinese tech companies?
DeepSeek is following the path of major Chinese tech companies like Huawei and Alibaba, which have also pushed into chip design to reduce dependence on foreign silicon. Huawei currently controls roughly half of China's data center chip market, while Alibaba and Baidu are also pursuing their own chip designs, making DeepSeek's entry into a crowded, state-encouraged race.
What does DeepSeek's chip development reveal about the effectiveness of US export controls?
DeepSeek's decision to build its own chips demonstrates how deeply US export controls have impacted Chinese AI companies, signaling that the "Nvidia workaround well is running dry." When a company that disrupted the LLM market with efficient training methods determines it must design silicon, it indicates that foreign semiconductor access has become critically constrained.
How does DeepSeek's chip strategy differ from OpenAI's approach?
While OpenAI partnered with Broadcom to announce Jalapeño, its first chip designed for inference at scale, DeepSeek is developing chips independently as a response to US export controls. DeepSeek's approach reflects the geopolitical constraints facing Chinese companies, whereas OpenAI operates without such restrictions and can collaborate with established semiconductor manufacturers.
Further Reading
- In AI chip trade war with China, there's one big mistake US can't make - CNBC
- DeepSeek shows the limits of US export controls on AI chips - Brookings Institution
- DeepSeek, Huawei, Export Controls, and the Future of the US-China AI Race - CSIS
- On DeepSeek and Export Controls - Dario Amodei
- House Select Committee Publishes Report on DeepSeek, as ... - Mintz
