DeepSeek, the Chinese startup that has rattled Silicon Valley with large language models rivaling those from OpenAI and Anthropic, is now planning to design its own chips. The move, reported by Reuters, is a direct response to US export controls that have cut off Chinese companies from Nvidia's most advanced semiconductors.
Why DeepSeek is building its own silicon
US export restrictions, tightened under the Biden and Trump administrations, bar Chinese firms from buying Nvidia's H100, H200, and B200 chips — the workhorses of modern AI training. DeepSeek, like other Chinese AI players, has been forced to rely on less powerful alternatives from Huawei and older Nvidia models. Founder Liang Wenfeng acknowledged the challenge in a rare 2024 interview, calling export controls a "constraint" on the company's ambitions.
A year in the making: DeepSeek's chip timeline
According to three people familiar with the matter, DeepSeek has been working on the silicon project for about a year. The startup has been meeting with potential partners in the hardware and semiconductor space and has been actively hiring engineers with chip design expertise. The project is still in early stages, and no timeline for a finished product has been disclosed.
What this means for China's AI ambitions
DeepSeek's chip push is part of a broader Chinese strategy to build domestic AI infrastructure independent of US technology. Beijing has been pressing its technology champions — including Huawei, Alibaba, and Baidu — to develop homegrown alternatives. If DeepSeek succeeds, it could reduce China's reliance on Nvidia and Huawei, giving the country more control over its AI supply chain. But chip design is notoriously difficult and capital-intensive, with even well-funded US startups struggling to compete with Nvidia's CUDA ecosystem and manufacturing scale.
How DeepSeek compares to OpenAI and others
DeepSeek is not alone in this strategy. OpenAI has also explored custom chip development, and companies like Google, Amazon, and Microsoft have designed their own AI accelerators. However, DeepSeek faces unique challenges: it must navigate US export controls, build a chip design team from scratch, and find a foundry willing to manufacture its chips — a problem compounded by Taiwan's TSMC being the only reliable advanced node manufacturer.
Confirmed facts vs what remains unclear
Confirmed: DeepSeek has been working on chip development for about a year. It has held meetings with hardware partners and is hiring engineers. Founder Liang Wenfeng has publicly stated that export controls are a challenge. Unclear: The chip's architecture, performance targets, manufacturing partner, and timeline. It is also unclear whether DeepSeek will design a general-purpose AI chip or a specialized accelerator. All speculation about specific technical capabilities or launch dates should be treated as unverified.
Risks and balanced view
Chip design is a high-risk, capital-intensive endeavor. Even established companies like Intel and AMD struggle to compete with Nvidia. DeepSeek's lack of experience in hardware, combined with the difficulty of accessing advanced manufacturing nodes, raises questions about feasibility. Critics argue that the startup may be overreaching, diverting resources from its core AI model development. Supporters counter that vertical integration is necessary for long-term survival under US export controls.
Wider trend: The US-China AI chip decoupling
DeepSeek's move is part of a larger decoupling of the US and Chinese AI ecosystems. US export controls have accelerated Chinese efforts to build domestic chip supply chains, from design tools to manufacturing. Companies like Huawei have already developed competitive AI chips, and startups like Biren Technology and Cambricon are emerging. However, the gap in manufacturing capability — particularly at the 3nm and 5nm nodes — remains a significant bottleneck.
What this means for investors and the AI industry
For investors, DeepSeek's chip plan signals that the Chinese AI sector is moving toward vertical integration, which could reduce dependence on US suppliers over the long term. For the global AI industry, it means a more fragmented supply chain, with potential for increased competition and innovation — but also higher costs and slower progress due to duplication of effort.
Future outlook
If DeepSeek successfully develops its own chip, it could become a significant player in China's AI hardware ecosystem. However, the path is fraught with technical, financial, and geopolitical obstacles. The most likely near-term outcome is that DeepSeek will produce a specialized chip for inference rather than training, leveraging its software expertise to optimize performance. A full-scale training chip competitive with Nvidia's latest offerings remains years away, if achievable at all.
Our Take
DeepSeek's chip plan is a logical response to an existential threat. US export controls have forced Chinese AI companies to either accept technological inferiority or build their own solutions. DeepSeek's decision to pursue chip design is ambitious, but it reflects a deeper reality: the US-China AI race is no longer just about algorithms — it's about hardware, manufacturing, and geopolitical leverage. Whether DeepSeek succeeds or fails, its move underscores the growing fragmentation of the global AI supply chain and the high stakes of the technology rivalry between the world's two largest economies.
Frequently Asked Questions
Why is DeepSeek making its own chips?
DeepSeek is developing its own AI chips to bypass US export controls that prevent Chinese companies from buying Nvidia's most advanced semiconductors. The startup aims to reduce dependence on US and Chinese suppliers like Nvidia and Huawei.
How long has DeepSeek been working on chip development?
According to sources familiar with the matter, DeepSeek has been working on the silicon project for about a year. The company has been meeting with hardware partners and hiring chip design engineers.
Will DeepSeek's chip be competitive with Nvidia?
It is too early to say. DeepSeek's chip project is still in early stages, and no technical specifications or performance targets have been disclosed. Competing with Nvidia's CUDA ecosystem and manufacturing scale is extremely difficult.
What does this mean for the US-China AI race?
DeepSeek's chip plan is part of a broader Chinese push for technological self-reliance. If successful, it could reduce China's dependence on US chip suppliers and accelerate the decoupling of the two countries' AI ecosystems.
