DeepSeek develops in-house AI chip to reduce dependence on Nvidia, Huawei

DeepSeek gained international attention last year after launching two highly efficient AI models that surprised Silicon Valley and Washington with their performance.

Chinese artificial intelligence (AI) startup DeepSeek is developing its own AI chip to reduce its reliance on Nvidia and Huawei processors, Reuters reports.The chip is being designed for inference—the stage of AI computing where trained models generate responses to users—rather than for training AI models, the sources said.If successful, the move would mark a significant strategic shift for DeepSeek, which has so far been recognised for developing highly efficient AI models rather than semiconductor hardware.

The development could also intensify competition in China’s AI chip market, where Huawei has emerged as a leading supplier following US export restrictions on Nvidia’s most advanced chips to China.DeepSeek gained international attention last year after launching two highly efficient AI models that surprised Silicon Valley and Washington with their performance.

Despite its growing prominence, the company has largely focused on advancing AI models instead of commercialising its technology.According to the reports, Huawei currently commands about half of China’s estimated $50 billion AI chip market, benefiting from US export controls.

However, rivals including Alibaba and Baidu have also begun developing their own AI chips, gradually increasing competition in the domestic market.DeepSeek’s chip project is still in its early stages.

The company has been holding discussions with chip-design companies, foundries and memory suppliers, while also increasing the hiring of chip-design engineers in recent months, the sources said.

The recruitment has been conducted privately, without public job postings.DeepSeek did not respond to Reuters’ request for comment.The company’s move mirrors a broader trend among leading AI developers seeking greater control over the hardware powering their models.For DeepSeek, the effort also reflects China’s broader push to develop domestic AI infrastructure as US export controls continue to restrict access to advanced semiconductors.DeepSeek founder Liang Wenfeng had said in a 2024 interview with a Chinese media outlet that chip export controls posed a significant challenge for the company.The startup has previously relied on both Nvidia and Huawei hardware.

It has said its R1 reasoning model was trained using Nvidia’s H800 chips, which were designed for the Chinese market before being barred from export in late 2023.

More recently, DeepSeek has increasingly adopted Huawei’s Ascend processors, releasing its V4 model adapted for the platform, while Huawei has said its chips were used for part of the training of V4-Flash, a lighter version of the model.An in-house inference chip would allow DeepSeek to tap into one of the fastest-growing segments of the AI computing market as demand increasingly shifts from training models to deploying AI applications.

Inference chips are typically more power-efficient and less expensive than general-purpose graphics processors.However, designing a competitive AI chip remains a complex undertaking requiring significant investment and years of development.

US restrictions continue to limit Chinese companies’ access to advanced overseas manufacturing facilities and high-bandwidth memory technologies critical for AI chips.The chip development effort also coincides with DeepSeek’s decision to seek outside funding.

In June, the company was preparing to raise $7 billion in its first funding round at a valuation of between $52 billion and $59 billion, marking a departure from its long-standing strategy of avoiding external investment.