China's DeepSeek's New Bid to Take On OpenAI Involves Custom AI Chips

Chinese AI startup DeepSeek is making its boldest move yet in the global artificial intelligence race. After shaking up the industry with its low-cost AI models, the company is now reportedly developing its first custom AI chip, signalling a shift from being just an AI model developer to becoming a full-stack AI company like OpenAI and Google.

According to a Reuters report citing people familiar with the matter, the chip is being designed specifically for AI inference, the stage where trained models generate responses to user prompts, rather than for training new AI models. If successful, the move could significantly reduce DeepSeek's dependence on Nvidia and even China's own Huawei, both of whose chips currently power its AI services.

Why DeepSeek Is Building Its Own AI Chip

The reported chip project represents a major strategic shift for DeepSeek. Until now, the Hangzhou-based startup has largely focused on building powerful AI models, including the viral R1 reasoning model and the newer V4 series. Hardware development marks an expansion into a space traditionally dominated by companies such as Nvidia, AMD, Broadcom and Huawei.

Sources told Reuters that the project began about a year ago and remains in its early stages. DeepSeek has reportedly been holding discussions with chip designers, semiconductor foundries and memory suppliers, while quietly hiring chip-design engineers without advertising public job openings.

Taking a Page Out of OpenAI's Playbook

DeepSeek isn't the only AI company trying to reduce its dependence on Nvidia.

Last month, OpenAI unveiled Jalapeño, its first custom inference chip developed in partnership with Broadcom. Anthropic has also been exploring in-house AI chips as companies seek greater control over performance, costs and long-term supply chains.

Inference has rapidly become the fastest-growing part of AI computing as millions of users interact with chatbots and AI assistants every day. Unlike training chips, inference processors are optimised to deliver responses efficiently while consuming less power.

For DeepSeek, building a dedicated inference chip could lower operating costs while allowing tighter integration between its hardware and AI models.

More Than Just a Business Decision

DeepSeek's hardware ambitions are also shaped by geopolitics. US export controls prevent Chinese companies from buying Nvidia's most advanced AI chips. Although DeepSeek initially trained its R1 model using Nvidia's H800 processors, those chips were later banned from export to China. Since then, the company has increasingly relied on Huawei's Ascend processors for newer models.

However, Huawei is no longer China's only domestic AI chip contender. Alibaba and Baidu are also developing their own processors, making the country's AI hardware market increasingly competitive. Reuters notes that DeepSeek's entry could further weaken Huawei's dominance in China's estimated $50 billion AI chip market.

Building Chips Won't Be Easy

Designing an AI chip is considerably more difficult than building an AI model.

Even if DeepSeek successfully completes the design, manufacturing presents another challenge. US restrictions continue to limit Chinese companies' access to the world's most advanced chip fabrication technologies and high-bandwidth memory, both of which are essential for competitive AI processors.

Industry analysts also note that any DeepSeek chip would likely remain focused on the domestic Chinese market unless the company gains access to leading-edge manufacturing capabilities.

A Bigger Shift in the AI Industry

DeepSeek's reported chip effort reflects a broader trend across the AI industry. Instead of relying entirely on Nvidia, leading AI companies are increasingly designing custom silicon tailored to their own models and workloads.

If DeepSeek succeeds, it won't just be competing with OpenAI on AI models. It will also be joining the race to control the hardware that powers the next generation of artificial intelligence, an increasingly important battleground as AI inference becomes the industry's biggest growth driver.