Chinese AI company DeepSeek is developing its own custom chip for AI inference, according to a Reuters report, as the company seeks to reduce its dependence on Nvidia and Huawei amid tightening US export controls. The Hangzhou-based firm, known for its efficient AI models, is making an unusual pivot from software into semiconductor design.
What's Happening
DeepSeek has been working on the chip project for about a year, Reuters reported, citing people familiar with the matter. The company is hiring engineers and holding discussions with external partners covering chip design, manufacturing, and memory. The focus is on inference chips—the specialized processors used when an AI model generates answers after being trained—rather than the training chips that dominated the early AI boom.
Inference is becoming a larger share of AI computing as companies shift from building models to running them in real-world applications, where cost and power efficiency matter more. This trend is also benefiting other firms, as noted in our coverage of Kioxia riding the AI shift to inference.
Why DeepSeek Needs Its Own Chip
The move is driven partly by supply constraints. DeepSeek previously used Nvidia's H800 chip, a version designed for the Chinese market that was later banned by Washington. The company has increasingly optimized its models, including the V4, for Huawei's Ascend chips, according to Reuters. But relying on a single supplier—especially one facing its own export restrictions—carries risk.
US export controls have narrowed China's access to advanced semiconductors and manufacturing equipment, making it harder for Chinese AI firms to buy cutting-edge chips from Nvidia or have them fabricated at leading foundries like TSMC. By designing its own chip, DeepSeek hopes to gain more control over its hardware stack and reduce vulnerability to geopolitical shifts.
If successful, DeepSeek would join a growing list of AI labs that tailor hardware to their own workloads. But the path is fraught with challenges: long development timelines, complex supply chains, and restrictions on access to advanced foundries and high-bandwidth memory, which is critical for AI performance.
What It Means for Investors
DeepSeek's chip ambitions help explain why Nvidia's stock slipped 2% in premarket trading on the news. Nvidia and Huawei don't just compete with each other in China; they face a broader trend of large AI developers trying to own more of the technology stack, from models to the chips that run them.
Inference-specific chips can trade some of a GPU's flexibility for better efficiency on a narrower set of tasks. That can lower costs and reduce dependence on outside suppliers. Reuters noted that Huawei currently holds roughly half of China's $50 billion domestic AI chip market, but that grip is weakening as Alibaba and Baidu work on their own designs. If DeepSeek joins that push, pricing power and market share could shift away from standalone chip vendors toward whoever controls the end-to-end inference platform.
For everyday investors, this story underscores a key theme: the AI hardware market is becoming more fragmented and competitive. Companies that rely on selling general-purpose chips may face pressure as their largest customers develop in-house alternatives. At the same time, the growing importance of inference could create opportunities for specialized chip designers and memory makers.
The broader market backdrop also matters. US export controls continue to reshape the global semiconductor landscape, and any major shift in China's AI chip market could have ripple effects for Nvidia, AMD, and other chip stocks. Investors should watch for further developments in DeepSeek's chip project and how Nvidia and Huawei respond to the growing trend of vertical integration among AI developers.
