Nvidia shares slip after report says DeepSeek is working on its own AI chip

Nvidia shares edged lower in premarket trading Tuesday after a Reuters report said Chinese AI startup DeepSeek is designing its own in-house artificial intelligence chip. The move, if it advances, could reduce how much computing Nvidia and other established chip vendors are needed to run DeepSeek’s systems, at least for certain stages of AI workloads.

Preliminary market reaction was immediate. Nvidia stock was indicated down about 1.6% in premarket trading following the report, according to the market coverage that circulated Monday into Tuesday.

Reuters reported, citing people familiar with the matter, that DeepSeek’s chip effort is aimed at inference. Inference is the part of AI computing where a trained model generates responses for users, as opposed to training new models from scratch.

Other coverage of the same Reuters account said the company has increased hiring for chip-design talent in recent months, framing the project as a strategic shift beyond DeepSeek’s earlier reputation for focusing on AI model performance rather than chip commercialization.

The potential implication for Nvidia is straightforward, but hard to quantify with public information. Nvidia’s data center GPUs are the dominant compute platform for training and inference across much of the AI industry. If DeepSeek can run some workloads on a purpose-built accelerator, it could alter procurement patterns over time, especially if Chinese firms face tighter hardware supply constraints or seek more control over performance and cost.

The competitive backdrop in China matters for how investors interpret the report. The Business Standard, summarizing the Reuters-based claims, noted that US export restrictions have helped Huawei gain a meaningful share of China’s domestic AI chip market, and that other Chinese tech companies such as Alibaba and Baidu are also pursuing their own AI chips. In that context, DeepSeek’s push would place another major model developer into the semiconductor arms race.

Still, crucial details were not disclosed in the public reporting. It remains unclear how far along DeepSeek is in development, whether the chip is designed to support training, how it compares with Nvidia’s most advanced accelerators on performance and efficiency, and how quickly it could move from prototypes to production deployment.

For Nvidia, the key watch item is whether the report leads to concrete changes in customer behavior. Market participants will likely focus on any subsequent disclosures from DeepSeek about pilot deployments, manufacturing plans, or compatibility with existing AI software stacks, since those steps determine whether an in-house accelerator meaningfully substitutes for external hardware.