DeepSeek Is Building Its Own AI Chip: Is This a Real Threat to Nvidia?

Chinese AI startup DeepSeek is developing its own AI chip, according to a Reuters report citing three people familiar with the matter. The goal is to reduce its reliance on Nvidia and Huawei, the two chipmakers it currently depends on. The news added to the recent wobble in AI and chip stocks, but Nvidia itself fell only about 1.6%. In our view, that muted reaction is telling, and it is worth understanding why before assuming this is a serious blow to the AI chip king.

What DeepSeek Is Actually Building (and What It Isn’t)

The first thing to get right is what kind of chip this is. There are two main jobs in AI computing: training, which is teaching a model, and inference, which is running that trained model to answer users’ questions. DeepSeek’s chip is for inference only.

That distinction is important. Training is where Nvidia’s lead is strongest and hardest to challenge, so DeepSeek is sensibly going after the easier, faster-growing target. Inference is cheaper to design for and is where the day-to-day cost of running AI adds up. The catch is that this is still an early-stage project, reportedly about a year in the making, with no finished chip, no named factory, and no launch date.

Building a competitive chip takes years and billions of dollars. Worse for DeepSeek, US export controls block Chinese firms from using Taiwan’s TSMC for the most advanced chips, forcing them to rely on domestic foundries like SMIC, which face severe production bottlenecks and older technology. So this is a statement of intent, not a product on shelves, and we would treat it as a signal to watch, not a threat that has arrived.

Why Nvidia’s Shares Barely Moved

Here is the key point many headlines missed: Nvidia’s grip on China is already severely compromised by US export rules, but it is not gone. Nvidia still earns billions from China through export-compliant chips like the H20, though tighter curbs have caused its share of the advanced-accelerator market there to collapse, and recent quarters have seen those China sales fall sharply.

A Chinese company building a home-grown chip mostly takes share from domestic rivals like Huawei, but it also chips away at the multi-billion-dollar business Nvidia is fighting to hold onto in the region. In other words, the DeepSeek news is a threat to that shrinking China business, not to the roughly 90% of Nvidia’s revenue that comes from everywhere else.

There is also a bigger picture. DeepSeek is not alone: OpenAI recently unveiled its first custom inference chip, named Jalapeño, developed with Broadcom, and other major AI players are exploring the same path. This tells us something important for investors: custom chips are becoming normal across the industry, but designing a chip is far harder than designing an AI model, and Nvidia’s advantages in software and scale remain formidable.

The Investor’s Takeaway for Nvidia

Our take: this is a long-term risk worth watching, not a reason to panic. The real story is not one Chinese chip but the slow, steady push by big AI customers to build their own silicon and depend less on Nvidia. Over many years, that trend could chip away at Nvidia’s dominance.

For now, though, Nvidia’s position looks secure. Its China business has already been hit hard by export controls, so DeepSeek’s chip changes little in the near term, and its lead in training and software is intact. That strength is a big part of why Nvidia has led the AI rally this year. For long-term Nvidia investors, the sensible response is to monitor how many big customers move toward their own chips, rather than react to a single early-stage project. The threat is real, but it is measured in years, not days.