DeepSeek Builds Its Own AI Chip, Pressuring Nvidia
Chinese startup DeepSeek is developing its own chip to run AI models, aiming to cut its reliance on Nvidia amid US export curbs. Nvidia shares fell about 1.6 percent.
- Reuters reported on July 7 that China's DeepSeek is developing its own AI chip, citing sources familiar with the plans.
- The chip is meant for inference, running already-trained models, not training new ones.
- The goal is to reduce reliance on Nvidia amid US curbs on exports of advanced chips.
- Nvidia shares fell about 1.6 percent in pre-market trading after the report.
Chinese artificial intelligence startup DeepSeek is developing its own chip to run AI models, Reuters reported on July 7, 2026, citing sources familiar with the company’s plans. The project is said to be at an early stage, but its purpose is clear: to reduce dependence on processors made by the US company Nvidia, which are increasingly hard for Chinese firms to obtain because of Washington’s export restrictions.
The report carried an immediate financial signal as well. Nvidia shares fell about 1.6 percent in pre-market trading after the news broke, a reflection of market nerves rather than any measurable damage. Nvidia still holds a dominant share of the global supply of AI chips, and a single startup in the early phase of hardware development does not change that position overnight.
Why does an inference chip differ from a training chip?
One detail easily lost in the first reactions is what the chip is actually for. According to the available information, DeepSeek is targeting inference, meaning it runs already-trained models when they respond to user queries. That is a demanding task, but a noticeably lighter one than training new models from scratch, where the demands on computing power and memory are far greater.
The distinction matters for gauging the real scale of the ambition. Building a chip that efficiently serves a finished model is a more achievable step than building one capable of training next-generation models, a job for which Nvidia’s most advanced processors remain the reference standard. In other words, DeepSeek is for now choosing ground where the odds of a result are better, rather than trying to match Nvidia head-on in its strongest discipline.
The context is also geopolitical. The United States has steadily tightened its rules on exporting advanced semiconductors to China in recent years, pushing Chinese technology companies to secure their own supply chains. An in-house inference chip is therefore as much a strategic move for DeepSeek as a technical one: less reliance on foreign suppliers also means less exposure to future shifts in US policy.
What this step does not yet prove
It is worth being precise about the limits of this news. The fact that a company is working on a chip does not mean the chip exists, that it works as intended, or that it will be manufactured in serious volumes any time soon. Producing advanced semiconductors depends on access to modern fabrication plants and equipment, and that is precisely the part of the chain made harder for Chinese firms by sanctions.
There is also an indirect indicator of how seriously the plan is being pursued. According to the sources cited by Reuters, DeepSeek is ramping up its hiring of chip-design engineers, and doing so without public job ads. Quietly recruiting hardware specialists suggests the company is committing resources, but staffing is not the same as a finished, tested and shipped product.
For now, the fairest description is that this is a symbolically important but only just-begun attempt to break Nvidia’s monopoly in one segment of the market. The coming months will show whether a chip that actually works stands behind the announcement. Do you think Chinese companies can build a competitive AI chip despite the US restrictions?
