Nvidia Corp., the biggest provider of chips used to train artificial intelligence software, said a new model released by Chinese startup DeepSeek is an “excellent AI advancement” that complies with US technology export controls.
“DeepSeek’s work illustrates how new models can be created” using the Test Time Scaling technique, the company said in an emailed statement, “leveraging widely available models and compute that is fully export control compliant.”
Nvidia’s statement indicates that it believes that the Chinese company didn’t violate US restrictions that limit access to advanced US chips in the creation of its technology. Most of Nvidia’s best products can’t be sold to Chinese companies without licenses from Washington — permits that are not typically granted.
The statement also appears to dismiss some analysts’ and experts’ suspicions that the Chinese startup couldn’t have made the breakthrough it has claimed.
After more than quintupling in the past two years, Nvidia’s shares plummeted Monday in one of the biggest selloffs in history, on concern that DeepSeek’s offering reduces the need for the expensive type of hardware that the US chip company offers. Nvidia’s statement, though, points out that inference, the work of running AI models, nonetheless requires a lot of its products.
“Inference requires significant numbers of Nvidia GPUs and high-performance networking,” the company said.
More stories like this are available on bloomberg.com