This is not investment advice. The author has no position in any stocks mentioned. Wccftech.com has a Disclosure and Ethics Policy.
Chinese artificial intelligence research institute DeepSeek’s artificial intelligence model R1 will take the industry by storm this month, giving it access to tens of thousands of NVIDIA’s GPUs for training, the AI company’s CEO believes. DeepSeek R1 is one of the most advanced AI models in the world and ranks alongside leading US platforms such as OpenAI’s o1 and Meta’s Llama.
NVIDIA’s Hopper chip is the company’s next-generation GPU and the most widely used in the world as the company ramps up shipments of its next-generation Blackwell chips. The latest details were shared by Scale AI founder and CEO Alexandr Wang in an interview with CNBC earlier this week, in which Wang said R1 is the company’s most difficult AI test. It added that it matched or beat all the highest-performing AI models.
AI CEO says DeepSeek R1 is equal to or better than OpenAI’s o1
Wang opened an interview with CNBC’s Andrew Ross Sorkin by sharing details about his company’s latest AI tests. Dubbed “mankind’s last test,” the test will feature “the most difficult questions” related to the latest research from “professors of mathematics, physics, biology, and chemistry.” After testing all the latest AI models in this test, Wang’s team found that DeepSeek’s latest model “actually performs the best, or is almost on par with o1, the best model in the US.” did. . ..”
When asked about the U.S.-China AI race, Wang said, “It’s been true for a long time that the U.S. is ahead.” However, he added that DeepSeek’s latest model attempts to change the calculation. Wang said he thinks it’s symbolic that a Chinese research institute would release an earth-shattering model on Christmas Day, when we’re all celebrating the holiday.
The conversation then turned to the amount of GPU that DeepSeek and Chinese AI players might put into their AI models. NVIDIA’s GPUs are among the most popular products in the world, and due to their potential, their export is also regulated by the US government. The company’s current generation Hopper GPU lineup consists of H100 and H200 GPUs.
The main differences between the two GPU models are memory clock speed and capacity. The Biden administration had restricted sales of these GPUs to China in 2022 with rules that prohibited NVIDIA from selling chips that were significantly better than the older A100 GPUs. These restrictions forced NVIDIA to develop alternative chips called H800 and A800, which were also banned a year later in October 2023.
As for whether the Chinese have access to NVIDIA’s advanced GPUs, Wang said, “The reality is both yes and no. Chinese labs, as you know, have more H100 than people think. “We understand that DeepSeek has approximately 50,000 H100s,” he added. “I obviously can’t talk about it because it goes against the export restrictions introduced by the United States,” Wang said. He also thinks that “they have more chips than other people expect.”
However, the Chinese may have difficulty procuring additional chips. “But in the future, it will also be limited by the chip regulations and export controls that we have in place,” Wang said.
