Nvidia CEO Jensen Huang credits DeepSeek with accelerating open-source AI shift

Speaking at the annual trade show, Huang said DeepSeek’s models – released earlier last year – had accelerated the growth of the open-source ecosystem, even after the company’s R1 model, which required fewer computing resources to train, helped trigger a short-lived sell-off in Nvidia shares.

“We saw the advance of DeepSeek R1, the first open model that’s a reasoning system,” Huang said, summarising key developments in AI in 2025. The Chinese model, he said, had “caught the world by surprise” and was helping to revolutionise AI and catalyse global innovation.

“Really, really exciting work – we’re so happy with it,” he added.

Huang said open-source models were rapidly closing the performance gap with the world’s leading “frontier” models. One slide in his presentation, which tracked the narrowing differences in capability, listed three China-developed models: DeepSeek-V3.2, Kimi K2 and Qwen.

Kimi K2 was developed by Beijing-based start-up Moonshot, while Qwen is the open-source model family developed by Alibaba Group Holding, the owner of the Post.

Despite the momentum behind open models, Nvidia has defended its position as a crucial hardware supplier for the AI boom, even as Chinese developers have demonstrated they can train high-performing models with limited chip access.