Nvidia CEO praises China’s open-source models as ‘very advanced’

Open source was “extremely powerful” for AI innovation and Chinese firms offer the “best open reasoning models”, Nvidia co-founder and CEO Jensen Huang said in Beijing on Thursday, ahead of the expected resumption of the company’s H20 chip sales to China.

Open source allows not just the contribution of each company but the combined resource of an ecosystem, which is what was “very clever about open source engineering here in China”, Huang said at the China International Supply Chain Expo.

DeepSeek’s models, Alibaba Group Holding’s Qwen and Chinese start-up Moonshot’s Kimi were “the best open reasoning models in the world today,” and were “very advanced”, Huang said during a fireside chat with Wang Jian, founder of Alibaba Cloud, the Chinese e-commerce giant’s AI and cloud unit.

Alibaba owns the South China Morning Post.

While Huang praised the “open source” movement in China, he also highlighted the global implications of the approach whereby companies share their research and collaborate to advance artificial intelligence technology.

“Don’t forget that open source has many global implications. Not only did the open source models help the Chinese ecosystem, [they are] helping ecosystems around the world,” Huang said.

Huang added that different kinds of companies - in areas such as fintech, healthcare and robotics - would be able to take advantage of open models.

Huang is on his third trip to China this year, attending the expo after Nvidia announced earlier this week that it had received Washington’s approval to resume shipping the H20, an AI graphics processing unit (GPU) tailor made for the Chinese market to comply with US export restrictions.

Despite being less powerful than other advanced Nvidia chips such as the H100, it remains the top option for Chinese firms and was in high demand before the US restricted its sales in April.

The recovery of a supply chain takes time, and it currently takes Nvidia about nine months from the placement of wafer orders to the delivery of finished computing products, Huang told Chinese media outlets on Wednesday.

“We are working at full speed to restore the production capacity for Hopper architecture products, but this indeed requires sustained efforts,” Huang reportedly said.

Chinese companies have not yet been able to place orders for the H20 because US government agencies were still processing the approvals, according to Huang. Nvidia said earlier this week that the US government had “assured” the company that licences would be granted to allow export of the H20 GPUs to China.

The availability of Nvidia’s H20 would “ease the anxiety” of domestic firms that were building large language models, and boost the share of foreign chip suppliers in China, analysts said. - South China Morning Post