- Nvidia CEO Jensen Huang warns of China’s big advantages in AI
- These include China’s rapid completion of data center construction and a robust energy infrastructure to meet AI’s power needs.
- Meanwhile, a new study shows that Chinese open source LLMs have driven nearly a third of global AI use.
Nvidia CEO Jensen Huang again warned of China’s rapid progress in artificial intelligence and the country’s advantage in development infrastructure.
Luck reports Late last month, Huang spoke with John Hamre, president of the Center for Strategic and International Studies (CSIS), noting, “If you want to build a data center here in the United States, it will probably take about three years from the start of construction to the deployment of an AI supercomputer. They (China) can build a hospital in a weekend.”
In other words, China is able to implement large construction projects at an incredibly fast speed and also has a huge advantage in energy infrastructure.
These are critical elements for AI development in quickly building massive data centers, managing computing needs, and providing the power to power everything.
Huang noted that China has “twice as much energy as we (US) as a nation and our economy is bigger than theirs” and that this “doesn’t make sense to me”, and that energy capacity growth in China is “increasing”, while in the US it remains more or less stable.
However, to offset the concerns raised, the CEO clarified that Nvidia is “generations ahead” of China when it comes to AI chip technology (this statement may be a bit biased), but Huang still said that this is no reason to rest on our laurels.
Huang has previously noted that China is “nanoseconds behind the US” in the AI race, but we’re told Nvidia’s CEO apparently remains optimistic about the Trump administration’s efforts to boost AI investment and domestic manufacturing jobs.
Attractiveness and rapid promotion.
Meanwhile, a separate article from This was reported by the South China Morning Post (SCMP). that nearly 30% of global AI use can now be attributed to Chinese open source models (LLM).
This figure comes from a report prepared by OpenRouter, an independent aggregator of AI models, in collaboration with venture capital firm Andreessen Horowitz. It is based on a study of 100 trillion tokens, the units of data processed by LLMs (or, more simply, the building blocks of how artificial intelligence works).
The bulk goes to Western closed source LLMs like ChatGPT, which owns the rest of the market (around 70%).
However, remember that Chinese open source LLMs accounted for just over 1% of tokens a year ago, so reaching 30% today is a steep growth trajectory to say the least.
Looking at open source LLMs alone, they tell us that Chinese models represent on average about 13% of weekly token usage, almost on par with the rest of the world’s 13.7%. (This is open source, remember: the remaining majority are closed, proprietary models like ChatGPT.)
Another interesting point is that China’s open source LLMs are now getting equal weight and it’s not just about DeepSeek (as it was originally). DeepSeek V3 is of course a major force in the use of artificial intelligence in China, but there are Alibaba’s Qwen and Moonshot AI’s Kimi K2 who are also major players.
The report says that messages in Chinese are now second only to English.
Overall, the growth of AI in China is a pretty steep climb and you can see where Huang’s concern comes from. Especially since it will be difficult to curb this growth in China in the short term, and the Nvidia CEO’s comments about the country’s energy infrastructure actually represent a significant advantage over the US – again, one that is unlikely to change anytime soon.
And as we’ve seen recently with the launch of DeepSeek’s new v3.2 models, China still has a lot to offer when it comes to lowering the cost of AI deployment. It seems serious competition for global AI dominance is imminent.