- Nvidia CEO Jensen Huang urges caution over China compute capacity
- China has "enormous" amounts of compute, including lots going unused
- Huang says China training a Mythos-esque model could cause issues
Nvidia CEO Jensen Huang has warned of the potential threat China poses in terms of harnessing huge amounts of compute power to train its next generation of AI models.
Speaking on the Dwarkesh Patel podcast, Huang said China could soon be able to train an AI model equivalent to the likes of Anthropic’s recently-announced Claude Mythos - which could have worrying effects on global cybersecurity.
Huang also raised concerns on the "enormous" amount of compute China currently has going unused, but said both his company and the US overall should still be in a strong position.
Chips and more chips
"The amount of capacity and the type of compute (Mythos) was trained on is abundantly available in China, so you just have to first realize that chips exist in China," Huang said, noting how Mythos was trained on a “fairly mundane capacity".
“They (China) have datacenters that are sitting completely empty, fully powered," Huang added. "You know, they have ghost cities, they have ghost datacenters too. They have so much infrastructure capacity. If they wanted to, they [could] just gang up more chips...Their capacity of building chips is one of the largest in the world.”
China already manufactures a huge amount of the chip hardware used across the world, as well as boasting many leading universities and AI researchers - which offers huge potential for collaboration, rather than rivalry, Huang noted.
“This is an area that is glaringly missing because of our current attitude about China as an adversary,” he said. “It is essential that our AI researchers and their AI researchers are actually talking.”
“Victimizing them, turning them into an enemy, likely isn’t the best answer,” he added. “They are an adversary.”
“We want the United States to win. But I think having a dialogue and having research dialogue is probably the safest thing to do.”
“It is essential that we try to both agree on what not to use the AI for.”
Mythos is a key part of Project Glasswing, a new cybersecurity initiative Anthropic is leading along with tech leaders to to identify and fix vulnerabilities in critical software.
It quickly gained widespread attention, and OpenAI also recently revealed GPT-5.4-Cyber, its Mythos rival, designed for cybersecurity pros to spot the next level of attacks.
