China is closing AI gap, says Schmidt

Posted by

May 9, 2026

On May 9, 2026

He also says US attempts to limit its rival’s progress through chip export controls are “largely beginning not to work”.

He attributes much of China’s – and specifically Huawei’s – progress to the way its engineers have used software “to get around the various latency and architectural problems”.

“What I like about this is that we have real competitors,” Schmidt said at this week’s AI+Expo in Washington DC. “What I don’t like about it is that China is very focused on the broad diffusion of this technology globally, and it’s all open-source, which means it’s largely uncontrolled and not controlled in any way by us.

“A year ago, I said I thought they were one-to-two years behind. It looks like they’ve caught up enough that the most recent analysis is China is within six months, which is a nanosecond in our world.”

The Expo is operated by Schmidt’s own technology think tank, the Special Competitive Studies Project. It also hosted a preview of research from SCSP’s Taiwanese partner on how China is innovating in AI hardware across back-end manufacturing by leveraging its expertise in PCB and substrate manufacturing.

Ines Chung, analyst at Taiwan’s Research Institute for Democracy, Society and Emerging Technology (DSET), said that patent analysis beyond Huawei’s Ascend 910C, towards the 950 and 970 series of AI accelerators indicates a move from a dual-die architecture today to a multi-die architecture and a potential “ninefold growth in memory bandwidth”.

Chung said it was less a question of winning on performance than on useability: “China is actually dedicated to another battlefield – one they are more competent in. They are trying to build good-enough performance, leveraging their industrial depth and manufacturing breadth to create deployable, operational compute at scale.

“My key takeaway is very straightforward,” said Chung. “The back-end is now the new front-end. With advanced packaging transforming AI scaling from transistor density to system-level integration, we should look beyond just chips.”

Importantly, the techniques Huawei is using are distinct from proprietary advanced packaging techniques (e.g., TSMC’s CoWoS or Intel’s EMIB (Embedded Multi-die Interconnect Bridge) processes) and use tools and materials not currently subject to US or other restrictions.

DSET will release the full research soon but is already pushing for an extension of the AI export controls regime to cover some advanced packaging equipment and key materials that China imports, including certain epoxy films and photoresists. This would likely require support from at least Japan and Germany as well as the US.

Schmidt, however, is also becoming increasingly concerned about maintaining a “multi-stakeholder” dialogue to prevent AI misuse, especially as China disseminates much of its innovation as open-source.

“This means that if you want to take DeepSeek V4 and you want to add powerful capabilities and you’re evil — which none of you are, and I’m not suggesting the Chinese would do this but somebody could do this, some evil person that exists in the world. How would we know that? How would we police it? How would we see it?” said Schmidt

“At least with the American models, the big ones are under corporate control and we know who to call, who the military can call, who the police can call and so forth. But if you release this stuff open source, it’s dangerous enough that a lone actor, a new kind of terrorist, could emerge. Obviously we want to avoid that. And we want to acknowledge it’s possible and get ahead of it. And it requires multi-stakeholder agreement.”