Manycore, the first of the Hangzhou ‘Little Dragons’ to go public, pushes ‘spatial intelligence’ as the next wave of AI development

Hong Kong’s AI IPO boom produces its latest entrant today, as design AI startup Manycore Tech begins trading after seeking up to 1.02 billion Hong Kong dollars ($130 million) in funding, becoming the first of China’s six celebrated “Little Dragons” from Hangzhou to reach public markets.

“The IPO is important for us to attract the most talented engineers to join us, to buy more GPUs, and to collect more data,” Victor Huang, Manycore’s chair and one of its cofounders, told Fortune ahead of the trading debut.

Manycore shares closed at 18.60 Hong Kong dollars ($2.38), 144% above the offer price of 7.62 Hong Kong dollars.

The Hangzhou-based startup is a bet on “spatial intelligence,” moving beyond the word- and language-based work of large language models like OpenAI’s GPT and DeepSeek’s V3 to instead create AI models that can autonomously work in the real world.

These programs, also called “world models,” are key to operations like robotics and autonomous driving, where machinery has to react to external stimuli, like how a robotaxi needs to slow down in response to changing traffic conditions. 

Huang described spatial intelligence as similar to a person or animal’s innate ability to understand the world around them. “When you enter a room, you can understand where you are and what’s in front of you. And if you want to take a seat, you can understand which seat is empty,” he explained.

“People are now trying to apply AI in the physical dimension,” Jixun Foo, senior managing partner at the Singapore-based venture capital firm Granite Asia, and an early backer of Manycore, said. He pointed out that viral videos of humanoid robots dancing, while impressive, are often carrying out pre-programmed routines. “If you want a different performance you have to program it again. You can’t just tell the robot to do this or that action.”

Some of AI’s biggest names are also working on “world models.” Both ImageNet creator Fei-Fei Li and former Meta chief scientist Yann LeCun see these models as the next step in AI development. 

LeCun has argued that video data can help train world models, but Manycore and Huang instead think the startup’s vast repository of 3D assets will be a more useful data set. “I don’t believe that if you........

© Fortune