The bell rang at 9:30 on a Friday morning in Hong Kong. By the time it stopped ringing, Manycore Tech was up 172%. By close, it settled at HKD 18.60 — a 144% gain over the HKD 7.62 offer price. Market cap: roughly HKD 35 billion, about $4.5 billion US.

For a company most Americans have never heard of, built on a product category most haven't either, that's a strange number.

Manycore is the first of Hangzhou's "Six Little Dragons" to go public — a cluster of Chinese AI startups the state has been quietly grooming. The company started in 2011 as a cloud tool for interior designers. The flagship product, Kujiale, lets Chinese furniture retailers render apartments in 3D before construction begins. The international version is called Coohom. Millions of users across 200-plus countries draw rooms on it every day.

Then, roughly eighteen months ago, Manycore rebranded. It is no longer a 3D design company. It is, according to its prospectus, the "world's first spatial intelligence" company.

What's actually in the box

Manycore owns the largest indoor spatial dataset on earth: around 500 million structured 3D environments and 440 million product models. That's not a metaphor. Every chair, desk, and doorframe its users have ever placed in a virtual room has been captured, labeled, and stored. Those scenes are now training data for what the company calls SpatialVerse — a model that's supposed to give robots and AI agents a sense of where things are, not just what they are.

They also announced SpatialTwin, an "industrial digital twin" product aimed at factories, and LuxReal, a 3D content generator. Co-founder Victor Huang (real name Huang Xiaohuang) spent years on Nvidia's CUDA team before coming back to China. This isn't a company that stumbled into the AI pitch. It spent a decade stockpiling the one thing LLMs famously lack — grounding in physical space — and then waited for the market to catch up.

On Friday, the market caught up hard.

The gap this exposes

Cerebras filed for a US IPO months ago and keeps pushing it back. Figure AI, Physical Intelligence, Skild, all the American "world model" startups are still private, still raising, still PR-postponing. The biggest American AI story of this week was OpenAI's shareholders accusing Anthropic of inflating revenue by $8 billion in a leaked memo. The week before that was a firebombing at Sam Altman's house.

Hong Kong listings raised nearly $14 billion in Q1 alone, most of it AI. The US spent the same quarter arguing about liability laws, open-sourcing Claude by accident, and forming defense pacts. One market is shipping companies. The other is shipping lawsuits.

My Opinion

Here's what bugs me: spatial intelligence is a real category. Fei-Fei Li's been saying this for two years — that LLMs will hit a ceiling, that the next wave needs models that understand the physical world the way a two-year-old understands a room. The American response was to start yet another startup (World Labs) and raise $230 million on a deck. The Chinese response was to take a profitable interior design business, quietly bolt a 500-million-scene dataset onto it, and IPO.

I'll be blunt. I think this is the playbook going forward, and I think US AI is going to miss it. When your "frontier labs" are burning $15 million a day on video generation that earns $2 million total, and your competition is a design software company that already has paying customers AND the world's biggest spatial dataset, the math stops favoring the burn-rate approach. Revenue, data, and a public-market exit beat another demo video.

The part that should worry people in San Francisco isn't that Manycore exists. It's that Manycore is the first of six. Five more of these "little dragons" are sitting in Hangzhou right now, lining up bankers. By the end of 2026, "spatial intelligence" won't be a category American VCs brag about being early to. It'll be a category American companies compete against, from behind, in a market they don't control.

Author: Yahor Kamarou (Mark) / www.humai.blog / 17 Apr 2026