Manycore's shares jumped 172% at their Hong Kong open.
Market welcome and the numbers
Manycore Tech's Hong Kong debut surprised investors — its shares jumped 172% at the open. The Hangzhou-based design and AI firm saw its shares open at HK$20.70, a 172% rise from the HK$7.62 offer price. The IPO raised about HK$1.09 billion. On the grey market the stock was trading near HK$20.52 before the first session, signalling strong demand.
Retail investors flooded the offer — the retail tranche was reportedly subscribed around 1,590 times on some platforms. That appetite underlined investor interest in China-origin AI plays that promise to take artificial intelligence beyond text and images and into real-world, physical tasks.
From design tools to 'spatial intelligence'
Manycore began as a maker of space-design software. Products such as Kujiale and its overseas counterpart Coohom helped the company build customers among designers, manufacturers and construction professionals. Over the last few years the firm has shifted strategy. It now markets what it calls "spatial intelligence"—AI systems meant to perceive and act inside three-dimensional environments.
Huang Xiaohuang, Manycore’s chairman and co-founder, told the South China Morning Post that spatial intelligence is a foundational capability that can apply across industries. Victor Huang, identified by Fortune as Manycore’s chair and one of its co-founders, framed the IPO as a step to hire top engineers, buy more GPUs and expand the company's data assets. Those are explicit capital uses tied to the firm’s stated pivot from design software to AI for the physical world.
Products and data
Manycore has rolled out a suite of products under the spatial label. The company lists SpatialVerse, a platform for generating realistic synthetic datasets; SpatialTwin, an industrial digital-twin service; SpatialLM and SpatialGen, a spatial language model and a generative model; and LuxReal, a 3D content-creation tool.
Manycore packages tools that create 3D data together with platforms that apply that data in real systems, covering both data generation and application.
Victor Huang has said Manycore has amassed a very large 3D data repository—almost 500 million assets, by his account. This company sees that pool as a competitive advantage.
Huang argues that 3D assets can better encode the rules of physical spaces than raw video, and he has pushed Manycore’s models into the open-source ecosystem, releasing tools such as SpatialLM to the broader developer community.
How Manycore fits the 'world model' debate
Manycore's emphasis on spatial or "world" models sits alongside research from well-known AI figures. Fei-Fei Li, who helped create ImageNet, and Yann LeCun, former Meta chief scientist, have both highlighted the potential of models that can reason about environments. But approaches differ—some researchers prioritise video as a training source, while Manycore places more weight on curated 3D assets and synthetic data.
Jixun Foo, senior managing partner at Granite Asia and an early backer of Manycore, pointed out that many impressive robotic demonstrations remain scripted. He said investors and founders want systems that generalize: tell a machine to do something new, and it should figure out how without bespoke programming. That, Foo and Manycore argue, is where spatial intelligence can add value.
Funding goals and near-term priorities
Proceeds from the IPO are meant to address three practical bottlenecks for an AI startup chasing physical-world capabilities. First, Manycore plans to recruit engineering talent. Second, it expects to spend heavily on compute, buying GPUs and other infrastructure to train larger models. Third, the firm intends to expand and curate its datasets—both synthetic and real-world 3D assets—to improve model robustness.
Those steps are easy to list but expensive to carry out — buying GPUs and hiring senior engineers costs real money. GPU capacity and senior AI hires are costly. And training models meant to interact with real environments isn't just a compute problem; it requires careful data engineering and validation to avoid brittle behaviour when models leave controlled settings and meet messy reality.
Investor sentiment and the IPO wave
Manycore's rally fits a broader trend. Hong Kong has attracted multiple AI listings in recent months, and investors have shown a willingness to pay premiums for companies that tie AI to tangible business models or assets. Manycore's mix of legacy revenue from design tools and a high-profile pivot to spatial intelligence gave retail and institutional buyers a narrative they could back.
But the rally looks driven by investor expectations as much as by underlying fundamentals; sentiment pushed the price higher on day one. A large first-day gain boosts the company's public profile and gives it more ammunition to recruit and invest.
It also makes people wonder about valuation compression for later investors and long-term return prospects—issues every newly listed tech firm must answer as they move from private funding rounds to public scrutiny.
Competition and technical hurdles
Manycore isn't alone in pursuing world models. Large research groups and other startups are building systems that let machines understand scenes, navigate, and predict physical interactions. Competitors are building similar 'world models', so Manycore will need to prove its tech in real settings — robotics, factory automation or digital twins — to land recurring, paying customers.
Another hurdle is validation. Fielding models in factories or on physical robots demands rigorous safety testing and often bespoke adaptation for each use case. Manycore's strategy of offering tools like SpatialTwin suggests it plans to sell not just models but end-to-end platforms that industrial customers can integrate. That approach can broaden revenue channels beyond licensing models, But it also raises costs for integration and testing, which can slow deployments in industrial settings.o increases the scope of customer support and deployment work.
Open source, global reach and regulatory context
Manycore has released open-source models, which the company says helped build credibility and global adoption. The open approach can accelerate developer uptake and create community-driven improvements. It also helps Manycore show tangible outputs to potential customers and partners in markets beyond China.
Listing in Hong Kong rather than a U.S. Exchange gives Manycore proximity to Chinese capital markets and regional investors keen on AI plays. That choice may ease fundraising and recruitment in Greater China while exposing the company to Hong Kong's disclosure and governance regime. For international customers, the combination of open-source releases and a Hong Kong listing may ease some concerns—though cross-border concerns about data and tech rules remain part of the broader industry conversation.
What to watch next
Investors and industry observers will be watching how Manycore deploys IPO proceeds. Concrete milestones to follow include hiring traction for senior AI engineers, announcements of new GPU infrastructure and partnerships or pilots that show SpatialLM, SpatialGen or SpatialTwin working in live industrial settings. Revenue trends from the legacy design products will also matter—those cash flows can fund R&D while the AI business scales.
Finally, the company's ability to demonstrate that its 3D-asset strategy produces more robust, adaptable models than video-trained alternatives will shape Manycore's long-term competitive position. If the claim holds, Manycore could anchor a niche in the physical-AI market; if not, the firm will face a tougher fight against deep-pocketed research labs and platform companies.
Related Articles
Manycore said its shares opened at HK$20.70, 172% higher than the HK$7.62 offer price.
This article was created with AI assistance.
