Manycore Tech Surges 144% in Hong Kong Debut as Spatial Intelligence Era Begins

Manycore Tech made history on Thursday when its shares hit the trading screen of the Hong Kong Stock Exchange for the first time, closing at HKD 18.60—a 144% premium over the HKD 7.62 IPO offer price, after surging as high as 185% intraday. In a single session, the Hangzhou-based startup became not only the world’s first publicly listed company focused on spatial intelligence, but also the breakout star of a Hong Kong IPO market that has been energized by a wave of Chinese AI listings in early 2026.

The debut lands at a pivotal moment in the physical AI story. As humanoid robots compete in half-marathons in Beijing, and as Gemini Robotics and Boston Dynamics race to deploy embodied AI in warehouses and homes, the bottleneck is increasingly clear: the 3D spatial data that robots need to understand and navigate the real world is scarce, inconsistent, and expensive to generate. Manycore is betting it has already built the infrastructure to solve that problem—and that the market agrees.

The Company Behind the Ticker

Founded in 2011 in Hangzhou, Manycore’s origins are surprisingly unglamorous for a company now trading as a spatial intelligence pioneer. The company started as a provider of cloud-based 3D design tools, primarily serving interior designers and architects who wanted to visualize spaces without expensive CAD software. Over fifteen years, it quietly assembled what it now describes as the world’s largest cloud-native spatial design platform—a library of interactive 3D environments, furniture models, material textures, and room layouts generated through billions of cumulative user interactions.

That library is the asset. Manycore has spent the past three years retrofitting it into an AI training data flywheel: spatial editing tools generate spatial data, which trains proprietary spatial large models, which improve the editing tools, which attract more users and generate more data. The company calls this its “spatial intelligence technology flywheel,” and it is the core thesis of the IPO.

The pivot to selling AI training data to robot makers is newer—announced in 2024 as the physical AI boom accelerated—but it flows naturally from the base asset. A robot learning to navigate a living room needs millions of examples of how objects relate to each other in three dimensions: where a sofa sits relative to a coffee table, how natural light falls across a kitchen at different times of day, what a partially opened door looks like from a robot’s eye-level camera. Manycore has those examples, at scale, in machine-readable form.

IPO Mechanics

Manycore priced its shares at HKD 7.62, the top end of the indicated range, issuing approximately 160.6 million shares globally. Total gross proceeds amounted to approximately HKD 1.224 billion, with net proceeds after fees of around HKD 1.092 billion—equivalent to roughly $140 million USD at current exchange rates. The stock trades on the HKEX Main Board under the ticker 00068.HK.

The company reportedly received strong institutional interest in the days before pricing, with the international tranche oversubscribed by a substantial multiple. The retail tranche in Hong Kong was similarly oversubscribed, in a sign that the spatial intelligence narrative resonated not just with tech-specialist funds but with the broader investor community.

Peak intraday gains of 185% are unusual even by the standards of Hong Kong’s notoriously volatile tech IPO market. The close at 144% above issue price still represents one of the strongest first-day finishes for a technology listing in Hong Kong since the AI investment wave began in earnest in 2025.

The Six Little Dragons

Manycore’s listing carries symbolic weight beyond its own business. The company is the first of the “Hangzhou Six Little Dragons”—a cohort of six AI startups from Hangzhou that captured the imagination of Chinese tech watchers after each attracted massive funding and global attention in rapid succession—to reach public markets.

The Six Little Dragons emerged as a group identity in late 2023 and 2024, when Hangzhou-based AI companies began producing models and products that credibly competed with Silicon Valley counterparts. The group is loosely defined but typically includes DeepSeek (large language models), Manycore Tech (spatial intelligence), Kuaishou’s AI research spinoff, and three others in robotics, code generation, and biotech AI. The shared “Hangzhou origin” narrative became a shorthand for a new generation of Chinese AI companies that were building on fundamentally different architectural choices—and, in DeepSeek’s case, achieving comparable capabilities at dramatically lower compute cost than Western rivals.

Manycore’s successful public debut puts the Six Little Dragons on the IPO path. Several of the others have reportedly filed or are preparing confidential filings for Hong Kong, New York, or both. The Manycore listing serves as a proof point that international investors are willing to pay premium valuations for Chinese spatial AI companies, even amid ongoing geopolitical friction over chip exports and AI governance.

What Spatial Intelligence Actually Means

The “spatial intelligence” label deserves scrutiny. In the industry, the term is used in at least two distinct ways. The neuroscience usage refers to the cognitive ability to understand and navigate three-dimensional space—an area where current large language models are notoriously weak. The commercial usage, as Manycore employs it, refers to AI systems trained on 3D spatial data that can reason about how objects, spaces, and physical forces interact.

Manycore’s spatial large models are designed for the latter. They can predict how a room will look when furniture is rearranged, how shadows will fall at different sun angles, and how a robot should plan a path through a cluttered kitchen. These capabilities are valuable for both the consumer design platform (helping users visualize renovations before buying paint) and the robot training use case (generating synthetic 3D environments for reinforcement learning).

The core technical question investors will eventually need to answer is whether Manycore’s moat is data-based or model-based. If the value lies primarily in the 3D data library—which is genuinely scarce and took fifteen years to accumulate—the business is defensible. If it lies primarily in model architecture choices that can be replicated by better-funded competitors with synthetic data generation pipelines, the defensibility is much narrower.

Looking Ahead

For now, markets have given Manycore a resounding vote of confidence. With a post-IPO market cap implying a valuation of approximately $1.8 billion USD at the first-day close, the company enters public life with significant resources and a clear mandate from investors to accelerate the pivot toward physical AI training data.

The company faces real competition. Nvidia’s Cosmos simulation platform, announced earlier this year, is targeting precisely the same synthetic physical-world data generation market. Google DeepMind’s robotic simulation infrastructure and Meta’s habitat simulation environments are also accumulating spatial training assets at scale.

What Manycore has that none of those competitors can easily replicate is fifteen years of real-world human interaction with 3D spaces—data generated by people who were actually trying to design, furnish, and visualize real rooms, not data generated by researchers running synthetic experiments. In a world where the quality of physical AI ultimately depends on how faithfully training data reflects the messiness and variability of the real world, that difference may matter more than any architectural advantage.

The market, at least, seems to think it does.