Unitree’s IPO Marks Physical AI’s Shift From Lab Demo to Public Market

Why Unitree’s IPO Is a Public-Market Test for Physical AI

Chinese robotics company Unitree Technology, one of the best-known humanoid robot makers, is moving toward an initial public offering—the clearest sign yet that physical AI is moving from research demonstrations into commercial validation and public-market pricing. Nvidia CEO Jensen Huang has defined physical AI as the shift from AI that understands the world to AI that changes it, requiring models to learn gravity, motion, collision and other real-world physical rules. Unitree's key strategic investor, Shoucheng Holdings, is using the moment to launch a 3.5 billion yuan fund with partners, dedicated to investing across the physical AI supply chain.

The financial backdrop is substantial. According to an industry tally cited in the report, more than $10 billion has flowed into physical AI over the past 18 months. In August alone, companies such as Feijie Kesi, Westlake Digital Intelligence and Huilun Technology raised hundreds of millions of yuan. But the destination of that money has changed sharply. In 2025, about 50% of physical AI funding went to early-stage startups; in 2026, first-round deals accounted for only 8%, while 92% went to companies with landed orders and mature products.

Unitree shipped more than 5,500 humanoid robots in 2025, among the largest volumes globally. Other embodied AI companies are also moving through capital markets: Zhiyuan Innovation has begun a Hong Kong listing process, and IPO applications from Leju Humanoid Robot and Deep Robotics have been accepted. UBTech's Walker S2 industrial humanoid is landing in car and electronics factories. In autonomous driving—the other major physical AI application—Momenta listed in Hong Kong in July, received approval from Germany's Federal Motor Transport Authority for nationwide urban L4 testing, and reports that its systems are installed in more than one million mass-produced vehicles across more than 210 designated models.

The growth ambition is large. Frost & Sullivan projects the physical AI industry will expand at a 47.2% compound annual rate to reach $3.25 trillion by 2040. Shenwan Hongyuan Securities describes 2026 as the key year for physical AI to separate from screen-based AI, and argues that real-world interaction data is now a resource as critical as lithium—because VLA models require trillions of physical interaction data points while public datasets remain in the millions. Shoucheng Holdings capital-markets manager Kang Yu told Securities Times that capital is no longer the main constraint; orders, offline deployment scenarios and genuine interaction data are.

How the Funding Shift, Data Scarcity and Momenta’s Milestones Shape the Race

Shoucheng’s 3.5 Billion Yuan Fund Is a Liquidity and Portfolio Play

As a core strategic investor in Unitree, Shoucheng Holdings is positioned to crystallize part of its earlier bet through the IPO. The new 3.5 billion yuan fund—created with multiple partners to invest across the physical AI chain—can reinvest that liquidity into suppliers, component makers and adjacent applications rather than relying on a single robotics asset. The logic is straightforward: if Unitree trades well, Shoucheng captures both a public-market benchmark and capital to deepen its physical AI exposure. The risk is concentration. A fund dedicated to one still-emerging industrial market will rise or fall with the pace of actual robot orders, not just financing rounds.

The Funding Shift From Early-Stage to Proven Players Is a Market Filter

The report's most striking data point is the change in capital allocation. In 2025, half of physical AI funding went to early-stage startups; in 2026, first-round deals fell to 8%, while 92% went to companies with landed orders and mature products. This suggests investors have stopped paying for laboratory prototypes. Instead, they are chasing companies that already have a deployable product and a distribution path—Unitree's 5,500 shipments, UBTech's factory deployments and Momenta's million-vehicle installed base are exactly the kind of proof now required. The practical consequence is a two-tier market: scale players can access late-stage capital and public listings, while pre-revenue startups face a sharply narrower funding window.

Real-World Data Is the New Bottleneck

Shoucheng Holdings capital-markets manager Kang Yu argues the industry is not short of money; it is short of orders, offline scenarios and real interaction data. The technical reason matters. VLA models require trillions of physical interaction data points, but public datasets remain in the millions. Collecting that data cannot be simulated indefinitely—it requires robots in warehouses, vehicles on roads and machines in ports. This is why deployment scale is now a competitive moat. Momenta's one million vehicles and 210-plus designated models feed its world model with real road dynamics; Unitree's robots in real environments can do the same for embodiment. Companies without such scenarios risk becoming data-poor even if their models are advanced.

Momenta’s German L4 Permit Adds a European Validation Layer

Momenta's listing in July and its subsequent approval to test L4 urban driving nationwide in Germany are separate but reinforcing signals. The listing gives public investors a physical AI benchmark outside robotics, while the permit shows a rigorous regulator accepting a real-world testing footprint. The limitation is that a test permit is not commercial deployment. Revenue, safety certification and operational conditions still need to be proven. But for competitors—Huawei ADS, Xpeng VLA and Pony.ai were named as following similar paths—Momenta's scale and regulatory progress raise the bar for data access and credibility.

What the Unitree Moment Means for Robotics and Autonomous-Driving Executives

The shift from broad early-stage funding to orders-backed capital changes what companies and investors need to demonstrate. Specific next steps follow from the data in this report.

• For robotics companies without committed orders: treat the 2026 funding mix as a closure of cheap early-stage capital. Unitree’s 5,500-unit 2025 shipment base, UBTech’s Walker S2 factory orders and Momenta’s one-million-vehicle installed base are the new proof points—not demo videos.
• For investors benchmarking the sector: use Momenta’s post-listing performance and Unitree’s IPO filing—when it publishes—to assess gross margin and order backlog, rather than extrapolating directly from the $3.25 trillion 2040 market forecast.
• For automakers and tier-one suppliers: evaluate autonomous-driving partners by accessible real-road data and designated-model count. Momenta reports one million vehicles and more than 210 models; Huawei ADS, Xpeng VLA and Pony.ai’s world-model systems should be compared on the same installed-base basis.
• For companies deploying physical AI: prioritize owned industrial scenarios—warehouses, ports, factories—because Kang Yu identifies offline scenarios and real interaction data, not capital, as the scarce resource. A documented port automation or industrial mobile robot rollout is strategically more valuable than a general-purpose prototype.