Pomegra
Beijing's RoboScience closed a Series A2 worth hundreds of millions of yuan led by Amber Capital, its third raise of 2026 as it pushes robot AI into logistics.
Key Takeaways
- RoboScience announced a Series A2 on November 2, 2026; the exact amount and valuation are undisclosed.
- Amber Capital, Hongruida Investment and Shanghai Semiconductor Industry Investment jointly led the round.
- The company was founded in December 2024 and has raised five rounds in under two years.
Lead
RoboScience, a Beijing company that builds AI models for robotic manipulation, announced on November 2, 2026 that it had closed a Series A2 round worth hundreds of millions of yuan. Amber Capital, Hongruida Investment and Shanghai Semiconductor Industry Investment jointly led. Shuang Capital, 360 strategic investment, Guolian Minsheng Capital, Jinding Capital and several listed-company investment platforms followed on. The company did not disclose the precise sum or its valuation.
Money will go toward iterating its embodied foundation models, running pilots in logistics, retail and industrial settings, and hiring senior talent.
What Does RoboScience Actually Build?
RoboScience sells software first and hardware second. Its core product is Visics, a general-purpose embodied model built on a vision-language-object-action design that separates a robot's perception and planning from its execution. The aim is a model that carries over between different robot bodies, objects and tasks without retraining from scratch.
A second product, RoboMirage, is a physics simulation engine released in September 2025 that generates training data at high concurrency. In June 2026 the company also launched the REX G1, a wheeled humanoid robot meant for logistics, retail, factory and home work.
Chief executive Tian Ye previously worked on Apple's AI platform and studied under Andrew Ng at Stanford. Chief scientist Shao Lin, a co-founder and former assistant professor at the National University of Singapore, also trained at Stanford.
How Has RoboScience Funded Itself So Far?
The company has moved quickly through five rounds since its December 2024 founding. A seed round of tens of millions of yuan came from Zero One Ventures. An angel round in July 2025, about 200 million yuan, was led by JD.com. A Pre-A in February 2026 followed, backed by Puhua Capital and Dayi Finance and Investment, and a Series A reported at 1 billion yuan came before the A2.
That sequence invites a few questions. The A2 is described as hundreds of millions of yuan, which suggests a smaller cheque than the preceding A. Splitting a Series A into A, A1 and A2 tranches is common in Chinese venture markets when a company is raising on a rolling basis, often to bring in strategic holders without resetting terms. Without a disclosed valuation, readers cannot tell whether the round was priced up, flat or structured around commercial commitments.
Why Are Industrial Investors Showing Up?
The investor list leans toward industrial capital, which points to commercial positioning over pure financial bets. A semiconductor-focused state-linked fund and multiple listed-company platforms sit alongside 360 and financial firms such as Amber Capital. Strategic holders typically bring distribution, components or customer access, and the company has said it wants proof-of-concept deployments in logistics, retail and factories.
RoboScience describes three routes to market: end-to-end embodied solutions, a cloud service delivered with Tencent Cloud, and edge controllers for developers and system integrators. Which of these produces revenue first is unstated. The company has disclosed no customer count or revenue figure.
Its pitch rests on cost. It says data collection runs about 0.05 yuan per sample, roughly one-twentieth to one-two-hundredth of conventional approaches, and that its video training set grew from 1 million clips in early 2026 to 4 million by October. Those are company figures, and no independent benchmark has verified them.
What Comes Next for Embodied AI Funding?
Capital is still flowing into Chinese embodied AI, but the questions are shifting from model demos to deployment economics. The sector has attracted large rounds across robot bodies, simulation tools and foundation models, and investors are now asking which companies can show repeat orders. A cross-embodiment model is only valuable if customers with different hardware actually adopt it.
Two scenarios stand out. If the Tencent Cloud channel and edge controllers draw integrators, RoboScience could build a software-licensing business that does not depend on selling its own robot. If pilots stall at the proof-of-concept stage, the A2 will look like a bridge round that postponed a harder pricing conversation.
Outlook
RoboScience has raised five rounds in under two years, anchored by a model, a simulator and a wheeled humanoid. The A2 adds industrial backers and fresh capital but few new facts about revenue or valuation. The next signals will be signed deployments in logistics and retail, and whether a larger round follows.
