An investor of OpenAI has invested in a Guangzhou-based PhD.

It is learned by PEdaily AI that Guangzhou Yundie Technology recently announced the completion of a Series B+ financing of several hundred million yuan, jointly invested by well-known institutions and individuals including StellarS Capital, Asia United Infrastructure (00711.HK), Changzhe Investment, Turing Quantum and other parties. The funds will be used for the iterative upgrading of embodied intelligence technology and large-scale implementation in real scenarios.

Sorting out the list of investors, the figure of StellarS Capital draws much attention. Probably unfamiliar to the public, this is a New York-headquartered investment institution founded by Mao Zhang in 2017, with approximately 15 billion US dollars of assets under management as of July 2026.

Yet its past investment track record is impressive — ByteDance, OpenAI, Anthropic... behind these world-class AI companies, there are traces of StellarS Capital's investments. Now, they have placed a bet on a Chinese embodied intelligence company.

From OpenAI, Anthropic to Guangzhou

Back in 2019, Tian Xuesong, together with a group of like-minded doctor partners, founded Yundie Technology in Guangzhou.

At that time, neither large models nor embodied intelligence were popular buzzwords, and their application scenarios were not clear either. From the very beginning, Yundie Technology chose a unique commercialization path: instead of developing a general-purpose robot first and then looking for its usage, it started from specific positions, searching for tasks that can be delivered continuously in vertical scenarios such as hotels, property management, medical health care and park inspection.

This in turn defines Yundie Technology's technical route. A single hotel delivery task requires the robot to read the work order, call the elevator and access control, and confirm the task result after arriving at the room. Medical triage, pharmacy operations and park inspection involve more systems, and the site may change at any time.

For customers, what they ultimately care about is always whether the robot can run stably, how much manual intervention is required, and the cost of completing a task.

Accordingly, Yundie Technology proposed Embodied Environment Intelligence (EEI): when the capability of robots still has boundaries, elevators, access control, cameras and business systems can be incorporated into the task system, so that robots, the environment and humans can complete the work together.

At the product level, the RoboForge system-level embodied brain is responsible for task decomposition, planning, skill invocation and exception recovery, while the WING task-type ontology goes to the site for execution.

In terms of business model, Yundie Technology is rolling out RaaS, namely Robot as a Service. Robots, scenario deployment and subsequent operation and maintenance are integrated into continuous services. The operation data, after authorization and desensitization, is returned to the R&D process to improve models and skills.

It is reported that Yundie Technology has served thousands of government and enterprise clients in the past few years and has achieved sustained profitability; the long-accumulated customer relationships and delivery teams have provided scenario entrances for new businesses, thus forming a positive cycle that is hard to replicate.

Behind the International Capital's Bet

This time, how did Yundie Technology catch the eye of international capital?

In StellarS Capital's investment methodology, they tend to value two indicators of a project: slope and curvature. In short, revenue, valuation and number of customers are only the position of a company at a certain point in time. The slope measures the speed of change of these indicators; the curvature further observes whether the growth is accelerating or decelerating.

Such investment criteria are reflected in StellarS Capital's past investment projects. For example, Anthropic achieved 47-fold annualized revenue growth in less than 18 months; OpenAI's growth rate is equally staggering. They are both the world's top AI companies, not only growing step by step, but also maintaining positive growth for a fairly long period of time.

If the slope represents the growth speed of a company, the curvature represents its long-term vitality. For investment institutions, both are equally important.

As for Yundie Technology, the slope comes from new scenarios, task volume and embodied business revenue; the curvature is ultimately reflected in delivery efficiency. Whether each completed project can leave reusable skills, interfaces and exception handling experience for the next project determines whether the company is just adding projects or building scalable capabilities.

In Mao Zhang's eyes, Yundie Technology belongs to the latter. In his words, what he values is the growth path of building "intelligent compound interest" relying on the system-level embodied brain with the environment-enhanced technical route as the core. "On-site tasks continuously generate success, failure and takeover data, which are used to improve the system, and the enhanced capabilities then serve more scenarios. As this cycle continues to operate, Yundie Technology precipitates these accumulations into reusable skills and system capabilities."

When AI Enters the Physical World

At the moment, the inflection point of embodied intelligence is emerging.

On the one hand, regulators and the secondary market have begun to reprice the authenticity of revenue. The growth curve of companies that only tell stories is declining, while the curve of companies that deliver real orders is rising; on the other hand, robots are shifting from one-time delivery to continuous operation, and RaaS is rewriting the revenue slope of this industry. With the superposition of the two inflection points, capital and resources in the industry are rapidly concentrating on a small number of enterprises with positive growth curvature.

The products of OpenAI and Anthropic run in the digital world, where software can reach massive users at nearly zero cost; embodied intelligence is slower and heavier — robots need to go through manufacturing, deployment and operation and maintenance, adapt to the on-site environment, and take responsibility for task results. Heavier assets, slower delivery and greater cash flow pressure are the most essential differences between the two.

However, no matter what type of company it is, they cannot avoid the same core proposition: whether technological progress can be transformed into sustainable revenue. The real competition always lies in commercialization capabilities.

The compound interest of large model companies comes from model capabilities, user scale and computing power investment; the compound interest of embodied intelligence companies occurs on site — the results, failure records and manual takeover data left after the robot completes the task, if they can be used to improve the system in reverse, will directly affect the speed and cost of the next delivery. This is the source of value that distinguishes embodied intelligence from pure software business, and it is also the part that is harder to replicate.

This is exactly why capital chooses to bet on Yundie Technology. It continuously accumulates real customers and data, which is then used to iterate technologies and products, thus bringing stronger product strength, and further consolidating and expanding the customer base.

Going forward, Yundie Technology has a clear path plan: after this round of financing, Yundie Technology plans to continue to improve the stability and deployment efficiency of robots, perfect the continuous operation capability, and expand the adaptation of RoboForge to different tasks and ontologies. Perhaps soon, the potential of the positive flywheel will be revealed.

"We hope to start from one real post after another, let robots run stably in actual businesses, and test the value of embodied intelligence with the results of continuous delivery." As Tian Xuesong, founder of Yundie Technology, said, when AI moves from the digital space to the physical world, the real competition has just begun.

This article is from the WeChat official account "PEdaily AI", author: Wu Qiong, published with authorization from 36Kr.