General-purpose embodied AI robotics company UniX AI has completed a 500 million yuan Pre-A funding round. Existing shareholders Xeno Investment and Yonghua Capital made further investments across multiple rounds, while new backers include Greenland Group, Keli Sensing, HG Capital, Xinjing Fuying, the Redbird Qihang Fund and Yulin Industrial Investment.

What stands out about the investor list is not the size of the round. Greenland is already a downstream application partner, Keli Sensing sits on the hardware supply-chain side, the Redbird Qihang Fund is backed by the Hong Kong University of Science and Technology, and Yulin Industrial Investment brings local industry resources. It is unusual in this year’s embodied AI market for a robotics company to bring customers, suppliers, a university and local government-backed capital into the same financing round.

The company says its customers have evolved from “product users” into “industry co-builders.” That may sound like corporate messaging, but it reflects a concrete shift. In the past, once a robotics company sold the hardware, everything else was the customer’s responsibility. Now that customers are also shareholders, both sides must work out how robots fit into existing workflows, how they are maintained day to day and how the next deployment can be replicated.

Capital Is Coming From Two Directions

Greenland Group joined the round as a strategic investor and is already a downstream application partner of UniX AI. The two companies plan to build a closed loop spanning robot models, hardware, data and solutions across security, hospitality and eldercare.

Under this arrangement, the deployment partner no longer merely sets requirements, signs off on delivery and pays the bill. As a shareholder, it now has a financial stake in whether the robots can keep operating in its buildings. For a robotics company still working to prove that its deployments are replicable, that alignment is worth more than a single order.

Upstream, there is Keli Sensing. The announcement groups it with the Redbird Qihang Fund, which is supported by the Hong Kong University of Science and Technology, Yulin Industrial Investment and other strategic investors from listed companies, upstream and downstream businesses, local industry and the university ecosystem. The focus is on hardware supply-chain coordination and the academic innovation ecosystem.

The bulk of a robot’s hardware cost typically lies in mechanical components such as joints, torque sensors and reducers. When a supplier becomes a shareholder, discussions extend beyond payment terms and pricing to joint development, production priority and how to share the cost of yield issues.

Verifiable Deliveries Gave Strategic Investors Confidence

“Deployment” is one of the easiest claims to inflate in a funding announcement. This time, UniX AI provided a set of figures that can be checked one by one.

Since the end of 2025, the company has consistently delivered more than 100 units a month, with monthly deliveries peaking above 300. It also highlighted solid collections and gross margins, two metrics that say more about a hardware business than shipment volume. Products can be pushed into distribution channels; cash collection and gross margins cannot.

Several flagship customers have each placed orders for more than 100 units.

Security, Hospitality and Retail Have Different Economics

The same three use cases come up repeatedly, but each presents a different set of challenges.

In security, the company has secured and fulfilled volume orders from several leading property developers, as well as major listed security companies in China and overseas, including Certis and Inspur. This use case requires long-duration monitoring, anomaly detection and coordination across multiple buildings. The key tests are reliability and battery life, not the precision of a dexterous hand.

In hospitality, its robots have entered routine operation at hotel chains including Hilton, Ji Hotel and Home Inn.

The retail example is the most concrete. At JD.com’s 7Fresh Coffee, a robot operated continuously for 24 hours and made 2,156 drinks without a single error, setting a Guinness World Record.

The significance of the 2,156-cup figure is that it measures output under continuous operating conditions. Making one cup of coffee is not difficult. The hard part is avoiding errors overnight without supervision and restarting the next morning without manual recalibration. Data like this is closer to what customers actually pay for than a laboratory measure of grasping success.

Three Algorithms, Each Handling One Part of the Task

On the algorithm side, the company divides its capabilities into three layers.

UniFlex is an efficient, generalizable imitation-learning framework that teaches robots operational skills from demonstrations. UniTouch is a visual-tactile fusion model for object perception and fine manipulation, determining whether the hand has made contact and what kind of material it is touching. UniCortex focuses on planning and executing long-horizon tasks: what to do first, what comes next and how to resume after an interruption.

According to the company, the system focuses less on improving the success rate of an individual grasp and more on understanding, linking and continuously executing tasks in complex environments. In other words, it is designed not to perfect a single action, but to ensure that an entire sequence can be completed autonomously.

On the research side, there is UniTac. In 2026, a paper on the unified multimodal tactile model, developed by UniX AI with universities and research institutions in China and overseas, was accepted by ECCV 2026. UniTac places object properties and sensor characteristics within a single framework for cross-sensor tactile understanding and generation. Different robots use tactile sensors from different vendors, with varying data formats and physical meanings. Understanding across sensors is a prerequisite for transferring experience from one customer deployment to another.

Modularity Shortens Adaptation Cycles, Not Just the Product List

On the hardware side, the company has established mature product platforms including Wanda and Panther. It develops key modules in-house, including joint modules, biomimetic robotic arms and motion controls, while supporting different end effectors, such as grippers and dexterous hands, for different tasks.

The announcement’s explanation of modularity is worth noting: the goal is not to add more product models, but to improve the reuse of hardware and software across tasks, shortening the time required to adapt to and deliver new use cases.

That gets to the real cost structure of commercial service robots. A customer may want a robot to patrol a building today and deliver items tomorrow. If every new task requires redesigning the hardware and repeating the entire calibration process, delivery costs will never come down. Modularity is about reducing the marginal cost of a new use case to an acceptable level.

The company says it has built an engineering system capable of supporting sustained volume deliveries, covering product design, supply-chain management, complete-unit testing, delivery and on-site deployment.

The Bottleneck Has Shifted From Demonstration to Maintenance

At the 2026 World Humanoid Robot Games, UniX AI won one gold, one silver and one bronze medal. It took first place in the hospitality category for the second consecutive year and also won a medal in the firefighting category.

The results themselves are not the main point. Hospitality and firefighting have clear rules but unforgiving environments. Winning the hospitality event two years in a row at least indicates that the capability was not a one-off.

The real question is what comes next. Founder and CEO Yang Fengyu said the robotics industry does not ultimately need to prove how many impressive demonstrations it can produce, but whether customers can keep using robots in daily operations. He said the company had previously devoted substantial resources to refining its algorithms, products and delivery capabilities in real-world settings. Now, a growing number of industry partners are not only buying its robots but also becoming investors and long-term collaborators.

The proceeds from this round point in the same direction: deeper research into embodied AI algorithms and further product iteration, improvements to the engineering and delivery system, and scaled deployment in existing use cases including property management, hospitality and retail.

At the demonstration stage, the question is whether a robot can perform a task. At the operations stage, the questions are who repairs it when it breaks, who adapts it to a new use case and who owns the data flowing back from deployment. The first can be answered at a product launch; the second requires organizational capability. In effect, strategic industry investors are helping share the cost of solving all three problems.