Physical AI companies are racing to prove their technology works in messy, unpredictable real-world environments, not just controlled lab settings. Two of China's most-funded embodied AI startups announced major milestones this week, signaling that the industry is moving past flashy demonstrations toward genuine commercial deployment in homes, factories, and logistics centers.

What's Driving the Shift From Lab Demos to Real Deployments?

For years, robotics companies have showcased impressive videos of humanoid robots performing tasks in pristine environments. But scaling robots to work reliably in the messy, unpredictable physical world is a fundamentally different challenge. X Square Robot, founded in 2023, has taken a deliberate approach by building what it calls a "full-stack" embodied AI system that combines foundation models, robotics hardware, and a proprietary data pipeline designed to improve robots through real-world experience.

The company announced it has secured four consecutive financing rounds culminating in a Series C, bringing its valuation to over $2.8 billion and making it the only embodied AI company in China backed by all four major internet technology leaders: Meituan, Alibaba, ByteDance, and Xiaomi. This level of backing from multiple tech giants suggests confidence that the company's approach to physical AI is viable at scale.

X Square Robot's strategy hinges on a critical insight: robots learn and improve when deployed in real environments. The company has launched two initiatives to test this theory. First, it partnered with 58.com, a Chinese property services platform, to deploy cleaning robots alongside human cleaners in real residential apartments in Shenzhen and Beijing. Second, it launched the "X Family Member Program," where robots live with families for up to one month as household companions, responding to everyday needs.

"As AI moves beyond digital experiences into the physical world, progress will depend on close integration between models, data and robotics," said Wang Qian, founder and CEO of X Square Robot.

Wang Qian, Founder and CEO of X Square Robot

How Are Companies Building Better Robot Foundation Models?

At the core of X Square Robot's approach is WALL-B, an embodied AI foundation model introduced in April 2026 that differs from earlier approaches in a fundamental way. Instead of connecting separate vision, language, and action systems, WALL-B trains all of these capabilities within a single unified network. This allows the model to develop stronger spatial reasoning and learn continuously from real-world interactions.

The company has also open-sourced two related models: WALL-OSS-0.5 and WALL-WM. WALL-OSS-0.5 achieved over 80% autonomous completion on four of 17 real-robot tasks without any additional training, a significant benchmark for generalization. WALL-WM introduces what researchers call "event-level prediction," which aligns language, vision, and action data around meaningful events to improve cross-modal learning.

To support rapid model improvement, X Square Robot built a scalable data pipeline that automates collection, cleaning, annotation, quality control, and augmentation of training data. This infrastructure is designed to create high-quality datasets for complex and long-tail scenarios, the kinds of edge cases that often trip up robots in real homes and factories.

What Is DEEP Robotics' Strategy for Global Expansion?

Meanwhile, DEEP Robotics, another major player in embodied AI, hosted its 2026 Global Partner Conference and unveiled a strategic framework called "1+X+N" designed to accelerate commercialization worldwide. The framework breaks down as follows:

• "1" (The Universal Brain): A unified, proprietary embodied AI system that integrates perception, decision-making, and motion control to power all of the company's robot products
• "X" (Versatile Physical Forms): Multiple robot platforms including humanoid, wheeled-legged, and quadruped designs engineered to adapt to different physical environments
• "N" (Diverse Industry Scenarios): Tailored solutions for specific industries such as energy, manufacturing, logistics, and public services

DEEP Robotics emphasized that the future of robotics will not be won by companies with a single great product, but rather by those that can integrate five core capabilities: versatile hardware platforms, advanced AI agents, scenario-specific solutions, robust supply chain ecosystems, and comprehensive service networks.

"The core competitiveness of future robotics enterprises will hinge on five integrated capabilities: versatile hardware platforms, advanced AI agents, scenario-specific solutions, robust supply chain ecosystems, and comprehensive service networks," said Zhu Qiuguo, CEO of DEEP Robotics.

Zhu Qiuguo, Founder and CEO of DEEP Robotics

DEEP Robotics has already begun scaling its operations. The company has established a manufacturing facility with an annual production capacity of 20,000 units and operates across more than 45 countries and regions. Its quadruped robots have been deployed in over 1,200 real-world industry scenarios globally, ranging from infrastructure inspections to emergency firefighting.

A major announcement at the conference was the commercial launch of the DR02, an all-weather humanoid robot designed for industrial applications. The company stated that the DR02 has completed its core technical validation and is now available for commercial deployment in 2026.

How to Evaluate Physical AI Companies for Real-World Viability

As the embodied AI market matures, investors and industry partners are learning to distinguish between companies with genuine commercial traction and those still relying on impressive demos. Here are key indicators of real progress:

• Real-World Deployment Data: Companies that can point to robots operating in uncontrolled environments like homes, factories, and public spaces are further along than those showing only lab results. X Square Robot's household cleaning service and family companion program represent genuine deployment, not staged demonstrations
• Backing From Multiple Strategic Investors: When major technology companies invest in different funding rounds, it signals confidence in the company's technology and market potential. X Square Robot's backing from Meituan, Alibaba, ByteDance, and Xiaomi suggests these companies see real commercial opportunity
• Proprietary Foundation Models: Companies building their own embodied AI models, rather than relying on generic large language models, are developing defensible competitive advantages. Both X Square Robot and DEEP Robotics have invested heavily in proprietary model development
• Manufacturing Scale: The ability to produce thousands of units annually indicates the company has solved supply chain challenges and has real customer demand. DEEP Robotics' 20,000-unit annual capacity is a concrete sign of commercial readiness
• Global Geographic Footprint: Deployment across multiple countries and industries demonstrates adaptability and reduces dependence on a single market. DEEP Robotics' presence in 45+ countries is a significant achievement

Why Does the Shift to Real-World Deployment Matter?

The robotics industry has long struggled with what researchers call the "sim-to-real gap," the challenge of taking robots trained in simulation or controlled environments and having them perform reliably in the messy, unpredictable physical world. Both X Square Robot and DEEP Robotics are tackling this problem head-on by deploying robots in real homes and factories, then using the data from those deployments to improve their AI models.

This creates a virtuous cycle: robots deployed in real environments generate high-quality training data, which improves the foundation models, which makes the robots more capable, which enables deployment in more complex scenarios. X Square Robot explicitly describes this as "a continuous feedback loop in which operational data improves model performance, helping accelerate progress toward general-purpose embodied intelligence".

The timing of these announcements suggests the embodied AI industry is reaching an inflection point. Rather than debating whether physical AI will work, companies are now focused on the harder problem of scaling it profitably. The next phase will determine which companies can maintain their technical advantages while building the supply chains, service networks, and partnerships needed to serve global customers reliably.