The artificial intelligence industry is shifting focus from language models to machines that can see, think, and act in the real world. A coalition of more than 80 companies spanning hardware makers, software platforms, and AI model developers announced a coordinated effort to standardize how intelligent robots and autonomous systems are built and deployed at scale.

What Is Physical AI and Why Does It Matter Now?

Physical AI refers to intelligent systems that combine AI models, software, sensors, and actuators into complete machines capable of operating independently in real-world environments. Unlike chatbots or content generators, physical AI systems must perceive their surroundings, reason about what they observe, and take precise actions in industries like mining, agriculture, manufacturing, and logistics.

The economic stakes are enormous. These industries represent trillions of dollars in global economic activity and represent a $200 billion annual compute opportunity in the 2030s, according to Arm's announcement. Yet they remain among the hardest to automate because robots must operate reliably in unpredictable, uncontrolled environments where mistakes can be costly.

How Is the Industry Standardizing Robot Capabilities?

The challenge facing robotics today mirrors a problem the autonomous vehicle industry solved years ago. When self-driving cars emerged, the industry lacked a common language to describe and compare automation levels. The solution was the SAE Levels framework, which created a shared vocabulary that engineers, regulators, and consumers could all understand.

Robotics needs something similar. Arm is introducing a Robotics Capability Framework designed to define levels of increasing sophistication for robotic systems. The framework connects real-world use cases with specific technical requirements including latency, compute placement, memory constraints, power consumption, determinism, and safety considerations.

"Robotics is advancing rapidly, but the industry still lacks a common way to describe, compare and communicate the capabilities of increasingly intelligent machines. This fragmentation makes robotic systems harder to design, integrate, and scale," explained Richard Grisenthwaite, chief architect at Arm.

Richard Grisenthwaite, Chief Architect, Arm

The framework's initial structure was informed by feedback from across the robotics ecosystem, including companies like ANYbotics, Fourier, Gravis Robotics, and others. Arm is inviting the broader industry to contribute expertise and help shape how the framework evolves.

Which Companies Are Leading This Effort?

Arm Total Design for Physical AI brings together a diverse ecosystem spanning the entire technology stack. Participants include cloud infrastructure providers, chip makers, open-source AI model platforms, and robotics specialists.

• Cloud and Infrastructure: AWS and other cloud providers offering compute resources and deployment platforms for physical AI workloads
• AI Models and Frameworks: Hugging Face, Liquid AI, and Qwen providing open-source models and tools that robots can run
• Hardware and Semiconductors: NXP and other chip designers creating processors optimized for edge computing and real-time control
• Robotics and Autonomous Systems: Unitree Robotics, PlusAI, and others building actual robots and autonomous vehicles
• Software and Operating Systems: QNX and Siemens providing real-time operating systems and industrial software stacks

This collaborative model has already proven effective in automotive. Arm, AWS, Google, HERE, RemotiveLabs, and Siemens collaborated on an integrated digital cockpit reference solution that enabled developers to test and validate complex automotive software before silicon was even available.

How Are Cloud Native Technologies Supporting AI at Scale?

While Arm focuses on physical AI standards, a parallel trend is reshaping how AI infrastructure operates globally. Cloud native technologies, which enable software to run reliably across different cloud environments, are becoming the foundation for AI systems moving from experimentation to production.

China offers a particularly revealing case study. The country is home to approximately 1.75 million cloud native developers as of the first quarter of 2026, with roughly 400,000 of those specializing in AI development. Professional industrial Internet of Things (IIoT) developers in China are more likely to be cloud native at 48 percent compared to the global average of 42 percent, signifying strong technical demand for complex deployments.

The adoption is especially pronounced among younger developers. Nearly 58 percent of Chinese backend developers under age 25 are classified as cloud native, up from 30 percent two years ago. This suggests cloud native practices are becoming embedded in how the next generation builds backend services.

What Path Do AI Teams Follow From Experimentation to Production?

As AI systems mature, they follow a distinct progression that relies on specific cloud native technologies and practices. Understanding this path helps explain why infrastructure choices matter as much as model quality.

• Data Pipeline Operations: Kubernetes and microservices provide the foundational infrastructure layer, followed by event-driven architecture, streaming services, and observability tools to support sophisticated data pipelines
• Training and Experimentation: Feature flagging, remote procedure calls (RPC), and immutable infrastructure support model version routing, distributed training, and reproducible environments
• Production Serving: Service meshes, chaos engineering, and multicluster management support traffic splitting, resilience testing, and distributed inference at scale

At the most advanced levels of cloud native maturity, these practices converge. The strongest technology association in recent research was between immutable infrastructure and chaos engineering, with a correlation lift of 2.22, meaning the two are adopted together at more than twice the expected rate. This "elite cluster" also includes multicluster management and service meshes, showing how developers operating complex systems adopt multiple advanced infrastructure practices simultaneously.

How Are Major Chinese Tech Companies Contributing to Open Source AI?

The push toward standardized, open-source AI infrastructure is accelerating in China, where leading technology companies are joining forces to advance the PyTorch ecosystem. PyTorch is an open-source machine learning framework that has become the default tool for AI researchers and engineers worldwide.

Alibaba Cloud and Cambricon joined the PyTorch Foundation as Platinum members, while Ant Group joined as a Gold member. These memberships grant the companies seats on the Foundation's Governing Board and Technical Advisory Council, allowing them to shape the direction of PyTorch development.

"Open source has become the default way the world builds AI, and the PyTorch Foundation is a leading hub for this innovation. Welcoming Alibaba Cloud, Cambricon, and Ant Group as members alongside Huawei further strengthens the global open source AI ecosystem," stated Mark Collier, Executive Director of the PyTorch Foundation.

Mark Collier, Executive Director, PyTorch Foundation

These companies are addressing different layers of the AI stack. Alibaba Cloud's Qwen infrastructure team is demonstrating how to serve large language models at massive scale. Huawei is exploring hardware-software co-design and interoperability between Chinese and global AI technology stacks. Cambricon is hardening PyTorch's device-agnostic foundation so any hardware accelerator can offer broader reach and richer native capabilities. Ant Group is showing how cloud native building blocks like Kubernetes and Kata Containers can be assembled into secure, on-demand runtimes for AI agents.

More than 250 organizations across China contribute to PyTorch Foundation projects including DeepSpeed, Helion, Ray, Safetensors, and vLLM, demonstrating the breadth of the open-source AI ecosystem in the region.

Steps to Understand the Shift Toward Production AI Infrastructure

• Recognize the Three Layers: Modern AI systems require standardization at the silicon level (chips and accelerators), the software level (frameworks and models), and the infrastructure level (cloud native platforms and deployment tools)
• Track Ecosystem Participation: Watch which companies join initiatives like Arm Total Design and the PyTorch Foundation, as membership indicates where industry leaders are investing resources and influence
• Monitor Maturity Progression: Understand that AI teams follow a predictable path from experimentation to production, and infrastructure choices differ at each stage
• Follow Open Source Developments: Open-source frameworks like PyTorch and tools like Kubernetes are becoming the foundation for production AI, making their evolution critical to enterprise AI strategy

The convergence of physical AI standardization, cloud native infrastructure maturity, and open-source ecosystem growth signals a fundamental shift in how AI moves from research labs to real-world deployment. Rather than relying on proprietary platforms or single vendors, enterprises are increasingly building on open standards and collaborative frameworks that allow flexibility while reducing fragmentation.