Arm reorganizes around Physical AI as enterprise robotics gains momentum

Arm has created a new Physical AI unit focused on robotics and automotive systems, a sign that enterprise AI is increasingly moving out of the data center and into machines operating in the physical world.

As part of the reorganization, Arm has split its operations into three core groups, separating cloud and AI technologies, edge products such as smartphones and PCs, and a newly formed Physical AI division that brings automotive and robotics under one roof, according to Reuters.

Arm’s decision comes as enterprises experiment with robotics beyond pilots, deploying autonomous systems in factories, warehouses, and logistics operations where real-time decision-making matters more than raw compute power.

This shifts AI workloads toward the edge, forcing CIOs to prioritize device reliability over cloud scale.

Enterprise implications

Arm’s move marks a structural shift in how computing is being aligned for robotics and automotive systems.

“The industry has moved through three distinct phases in the three years since the ‘ChatGPT moment’, from generative AI to agentic AI and now Physical AI,” said Neil Shah, vice president for research at Counterpoint Research. “Bridging digital agents to physical robots requires a massive investment in synthetic data. Unlike agentic AI, which can be trained on text or code, Physical AI requires ‘world models’ trained on high-fidelity video and physics simulations.”

For enterprises, this means planning infrastructure capable of supporting heavy, simulation-driven workloads needed to train robots across a wide range of real-world scenarios, Shah added.

Physical AI is also changing where AI workloads are executed. Arm’s approach shifts more inference and control functions toward edge and on-device environments, particularly for robotics and other real-time systems.

“These workloads require ultra-low latency, energy efficiency, and resilience, which centralized cloud cannot always deliver,” said Biswajeet Mahapatra, principal analyst at Forrester. “CIOs should adopt hybrid architectures: inference and control tasks at the edge or on-device using Arm-based accelerators, while training and large-scale analytics remain in the cloud.”

Networking also becomes a critical factor. Physical AI systems depend on predictable, low-latency connectivity to coordinate sensors and controllers in real time, particularly in factories and warehouses. This can push enterprises to revisit industrial networking designs, with greater emphasis on deterministic performance using technologies such as private 5G, Wi-Fi 7, and time-sensitive networking.

“The result is not cloud displacement, but a rebalance: the cloud serves as the system of learning and coordination, while Arm-based edge and device environments handle real-time perception, decisions, and physical action,” said Manish Rawat, semiconductor analyst at TechInsights.

Steps for CIOs

Preparing for Physical AI requires changes across the technology stack. “IT leaders need to optimize operating systems, AI frameworks, and container platforms for Arm architectures,” Mahapatra said. “Security and lifecycle management for distributed robotics systems must be strengthened. Running pilot projects with Arm-based robotic applications will help validate performance and integration before scaling.”

Rawat noted that enterprises should start by treating robotics and Physical AI as an extension of their core IT stack, not a niche OT experiment.

“This means designing applications with clear separations between training, orchestration, and real-time execution, so components can move cleanly between cloud and Arm-based edge or device platforms,” Rawat said.

The guidance reflects a shift toward treating robotics and Physical AI as long-term infrastructure investments, rather than standalone automation projects.

Arm’s enterprise strategy

With its increased focus on Physical AI, Arm is aiming to design highly optimized architectures as the AI economy shifts from paying for tokens generated to paying for precision in real-time decision-making in physical environments.

“Arm is designing end-to-end architecture to support decision making at the edge,” Shah said. “By standardizing on Arm across both the server and the robot, enterprises can create a ‘seamless compute fabric’ that allows these AI models to move from the cloud to the edge without rebuilding the underlying software stack.”

Standardizing on Arm can reduce fragmentation across device classes, streamline developer skills, and improve portability of workloads from data center to edge to machine.

“However, the risk lies less in vendor lock-in and more in dependency on Arm’s licensing and roadmap decisions as it moves closer to full chip designs,” Rawat said.

For most enterprises, adoption is expected to be gradual. CIOs are likely to begin with targeted deployments in controlled settings, such as factories or warehouses, before scaling robotics and autonomous systems more broadly across their operations.‘A wild future’: How economists are handling AI uncertainty in forecasts – ComputerworldRead More