Arm Rolls AI-Native Compute Platforms for Edge, Physical, and Cloud Systems

Today, Arm launches a new suite of compute platforms and programs targeting the specific demands of agentic AI across mobile devices, robotic systems, and cloud infrastructure. The newly announced architectures and ecoystem address the shift from basic AI inference to persistent, agentic workflows where models autonomously maintain context, coordinate tasks, and act on behalf of users.

Arm’s new compute platforms are geared for edge AI, physical AI, and cloud AI.

To support this transition, Arm unveiled the CSS for Mobile 2 platform, the Total Design for Physical AI ecosystem, and the Neoverse CSS N4 for data centers.

CSS for Mobile 2 and the Mali G2-Ultra NX GPU

For the mobile edge, Arm introduced CSS for Mobile 2, an AI-native compute platform that combines the new Mali G2-Ultra NX GPU and the C2 CPU cluster. The Mali G2-Ultra NX is the industry's first mobile GPU to integrate dedicated neural accelerators directly into the graphics processing pipeline. This tight integration allows neural workloads to run where graphics already reside, enabling the generation and enhancement of visual detail without relying entirely on traditional rendering techniques.

CSS for Mobile 2 blends the new Mali G2-Ultra NX GPU and the C2 CPU cluster.

The Mali G2-Ultra NX features a new execution engine and a third-generation Ray Tracing Unit built for modern game engines. By combining these elements with Arm Neural Technology, the GPU achieves up to four times higher performance per watt for neural graphics and a four times efficiency boost overall. For traditional gaming content, the new architecture delivers up to 14% higher performance compared to previous generations.

Alongside the GPU, the CSS for Mobile 2 platform includes the C2 CPU cluster, which acts as an orchestration engine for complex AI workflows. The cluster pairs the high-performance C2-Ultra and efficiency-focused C2-Pro CPUs with two SME2 units. This doubled SME2 capability enables a 70% speedup on the latest Small Language Models.

Arm says the C2-Ultra processor provides up to 1.7 times higher AI performance and 15% greater single-thread performance than the older C1-Ultra, all while utilizing up to 38% less power. In practical applications, the C2-Ultra combined with SME2 enables agentic operations to complete 24% faster. The CPU also yields a 15% increase in web browsing speeds and a 12% improvement in both app launch times and multi-thread performance.

Advancing Physical AI with a Robotics Capability Framework

To support the physical AI sector, which represents an estimated $200 billion annual compute opportunity in the 2030s, Arm expanded its collaborative ecosystem to create Arm Total Design for Physical AI. More than 80 companies, including AWS, ECARX, Siemens, NXP, and Liquid AI, have joined the initiative. This program aims to accelerate the development of autonomous systems by providing a shared foundation of software stacks, AI models, sensors, and compute hardware.

As part of this collaborative effort, Arm also introduced a Robotics Capability Framework. This framework establishes a common language for describing and comparing the capabilities of robotic systems, similar to the standard SAE Levels used for autonomous driving.

Arm’s Robotics Capability Framework

The structure defines robotic sophistication across six tiers, ranging from RL0 for simple reactive behaviors to RL5 for self-improving, cognitive systems. The framework is intended to help developers connect real-world use cases with specific system requirements, such as latency, compute placement, and power constraints, thereby reducing integration risk.

Neoverse CSS N4 for Cloud Infrastructure

Agentic AI also drives significant CPU-based computation in the data center as AI agents retrieve data, access tools, and interact with databases. To accommodate these heterogeneous workloads, Arm released the Neoverse Compute Subsystem (CSS) N4. The CSS N4 is characterized as the company's most configurable CSS to date, designed to provide a fast path to custom silicon for scale-out and data-plane use cases.

Arm claims the new Neoverse Compute Subsystem (CSS) N4 as its most configurable CSS to date.

The CSS N4 architecture supports up to 128 cores per die. It features compatibility with LPDDR6 memory and PCIe Gen 7 connectivity. Compared to the preceding Neoverse CSS N3, the new N4 configuration delivers up to twice the socket performance, 1.25 times the performance per watt, and 1.75 times the memory bandwidth.

For cloud providers and OEMs seeking production-ready solutions without designing custom chips, Arm is promoting the continued adoption of its Arm AGI CPU. The AGI CPU provides a highly responsive foundation for agentic AI tasks, with multiple cloud providers, including Google Cloud and Microsoft Azure, currently using Arm-based infrastructure to run and accelerate agent sandboxes.

All images used courtesy of Arm.