Arm Unveils AI-Native Chip Platforms for Mobile, Robotics, Cloud

Arm has introduced three specialized compute platforms designed to support agentic AI—autonomous systems that maintain context, coordinate tasks, and act independently across mobile devices, robotic systems, and cloud infrastructure.

The company announced CSS for Mobile 2 for edge devices, Total Design for Physical AI for robotics, and Neoverse CSS N4 for data centers. Each architecture addresses the computational shift from simple AI inference to persistent, autonomous workflows that require coordinated processing across multiple domains.

Mobile edge gets neural-integrated GPU

The CSS for Mobile 2 platform combines Arm's new Mali G2-Ultra NX GPU with the C2 CPU cluster. The Mali G2-Ultra NX marks the first mobile GPU to integrate dedicated neural accelerators directly into the graphics pipeline, allowing neural workloads to run alongside graphics processing without relying solely on traditional rendering.

The GPU delivers up to four times higher performance per watt for neural graphics tasks and a fourfold efficiency improvement overall. For conventional gaming, the architecture provides 14% higher performance versus previous generations. The platform also includes a third-generation Ray Tracing Unit and a new execution engine built for modern game engines.

The accompanying C2 CPU cluster pairs the C2-Ultra and C2-Pro processors with dual SME2 units, enabling a 70% speedup on recent Small Language Models. The C2-Ultra delivers 1.7 times higher AI performance and 15% greater single-thread performance than the C1-Ultra while consuming 38% less power. Combined with SME2, the processor completes agentic operations 24% faster and improves web browsing speeds by 15%.

Robotics framework standardizes capability levels

For physical AI—a sector Arm estimates will represent a $200 billion annual compute opportunity in the 2030s—the company expanded its Total Design ecosystem to more than 80 partners, including AWS, ECARX, Siemens, NXP, and Liquid AI.

Arm also introduced a Robotics Capability Framework that establishes six tiers of robotic sophistication, from RL0 (simple reactive behaviors) to RL5 (self-improving cognitive systems). The framework provides a common language for describing robotic capabilities, similar to SAE Levels for autonomous vehicles, helping developers map use cases to specific system requirements including latency, compute placement, and power constraints.

Data center platform scales to 128 cores

The Neoverse CSS N4 targets cloud infrastructure where AI agents retrieve data, access tools, and interact with databases. The architecture supports up to 128 cores per die and includes compatibility with LPDDR6 memory and PCIe Gen 7 connectivity.

Compared to the Neoverse CSS N3, the N4 configuration delivers twice the socket performance, 1.25 times the performance per watt, and 1.75 times the memory bandwidth. Cloud providers including Google Cloud and Microsoft Azure currently use Arm-based infrastructure to run agent sandboxes.

Why it matters

The shift from stateless AI inference to persistent agentic workflows requires fundamentally different compute architectures. By delivering specialized platforms for mobile, physical, and cloud domains simultaneously, Arm is positioning its IP across the full stack of autonomous AI deployment—from edge devices that must balance power and performance to data centers that orchestrate heterogeneous workloads. The robotics framework also addresses a critical gap in the physical AI sector by providing standardized capability definitions that reduce integration risk.

The platforms were detailed by All About Circuits, which first reported the announcement.