Arm Introduces Arm CSS for Mobile 2: AI-Native Platform with C2 CPU Cluster for Agentic AI

Arm officially announced Arm Compute Subsystem (CSS) for Mobile 2, a next-generation AI-native semiconductor platform engineered from the ground up for on-device autonomous agentic AI and neural graphics. Anchored by the all-new Arm C2 CPU cluster and Mali G2-Ultra NX GPU, CSS for Mobile 2 delivers a 70% leap in on-device AI throughput and 30% lower energy consumption, establishing the silicon blueprint for upcoming 2nm and 3nm flagship smartphone processors.

Architectural Breakthrough: The Arm C2 CPU Cluster & SME2

To support multi-step agentic workflows where local devices autonomously parse user intent and execute operating system tools, Arm redesigned its core CPU hierarchy. The C2 cluster combines the high-IPC C2-Ultra core for single-thread tool calling, C2-Pro cores for multi-agent reasoning, and C2-Nano cores for always-on background telemetry. Crucially, Arm Scalable Matrix Extension 2 (SME2) is embedded directly into the CPU pipeline, enabling high-throughput INT4 and FP8 matrix math for localized small language models (SLMs) without waiting for discrete NPU wake-up cycles.

Subsystem Architecture & Capabilities

• Arm C2-Ultra Core: Flagship single-thread CPU delivering maximum instruction throughput for rapid agent planning and app interaction.
• Scalable Matrix Extension 2 (SME2): On-core matrix execution pipeline slashing LLM prefill latency by over 50%.
• Arm Mali G2-Ultra NX GPU: Next-gen graphics engine with dedicated neural tensor units and hardware ray tracing for interactive 3D rendering.
• Coordinated Power Management: Dynamic workload balancer directing agent sub-tasks across CPU, GPU, and NPU with 30% power savings.
• Turnkey 2nm Physical Implementation: Pre-validated silicon layout optimized for TSMC and Samsung 2nm Gate-All-Around (GAA) foundry nodes.

Driving the Next Wave of Flagship Mobile Silicon

Arm CSS for Mobile 2 will serve as the architectural foundation for premier 2027 mobile chipsets, enabling smartphone manufacturers to embed low-latency Large Action Models (LAMs) into operating systems without sacrificing thermal performance or battery longevity.

Platform Summary