What does Arm's new mobile platform change for the phone in your pocket?

The CPU cluster that orchestrates the workflow

Agentic AI does not sit in one corner of a chip. It moves across speech recognition, memory retrieval, reasoning, application execution and web browsing in a single interaction. Arm designed the C2 CPU cluster to coordinate that movement.

The cluster pairs C2-Ultra and C2-Pro cores with two Scalable Matrix Extension 2 units. That doubles the SME2 capacity of the previous generation and gives the CPU more headroom for latency-sensitive on-device model work.

C2-Ultra delivers up to 1.7x performance improvement across the latest AI models. Single-thread performance rises up to 15 percent. Web browsing becomes 15 percent faster. Application launch improves by 12 percent. Multi-thread performance at the cluster level grows 12 percent.

The orchestration layer maintains task state and assembles relevant context. It decides where each stage should execute. That decision can route work to local applications or on-device models. It can also use other compute resources or cloud services. The CPU coordinates permissions and decisions. It also drives the progression of the task.

Across a representative end-to-end agentic workflow spanning speech processing, memory retrieval, reasoning, application execution and web browsing, the C2-Ultra with two SME2 units completes the workflow 24 percent faster than the previous generation.

Neural acceleration inside the GPU

Graphics workloads are becoming neural. Mali G2-Ultra NX places dedicated neural accelerators directly inside the shader cores. A new execution engine and a third-generation ray tracing unit sit alongside them. This gives the GPU a native path for neural graphics without bouncing data back to the CPU. The GPU acceleration complements the agentic AI capabilities of the C2 cluster rather than duplicating them.

System interconnect and software that tie it together

The SI L2 system interconnect provides lower-latency access and higher bandwidth across the platform. Coherency and quality-of-service features help manage and prioritize activities when CPU and GPU operate concurrently and share memory. Without that fabric, the heterogeneous workload would stall on data movement.

Software support arrives on day one. Arm KleidiAI and framework integrations give more than 22 million developers optimized paths for Arm CPUs. This includes workloads that use SME2. The Arm AI Portal adds a single place to discover optimized models across Arm platforms. Developers can compare and evaluate them there. The Arm MCP Server brings those capabilities into agentic development environments. Developers targeting CSS for Mobile 2 can take advantage of SME2-based AI acceleration on the C2 CPUs and integrated neural acceleration on Mali G2-Ultra NX.

Where the platform stops and the silicon partner begins

CSS for Mobile 2 is a configurable platform. Individual components can be used independently or combined within a broader CSS configuration alongside custom and third-party IP. Partners shape systems around their product and performance targets. Physical implementation optimization is included, allowing performance, power and area to be considered alongside the architecture before SoC integration begins. The announcement does not name launch devices, pricing or which partners have committed to tape-out.