Key Takeaways

  • Arm CSS for Mobile 2 combines new CPU, GPU, system IP, physical implementations and software components for AI-centric mobile silicon.
  • Arm Mali G2-Ultra NX adds dedicated neural accelerators, while the Arm C2 CPU cluster doubles SME2 capability for local AI workloads.
  • The platform could shorten development cycles for chip partners, but adoption will depend on software support, sustained performance and shipping device designs.

Arm has introduced CSS for Mobile 2, an integrated compute platform designed to support persistent AI agents and neural graphics within smartphone power and thermal constraints. Announced September 8, the platform brings together the Arm Mali G2-Ultra NX GPU, an Arm C2 CPU cluster, enhanced system IP, physical implementations and a developer software environment.

Rather than treating AI as a workload assigned mainly to a standalone NPU, Arm positions the CPU, GPU and other system components as participants in a coordinated AI pipeline. That reflects the growing complexity of mobile agents, which preserve context, run applications, schedule models, call services and respond quickly while operating partly offline.

IDC reported that worldwide smartphone shipments grew 2.6% in Q3 2025, while analysts expect AI-capable models to account for an increasing share of future sales. Gartner projects end-user spending on generative AI-enabled smartphones to rise from $298.2 billion in 2025 to $393.3 billion in 2026. A Gartner edge-computing outlook projects that at least 60% of edge computing deployments will use composite AI, combining predictive and generative techniques, through 2028.

Running a single model is only one part of an agentic experience. The device must also coordinate memory, applications, sensors, security controls and communications with cloud services. That puts renewed emphasis on general-purpose CPU performance, even as dedicated accelerators handle highly parallel inference.

Arm addresses this requirement with the C2 CPU cluster, which combines the performance-oriented Arm C2-Ultra with efficiency-focused C2-Pro CPUs and two SME2 units. Arm reports the doubled SME2 capability provides a 70% speedup on the latest small language models. Compared with the C1-Ultra, the C2-Ultra reportedly delivers up to 1.7 times higher AI performance and 15% higher single-thread performance, while consuming up to 38% less power at the same performance tier.

An agent that pauses while opening an application or coordinating several services will degrade the user experience regardless of its baseline inference throughput. Maintaining single-thread responsiveness is critical to preventing system-level bottlenecks during these complex operations.

Arm Mali G2-Ultra NX is the first Mali GPU equipped with dedicated neural accelerators, allowing neural and conventional graphics processing to operate within the same pipeline. According to the company, when paired with Arm Neural Technology, the GPU offers up to four times higher performance per watt for neural graphics and up to 14% more performance on existing game content than the previous generation.

Neural rendering reconstructs detail, improves resolution or generates selected visual elements without requiring every pixel to be produced through conventional rendering. This allows game developers more computing headroom for cinematic lighting, ray tracing and higher perceived image quality, provided latency and image stability remain under control. Mobile graphics continue to rely on established APIs such as Vulkan and OpenGL ES maintained by the Khronos Group, making integration with familiar engine workflows a primary adoption factor.

Arm is demonstrating this ecosystem approach through multiple developer partnerships. Neural Dawn, developed with Sumo Digital, showcases the GPU's intended capabilities. Tencent Games Central Tech is integrating Arm Neural Technology through MagicDawn, while Unity China is supporting it through Tuanjie Engine. NetEase plans an NSS-enabled version of Where Winds Meet this year. Arm also demonstrated an NSSD build using Tencent Games' Arena Breakout Infinite and NSS integration with Infold Games' Infinity Nikki.

Arm KleidiAI, the Arm Neural Graphics Development Kit and the new AI Portal supply developers with optimized models, code and tools to avoid engineering separately for every hardware implementation. For silicon partners, the combined hardware and physical designs aim to reduce integration work and accelerate differentiated products.

Arm has not yet provided commercial device names or a shipment schedule. The platform's market impact depends heavily on chip designers, handset manufacturers and developers translating these component-level gains into sustained performance within commercial shipping devices.