Arm CSS for Mobile 2: AI-native platform for agentic experiences and neural graphics

Arm announced CSS for Mobile 2 on September 8, 2026, a new compute platform designed to address the system-level demands of agentic AI and AI-native graphics on mobile devices. The platform integrates the new Mali G2-Ultra NX GPU, the C2 CPU cluster with SME2 support, enhanced system IP, and developer tools. It targets next-generation Android smartphones and aims to deliver more responsive on-device AI agents alongside richer visual experiences.

Key hardware components

The Mali G2-Ultra NX introduces dedicated neural accelerators integrated directly into the graphics pipeline. Arm claims this enables up to 4x higher performance per watt for neural graphics workloads compared with prior generations. Traditional graphics performance sees up to 14% improvement on existing game content. The GPU also features a new execution engine and next-generation Ray Tracing Unit.

The C2 CPU cluster pairs the high-performance C2-Ultra core with efficiency-focused C2-Pro cores and two SME2 units. This configuration supports low-latency on-device AI directly on the CPU. Arm reports up to 1.7x higher AI performance and 15% higher single-thread performance versus the previous C1-Ultra, with power efficiency gains of up to 38% at equivalent performance levels.

Practical implications for users and developers

For end users, the platform promises more capable on-device agents that can maintain context, coordinate models, and execute multi-step tasks without constant cloud offload, all while staying within smartphone thermal and battery limits. Gaming and graphics experiences could benefit from neural reconstruction, frame generation, and upscaling techniques that were previously impractical on mobile.

Developers gain access through familiar Arm tools including KleidiAI libraries, the Neural Graphics Development Kit, and a new AI Portal for optimized models and code. Early integrations are appearing in game engines from Unity China, Tencent, and titles such as NetEase’s Where Winds Meet and Infold Games’ Infinity Nikki.

Who should pay attention

• Smartphone buyers interested in future on-device AI features and sustained gaming performance on flagship Android devices expected in 2027 and beyond.
• Game developers and engine teams building for mobile who want early access to neural graphics capabilities.
• Chip partners such as MediaTek and Google (Tensor) evaluating the platform for differentiated silicon.

Skip this platform announcement if your primary interest is immediate availability in current devices or cross-platform iOS/macOS consistency, as initial implementations are Android-focused and partner-dependent.

Limitations and trade-offs

Real-world gains depend on silicon partner implementations, game engine adoption, and optimized model availability. The first announced device integration is Xiaomi’s 18 Fold with its XRing O3 chip, but broader availability will take time. Vendor performance claims require independent validation once devices ship. Power and thermal constraints of mobile form factors remain the ultimate limiter for persistent agentic workloads.

Comparison to prior generation

Overall, CSS for Mobile 2 represents a meaningful architectural step toward integrated agentic AI and neural graphics on mobile. Its value will be determined by how quickly partners deliver devices and how effectively developers leverage the new capabilities.

Sources

• newsroom.arm.com
• 9to5google.com
• www.pocketgamer.biz

Cover photo by Merlin Lightpainting on Pexels, used under the Pexels License.