Arm Debuts CSS for Mobile 2 to Power Agentic AI and High-End On-Device Graphics

Arm has announced its next-generation smartphone platform, the Compute Subsystem (CSS) for Mobile 2, designed to bring advanced AI agents and desktop-grade graphics to mobile devices. The platform marks a strategic shift toward "agentic AI," moving complex processing from centralized data centers directly to the device edge.

At the heart of the platform are the new C2-Ultra CPU core and the AI-native Mali G2-Ultra NX GPU. Technical specifications show the C2-Ultra core delivers up to 15% higher single-thread performance than its predecessor, the C1-Ultra. To accelerate AI tasks, the platform doubles its Scalable Matrix Extensions (SME2) units to two in flagship configurations, which Arm states delivers up to a 1.7x performance improvement across the latest tested AI models.

The Shift to Agentic AI

This platform represents a pivot toward supporting AI that does more than simply assist users. Agentic AI is designed to interpret complex objectives and execute sequences of actions autonomously—a requirement that demands higher parallel CPU performance and significantly lower latency than traditional cloud-based models.

Chris Bergey, Arm's EVP of Edge AI, noted that relying on data centers for every interaction creates significant implications for latency, connectivity, and the cost of delivering services at scale. By distributing these workloads to the device, Arm aims to increase responsiveness and user privacy while reducing the operational burden on cloud infrastructure.

Desktop-Class Graphics on Mobile

Beyond AI, the Mali G2-Ultra NX GPU focuses on bringing high-fidelity visuals to smartphones. The GPU introduces Neural Super Sampling (NSS) and Neural Frame Rate Upscaling (NFRU), technologies that cut external memory traffic by as much as 70% compared to native rendering. Arm claims this results in up to four times higher performance efficiency.

The GPU also features a third-generation ray tracing unit (RTUv3), which reduces ray tracing workloads by 70%. Deyan Lazarov, GPU Product Manager, stated that the G2-Ultra NX can sustain a 30 frames per second (FPS) gaming session on content with rich and intricate ray-traced scenery. Furthermore, the NFRU technology can upscale a 30 FPS render into a 60 FPS output, bridging the gap between mobile and desktop gaming experiences.

Industry Implications

By enabling high-performance AI agents and sophisticated gaming on-device, Arm is attempting to decouple the mobile experience from constant cloud dependency. This transition could lower the cost of AI services for providers and improve the reliability of AI features in areas with poor connectivity.

Industry observers will now look to see which smartphone manufacturers adopt the CSS for Mobile 2 and how the 1.7x AI performance boost translates to real-world agentic behavior in consumer applications. While the technical foundation is set, the success of the platform depends on the availability of software optimized for the new SME2 units and the RTUv3 hardware.