Arm Unveils CSS for Mobile 2 Platform for 2027 AI-Native Flagships
The new compute foundation integrates C2 CPU cores and Mali G2-Ultra NX GPU to support agentic AI and high-fidelity gaming.
Arm has introduced the CSS for Mobile 2 platform, an AI-native compute foundation designed to drive the next generation of flagship Android smartphones. Unveiled on September 8, 2026, at the Arm Everywhere China conference in Shanghai, the platform marks a strategic shift toward hardware optimized for autonomous AI agents and neural graphics.
The platform integrates the new Arm C2-Ultra and C2-Pro CPU cores alongside the Mali G2-Ultra NX GPU. Technical specifications reveal that the C2 CPU cores are equipped with Scalable Matrix Extension 2 (SME2) units to accelerate AI workloads. Performance benchmarks indicate that the C2-Ultra CPU delivers 1.7x higher AI performance compared to the previous C1-Ultra, while providing a 15% increase in single-thread performance.
The Shift to AI-Native Architecture
This architectural evolution arrives as mobile computing moves away from general-purpose processing toward "agentic AI," where devices take autonomous actions on behalf of the user. To support this, Arm is implementing tighter integration between the CPU and GPU. This approach aims to reduce the latency associated with Large Language Models (LLMs) and neural rendering, moving beyond a reliance on standalone Neural Processing Units (NPUs) to distribute AI tasks across the entire compute complex.
Neural Graphics and Gaming
On the graphics front, the Mali G2-Ultra NX GPU introduces dedicated neural accelerators specifically for native neural graphics. These include Neural Super Sampling (NSS) and Neural Frame Rate Upscaling (NFRU), technologies designed to enhance visual fidelity without a linear increase in power consumption. These advancements enable the platform to support high-performance mobile gaming sessions at rates up to 120 FPS, catering to the increasing demand for console-quality experiences on handheld devices.
Industry Implications and Outlook
By embedding neural acceleration directly into the GPU and CPU cores, Arm is positioning itself to define the hardware requirements for the 2027 flagship cycle. The ability to handle complex AI agents locally on-device is expected to reduce cloud dependency and improve user privacy and response times. Industry observers will now look to chip designers and OEMs to see how these C2 and Mali G2 components are implemented in the first wave of 2027 Android silicon, particularly regarding real-world battery efficiency and the actual deployment of agentic AI features.