Arm has launched CSS for Mobile 2, a new AI-native compute platform that integrates a CPU cluster, GPU, and system IP specifically for the workload demands of agentic AI and cinematic mobile graphics. The platform arrives as mobile AI shifts from discrete features toward persistent agents that must maintain context, run applications, and coordinate services within the power constraints of a smartphone.
A GPU built for AI-native graphics
The centerpiece of the graphics push is the Arm Mali G2-Ultra NX, which Arm calls the industry's most advanced mobile GPU. It is the first Mali GPU to include dedicated neural accelerators, tightly coupling neural processing with traditional graphics pipelines. This design lets neural workloads run alongside graphics where the data already lives, rather than shuttling it to a separate block.
Arm said the architecture delivers up to 4x higher performance per watt for neural graphics compared to previous approaches. Traditional rendering also gets a lift, with up to 14% higher performance on existing game content versus the prior generation. The GPU introduces a new execution engine and a next-generation Ray Tracing Unit aimed at modern game engine features.
Several game studios are already building on the technology. Sumo Digital collaborated with Arm on a demo called Neural Dawn to showcase the GPU's capabilities. Tencent Games Central Tech's MagicDawn and Unity China's Tuanjie Engine are integrating Arm Neural Technology into developer workflows. NetEase plans to release an NSS-enabled version of Where Winds Meet this year, and technology demonstrations are underway for Arena Breakout Infinite and Infinity Nikki with Infold Games.
The CPU as orchestration engine
Agentic AI workloads change the CPU's role. Rather than just running inference, the CPU must schedule tasks, maintain context across applications, and coordinate accelerators - all while keeping latency low enough for an agent that reasons and acts on a user's behalf.
The new Arm C2 CPU cluster pairs the C2-Ultra, Arm's highest-performance mobile CPU, with efficiency-focused C2-Pro CPUs and two SME2 units. Doubling the SME2 capability delivers a 70% speedup on the latest Small Language Models. Against the previous C1-Ultra, the C2-Ultra provides up to 1.7x higher AI performance and 15% higher single-thread performance, while drawing up to 38% less power at equivalent performance. SME2 already appears in leading Android and iOS handsets, with support from Alipay, Google AI Edge Gallery, OPPO, and vivo.
For developers, Arm is pairing the hardware with a software stack that includes Arm KleidiAI and the Arm Neural Graphics Development Kit. A new AI Portal serves as a single entry point to discover, evaluate, and use optimized models, code, and tools - a practical consideration for teams working across AI for IT & Development and AI for Product Development.
Why this matters for product and development teams
CSS for Mobile 2 signals a shift in how mobile silicon is packaged for AI. Rather than optimizing one accelerator in isolation, Arm is selling a pre-integrated platform where CPU, GPU, and neural processing share the same pipeline. For silicon partners, that shortens the path to differentiated chips. For developers, it means a common target across devices - one where neural graphics and agentic workloads can be built with familiar frameworks and deployed at scale without hand-tuning for every hardware variant.
Your membership also unlocks:
