Arm has unveiled its next-generation mobile computing platform, CSS for Mobile 2, featuring AI-native CPU and GPU architectures designed to bring advanced artificial intelligence directly to smartphones and edge devices without draining battery life. The announcement marks a significant shift in how mobile chipmakers approach AI processing, embedding neural accelerators directly into graphics hardware rather than treating AI as a separate component.

What Makes Arm's New Mobile AI Architecture Different?

The CSS for Mobile 2 platform introduces two key innovations: the Arm C2 CPU Cluster and the Mali G2-Ultra NX GPU. The C2-Ultra CPU delivers up to 15 percent higher single-thread performance compared to its predecessor while incorporating dual Scalable Matrix Extension 2 (SME2) units that accelerate AI workloads directly on the processor. For AI-related tasks like language models and voice assistants, Arm claims performance improvements of up to 70 percent in selected scenarios.

The Mali G2-Ultra NX GPU represents the more radical departure from traditional mobile design. Rather than routing AI inference to a separate neural processing unit, Arm embeds dedicated neural accelerators directly inside each shader core, the fundamental processing units of a GPU. This architectural choice allows AI-powered rendering to access the same high-speed memory caches that graphics workloads use, eliminating the energy-expensive process of writing data to main memory and back.

How Does This Compare to Qualcomm's Competing Approach?

Just six days before Arm's announcement, Qualcomm revealed its own answer to mobile AI graphics: the Adreno Neural Fusion architecture. Both companies are solving the same problem, but they diverge on implementation. Qualcomm places Matrix Cores at the GPU slice level alongside an 18MB dedicated high-bandwidth memory cache, while Arm integrates neural accelerators into individual shader cores.

The practical difference matters for real-world performance. Arm's approach keeps AI inference workloads "where the graphics already live," reading from and writing to the same on-chip tile memory used for rendering. Qualcomm's slice-level design with dedicated memory eliminates most round-trips to main memory but operates at a different architectural granularity. Neither company has yet published independent benchmarks comparing the two approaches on shipping devices, though MediaTek is expected to bring CSS for Mobile 2 silicon to market within weeks.

What Gaming and AI Features Will Users Actually See?

The Mali G2-Ultra NX enables three distinct neural graphics technologies designed to transform mobile gaming experiences:

• Neural Super Sampling (NSS): Reconstructs higher-resolution output from lower-resolution renders, producing 1080p images from 540p source material in approximately 4 milliseconds per frame, similar to NVIDIA's DLSS technology on PC graphics cards.
• Neural Frame Rate Upscaling (NFRU): Generates synthetic intermediate frames from motion vectors and scene data, turning 30 frames per second into 60 FPS output or 60 FPS into 120 FPS without rendering each frame natively, though synthesized frames reflect slightly older input data.
• Neural Super Sampling and Denoising (NSSD): Combines upscaling with ray reconstruction for ray-traced scenes, targeting desktop-quality visuals on mobile power budgets by delivering up to 4 times higher efficiency and reducing external memory traffic by as much as 70 percent compared with native rendering.

Real games are already integrating this technology. NetEase's "Where Winds Meet" includes NSS support, while Tencent Games' "Arena Breakout Infinite" demonstrates NSSD capabilities. Infold Games' "Infinity Nikki" integrates NSS technology as well. Arm has also created a Neural Dawn demo built with Sumo Digital on Unreal Engine, showcasing how NFRU and NSSD can deliver the efficiency gains needed to make advanced Unreal Engine 5 features practical on smartphones.

Why Should Smartphone Makers Care About This Platform?

CSS for Mobile 2 is not a single fixed chip design; it is a reference platform that semiconductor companies can customize extensively. Chipmakers adopting CSS for Mobile 2 can use individual IP blocks, combine them with third-party designs, or configure the cluster for their own requirements. The example flagship configuration pairs two C2-Ultra performance cores with six C2-Pro efficiency cores and two SME2 units, a layout that closely matches expectations for MediaTek's upcoming Dimensity 9600 Pro chip.

This modularity accelerates development timelines. Arm has packaged CPUs, GPUs, system interconnects, software frameworks, and developer tools into a comprehensive ecosystem that simplifies chip design while allowing extensive customization. Samsung could substitute the Mali GPU for its in-house Radeon-derived design in future Exynos processors, and Google could do the same with Imagination Technologies IP, just as both companies have done in prior generations.

How to Understand the Real-World Impact of Neural Graphics

For consumers and developers, the practical implications of CSS for Mobile 2 break down into several key areas:

• Gaming Performance: Mobile titles will deliver richer visuals, smoother frame rates, and more immersive environments without excessive battery consumption, with Arm claiming up to four times higher performance per watt for neural graphics workloads compared with traditional rendering methods.
• AI Assistant Responsiveness: The platform supports agentic AI systems that understand intent, maintain context, plan actions, and coordinate tasks across multiple applications, making devices more responsive when running language models, voice assistants, and translation tools directly on the phone.
• Multimodal Processing: CSS for Mobile 2 enables next-generation mobile assistants capable of processing voice, text, and images simultaneously, with workloads intelligently distributed across CPUs, GPUs, AI accelerators, and cloud services for optimal efficiency.

The neural accelerators in the Mali G2-Ultra NX support INT8 and INT16 precision, the numerical formats most common in deployed neural network inference, which require less memory bandwidth and power than higher-precision calculations.

What Does This Mean for the Broader Mobile AI Race?

Industry analysts view the CSS for Mobile 2 launch as part of a broader shift toward AI-native computing, where artificial intelligence is no longer treated as an optional feature but becomes a foundational component of device experience. As smartphones evolve into intelligent personal assistants capable of understanding context and executing tasks autonomously, platforms like CSS for Mobile 2 could play a crucial role in shaping the future of mobile computing.

The real test of which architectural approach works better in practice will arrive when independent third-party benchmarks run on shipping devices. Arm's performance claims compare its hardware to its own prior generation, while Qualcomm's claims compare its implementation to its own prior Snapdragon Game Super Resolution system. Neither set of numbers has yet been measured by an independent reviewer with access to production silicon. MediaTek's expected launch of CSS for Mobile 2 silicon within weeks will provide the first opportunity for real-world comparison.