SAN FRANCISCO — In a move poised to reshape the architecture of consumer electronics, Arm has officially announced its next-generation Compute Subsystem (CSS) for Mobile 2. Designed to serve as the foundational blueprint for silicon manufacturers building the brains of tomorrow’s premium Android smartphones, this latest release leans heavily into the artificial intelligence revolution.
By deeply integrating machine learning directly into the graphics pipeline and bolstering CPU clusters for complex, multi-tiered AI agent tasks, Arm’s new platform promises to alter how mobile devices render graphics, consume battery life, and execute autonomous software commands. The first consumer-ready silicon incorporating these advancements is expected to arrive in flagship devices by next year.
Main Facts: What is Arm CSS for Mobile 2?
At its core, the CSS for Mobile 2 is a comprehensive hardware and software design package that chipmakers—such as MediaTek, Qualcomm, and others—can license and integrate into their own System-on-Chips (SoCs). While every smartphone requires a central processing unit (CPU) and a graphics processing unit (GPU) to function, the modern high-end smartphone market demands an unprecedented level of specialized compute power to handle generative AI models, advanced mobile gaming, and complex OS-level automation.
The headline features of Arm’s CSS for Mobile 2 include:
- The Mali G2-Ultra NX GPU: Touted by the company as its first truly "AI-native" graphics processor, which embeds AI acceleration directly into the graphics pipeline rather than relying solely on a separate Neural Processing Unit (NPU).
- New C2 CPU Cores: Upgraded central processing cores specifically optimized for device-side "AI agent" workflows, offering significant performance gains while drastically reducing power consumption.
- Drastic Efficiency Gains: Proof-of-concept testing reveals up to a 4x increase in performance efficiency and a 70% reduction in external memory traffic compared to conventional rendering techniques.
- Overall Performance Lifts: A 24% boost in standard benchmark performance and a 14% improvement in ray-tracing capabilities.
Chronology: The Evolution of Mobile Silicon Architecture
To understand the significance of Arm’s latest announcement, it helps to examine the trajectory of mobile hardware over recent years.
The Shift Toward Heterogeneous Computing
For over a decade, smartphone chips evolved primarily through brute-force scaling—adding more cores, increasing clock speeds, and shrinking transistor nodes (moving from 7nm down to 3nm processes). However, as thermal limits and battery capacities hit natural plateaus, chip designers had to pivot toward heterogeneous computing. This meant offloading specific workloads to specialized hardware: ISPs for cameras, DSPs for audio, and eventually, NPUs for machine learning tasks like facial recognition and voice assistants.
The Rise of Generative AI (2023–2024)
The widespread emergence of generative AI in 2023 changed the requirements for mobile processors overnight. Suddenly, smartphones weren’t just running static algorithms; they were tasked with running large language models (LLMs) locally, generating images, and processing real-time translations. Last year, chipmakers like MediaTek rolled out advanced chips—such as the MediaTek Dimensity 9500, found in high-end devices like the Oppo Find X9 Pro—which utilized Arm’s first-generation C1 CPU cores and Mali G1-Ultra GPUs.
The AI-Native Transition (2025 and Beyond)
While first-generation AI phones relied on passing data back and forth between the GPU and the NPU, Arm’s announcement of CSS for Mobile 2 marks the next logical step: making the silicon components natively intelligent. By baking AI directly into the graphics rendering pipeline and redesigning CPU cores for multi-step agentic workflows, Arm is laying the groundwork for the 2026 smartphone generation, where AI will no longer be a separate feature, but the foundational operating layer of the device.
Supporting Data: Under the Hood of the Mali G2-Ultra and C2 CPUs
Arm’s technical briefings and whitepapers reveal substantial generational leaps across both graphical and computational metrics.
The Mali G2-Ultra NX: AI-Native Graphics
Traditionally, when a game or application required AI assistance to enhance graphics—such as upscaling a low-resolution image or frame generation—the GPU had to export texture and frame data across the chip to a dedicated NPU or general-purpose compute blocks. This data transit consumed precious milliseconds (adding latency) and burned through battery life due to high external memory traffic.
The Mali G2-Ultra solves this by introducing three core AI-driven rendering methodologies natively within the graphics pipeline:
- Neural Super Sampling: This technique takes a natively low-resolution render (such as 540p) and intelligently upscales it to a crisp high-definition image (such as 1080p) using neural networks, preserving edge fidelity and texture detail without the heavy rendering cost of native 1080p.
- Neural Frame Rate Upscaling: Instead of forcing the GPU to render every single frame, the chip generates intermediate frames via AI. For instance, a game running at a base render rate of 30 frames per second can be smoothly doubled to 60 frames per second.
- Neural Super Sampling and Denoising: Graphically demanding modern techniques like real-time ray tracing often introduce visual "noise" (dot artifacts or shimmering edges) in complex lighting scenarios. The G2-Ultra uses localized AI to scrub out this noise instantaneously within the pipeline.
To demonstrate these capabilities, Arm partnered with developer Sumo Digital to create a proof-of-concept fantasy game featuring a character exploring a subterranean cave system illuminated by vibrant, neon-lit giant mushrooms. The results of the internal testing were stark: the implementation of these AI methods yielded 4x higher performance efficiency and up to 70% lower external memory traffic compared to legacy rendering pipelines. Furthermore, overall benchmark performance saw a 24% improvement, while ray-tracing workloads benefited from a 14% speedup.
C2 CPU Cores: Built for AI Agents
While the GPU handles visual fidelity, the CPU remains the master conductor of the smartphone. Arm’s new C2 CPU cores have been architected specifically with "AI agent" workflows in mind—software designed to execute autonomous multi-step tasks on behalf of the user.
Compared to the previous-generation C1 cores, the new C2 architecture delivers:
- 15% higher single-thread performance (which directly reduces latency for interactive tasks).
- 12% higher multi-thread performance (allowing multiple background processes to run concurrently).
- 1.7x performance acceleration across AI models while consuming 38% less power.
- Faster application launch times and smoother web browsing.
To test the real-world impact of these upgrades, Arm simulated an agentic workflow that chained together several demanding tasks: real-time speech processing, internal memory retrieval, logical reasoning, application execution, and web browsing. The C2-powered setup completed this complex sequence 24% faster than its predecessor.
It is worth noting that individual silicon vendors retain ultimate control over how they configure these building blocks. For context, last year’s MediaTek Dimensity 9500 employed an eight-core CPU layout consisting of one C1-Ultra, three C1-Premium, and four C1-Pro cores. Future chipsets utilizing the CSS for Mobile 2 will similarly mix and match C2 variants to strike a balance between thermal output, cost, and peak performance.
Official Responses and Industry Perspectives
During a press briefing ahead of the formal announcement, Chris Bergey, executive vice president of Arm’s Edge AI Business unit, emphasized that these technological leaps are not merely about making phones faster, but about fundamentally rethinking device economics and physical design.
According to Bergey, the ability to offload heavy graphical lifting to AI pipelines opens up new possibilities for hardware form factors.
"If we can provide an amazing 540p, 30 frames per second experience [augmented by neural super sampling], and it allows you to use a smaller battery, that allows for a smaller form factor and potentially allows for a lower-cost handheld," Bergey explained.
By shrinking the physical battery size required to power high-end mobile gaming sessions—without sacrificing the perceived visual quality—manufacturers gain valuable internal chassis space. This could lead to thinner, lighter smartphones, or alternatively, make high-performance gaming handhelds more accessible and affordable to a wider consumer base.
Industry analysts have noted that Arm’s vertical integration of these subsystems places immense pressure on competitors. By offering a turnkey solution that handles both heavy CPU agentic workloads and AI-native graphics out of the box, Arm continues to solidify its near-monopoly grip over the architecture powering the vast majority of the world’s smartphones.
Implications: What This Means for Consumers and Developers
The ripple effects of Arm’s CSS for Mobile 2 announcement will be felt across the entire mobile ecosystem over the next twelve to twenty-four months.
For Mobile Gamers
Console-quality gaming on smartphones has long been hampered by thermal throttling and battery drainage. With the Mali G2-Ultra enabling efficient neural upscaling and frame interpolation, gamers can expect smoother frame rates (a stable 60 FPS or higher), richer ray-traced lighting effects, and longer gaming sessions before the device heats up or runs out of power.
For Everyday Smartphone Users
The upgrade to C2 CPU cores directly impacts day-to-day usability. As smartphone operating systems transition toward "agentic AI"—where personal assistants actively book flights, organize schedules, summarize communications, and execute multi-app workflows autonomously—the bottleneck has shifted from cloud servers to on-device processing. The C2 cores’ 1.7x AI model performance boost and 38% power reduction mean these autonomous agents will respond faster, run locally for better privacy, and drain the battery far less quickly.
For Hardware Manufacturers and Developers
For OEMs like Oppo, Vivo, Xiaomi, and MediaTek, Arm’s new subsystem provides a competitive roadmap for next year’s flagship releases. Developers, meanwhile, will need to adapt their engines to leverage Arm’s neural super sampling and frame generation toolkits. By standardizing these AI rendering techniques at the architectural level, Arm is making it easier for game studios to optimize titles for Android hardware without needing to custom-code proprietary upscalers for every individual chipset on the market.
As the mobile industry prepares for the commercial rollout of chips based on CSS for Mobile 2 next year, the line between traditional desktop-grade rendering and mobile graphics continues to blur—ushering in an era where artificial intelligence isn’t just an app on your phone, but the very fabric of how your phone operates.
