Arm today introduced the Mali G2-Ultra NX, which it calls its first AI-native mobile GPU. The product is aimed at phones and other edge devices that now run on-device assistants, image models, and small agents instead of shipping every token to the cloud.
That shift is no longer theoretical. Memory makers have been diverting capacity toward high-bandwidth memory for AI servers, and conventional DRAM prices have climbed sharply. Phone and PC vendors face a squeeze: cloud inference is costly at scale, and the DRAM that on-device models need is getting more expensive too.
An AI-native GPU is Arm’s answer to part of that squeeze. Graphics silicon that can also sustain transformer-style math lets SoC partners advertise local models without a separate giant NPU story. It also keeps Arm at the center of the Android hardware stack as Qualcomm, MediaTek, and others chase on-device agents.
The announcement will be judged in 2027 flagships, not in a press slide. If the NX family delivers usable tokens-per-watt, more of the “AI phone” pitch becomes silicon rather than a cloud demo.
The next mobile GPU war is about running models next to the camera and keyboard—not only rendering games—because the cloud cannot cheaply host every personal agent.
Key takeaway
Photo: Unsplash (smartphone). Sources: Unite.AI product report, September 8, 2026; broader reporting on HBM/DRAM tightness.
