Arm has officially unveiled its Lumex Compute Subsystem (CSS), a new integrated platform designed to dramatically accelerate on-device AI for flagship smartphones and next-generation PCs. An AI-First Architecture Built on SME2At the core of the new platform is the Scalable Matrix Extension v2 (SME2) architecture, integrated directly into the CPU cluster. The CPU as the Center of On-Device AIArm is making a strategic argument that the CPU is the most reliable and universal processor for on-device AI. KleidiAI: Bridging Hardware and Developer EcosystemsA novel hardware architecture is only as powerful as the software that can harness it. The libraries integrate SME2 acceleration directly into major AI frameworks, including PyTorch ExecuTorch, Google LiteRT, and Microsoft ONNX Runtime, according to Arm’s press release.