Arm Now Making Its Own Chips, Targeting AI and Data Centre Markets

KOMPAS.com – The British chip design company, Arm, has officially stepped out of its “comfort zone”. Previously, for more than 30 years, the company had only designed chip architectures to be licensed to third-party manufacturers; now Arm is beginning to produce its own chips. Arm CEO Rene Haas stated that this move is driven by the surge in computing demand, particularly for artificial intelligence (AI) needs. “This is a new business for Arm. We are now supplying our own-made processors (CPUs),” Haas said in a statement, quoted by KompasTekno from Wired. Arm stated that this step is also driven by demands from various industry partners who require ready-to-use computing solutions for large-scale AI. The first Arm chip is named the Arm AGI CPU. This chip is specifically designed for generative AI and agentic AI needs, which are AI systems that can act independently. In terms of its primary function, the chip is intended for high-performance servers in data centres, not for mobile phones or PCs. Arm described the CPU as the foundation of computing for the “agentic AI cloud” era, where AI systems run continuously and coordinate automatically without human intervention. In this context, Arm noted that the role of the CPU is becoming increasingly crucial. This is because the CPU now must manage thousands of AI tasks in parallel in a short time. These tasks include memory and storage management, workload orchestration, and coordination of interactions between AI agents on a large scale. Technically, the Arm AGI CPU is built using a 3 nanometre (nm) fabrication process by Taiwan Semiconductor Manufacturing Company (TSMC). The Arm AGI CPU has up to 136 cores based on the Neoverse V3 architecture and is designed for 1U rack servers. In one reference server configuration, two chips can be combined in a single blade (modular server unit) with a total of 272 cores. In terms of performance, Arm claims the Arm AGI CPU can deliver up to twice the performance compared to CPUs based on x86 architecture in agentic AI workloads.