arm agi cpu / neoverse / rack-scale ai

Arm announced the Arm AGI CPU, its first Arm-designed production silicon built on the Neoverse platform to target “agentic” AI cloud workloads that run continuously at global scale. The processor and reference server designs (1OU, 2-node blades with 272 cores; air- and liquid-cooled rack options up to ~45,000 cores) emphasize sustained per-task performance, memory bandwidth, and power-efficiency to orchestrate thousands of parallel agents and accelerators. Arm claims over 2x rack-level performance versus recent x86 systems by improving single-thread work and usable threads under sustained load. The move marks Arm extending from IP to full silicon to meet hyperscale customers’ demand; early ecosystem partnerships and deployments are already in progress.

Arm announced on March 24, 2026 that it will produce production silicon for the first time, launching the Arm AGI CPU—an Arm-designed data-center processor for agentic AI workloads. Developed with lead partner Meta and supported by ODMs and other customers, the Arm AGI CPU packs up to 136 Neoverse V3 cores per chip, 6 GB/s memory bandwidth per core, a 300 W TDP with one core per program thread, and designs targeting high-density air- and liquid-cooled racks. Arm says the chip delivers more than 2x performance per rack versus x86 for agentic AI, aiming to address rising CPU demand from always-on AI agents and offer partners options beyond IP licensing and compute subsystems. The move signals Arm’s deeper push into cloud infrastructure and competition with x86 server incumbents.

Arm announced on March 24, 2026 that it will build production silicon for the first time, unveiling the Arm AGI CPU — an Arm-designed data center processor targeting agentic AI workloads. Developed with lead partner Meta and backed by ODM commitments, the AGI CPU packs up to 136 Neoverse V3 cores, high per-core memory bandwidth (6GB/s), a 300W TDP with one core per program thread, and configurations for dense air- and liquid-cooled racks. Arm says the chip delivers over 2x performance per rack versus x86 for agentic AI token processing and aims to simplify infrastructure by reducing x86 complexity. The move broadens Arm’s platform options (IP, CSS, and silicon) and signals deeper vertical integration to meet surging CPU demand from persistent AI agents.

Arm announced the Arm AGI CPU, its first Arm-designed production silicon, targeting rack-scale agentic AI infrastructure by optimizing per-thread performance, memory bandwidth, and I/O for sustained, massively parallel workloads. Built on the Neoverse platform, the 1OU, 2-node reference blade packs 272 cores and scales to 8,160 cores per air-cooled rack or over 45,000 cores in a Supermicro liquid-cooled 200kW system, claiming more than 2x rack-level performance versus latest x86 platforms. Arm positions the CPU to meet demands of continuous, agentic AI that coordinates thousands of tasks and accelerators, and highlights partners and hyperscalers already using Neoverse-based solutions. This represents a strategic shift as Arm moves from IP licensor to silicon provider for AI datacenter deployments.

Arm has announced the Arm AGI CPU, its first Arm-designed production silicon built on the Neoverse platform to power rack-scale, agentic AI infrastructure. Aimed at workloads that run continuously and coordinate thousands of software agents, the chip focuses on sustained per-task performance, memory bandwidth, I/O, and energy efficiency across dense rack deployments. Reference designs include a 1OU, 2-node blade with 272 cores (8,160 cores per 36kW air-cooled rack) and a Supermicro liquid-cooled variant supporting over 45,000 cores per rack. Arm says the design delivers more than 2x rack-level performance versus current x86 servers by improving usable threads, single-thread work, and memory bandwidth, and is already gaining early ecosystem partner momentum.