Arm has officially launched its AGI CPU, a significant development that marks the company's first foray into producing its own silicon rather than solely licensing its chip designs. This strategic pivot positions Arm to directly address the surging demand for specialized processors in the burgeoning agentic AI infrastructure market.
The Arm AGI CPU, co-developed with lead partner Meta, is purpose-built for agentic AI workloads in data centers. According to Arm, the new chip offers "more than 2x performance per rack compared with x86 platforms," aiming to deliver substantial CAPEX savings for AI data center operators. Meta plans to deploy the AGI CPU alongside its custom MTIA accelerators, with Santosh Janardhan, Head of Infrastructure at Meta, stating, "We worked alongside Arm to develop the Arm AGI CPU to deploy an efficient compute platform that significantly improves our data center performance density."
Built on TSMC's 3nm process, the AGI CPU features up to 136 Neoverse V3 cores, a maximum frequency of 3.7GHz, and a 300W TDP. It supports 12 channels of DDR5 memory and PCIe Gen6 with CXL 3.0. This move represents a profound shift in Arm's 35-year business model, expanding its compute platform beyond intellectual property (IP) and compute subsystems into production silicon.
The AGI CPU has already garnered strong industry support, with companies like OpenAI, SK Telecom, SAP, Cloudflare, F5, and Cerebras listed as early customers and partners. Synopsys also played a crucial role, providing full-stack design solutions for the CPU's development. This strategic shift underscores Arm's ambition to capture a larger share of the high-growth AI market by offering integrated hardware solutions.
