Arm Holdings, the British semiconductor giant that has spent nearly 36 years licensing chip designs to the world's biggest technology companies, has officially entered the hardware business. The company unveiled the Arm AGI CPU at an event in San Francisco on Tuesday a production-ready processor built for running AI inference workloads in data centres.
It is the first chip Arm has ever manufactured itself, marking a historic departure from a business model that powered everything from iPhones to Nvidia GPUs without the company ever producing its own silicon.
Meta is the chip's first customer. OpenAI, Cerebras, and Cloudflare are among the additional launch partners.
What the Chip Does
The Arm AGI CPU was developed using the company's Neoverse family of CPU intellectual property cores and built in partnership with Meta. It is designed to work alongside Meta's existing training and inference accelerators inside AI data centres.
While GPUs have dominated headlines as the engines behind AI model training, Arm is making its case for the CPU. The company argues that CPUs remain a critical and underappreciated part of data centre infrastructure. These processors manage thousands of distributed tasks handling memory and storage, scheduling workloads, and moving data across systems. Arm describes the CPU as the pacing element of modern infrastructure, responsible for keeping distributed AI systems running efficiently at scale.
The company says the growing demands of AI workloads require an evolution of the CPU, and that the Arm AGI is purpose-built to meet that challenge.
Years in the Making
The move did not happen overnight. According to reporting from CNBC, Arm began developing the chip back in 2023, and the processors are already available to order. The company had been quietly signalling a shift toward in-house production for some time, but Tuesday's announcement made it official.
A spokesperson confirmed the chip is production-ready but declined to share further details about the development timeline or pricing.
Why It Matters
For more than three decades, Arm operated as the architecture behind other companies' chips. Its designs are found in virtually every smartphone on the planet and increasingly in laptops, servers, and data centre hardware. Apple, Qualcomm, Samsung, and Nvidia are among its most prominent licensees.
By manufacturing its own silicon, Arm is now stepping into a competitive position alongside many of its own partners and customers. That is a significant strategic gamble for a company whose entire business model was built on neutrality providing the blueprints without competing for the finished product.
The company, which is majority owned by Japanese conglomerate SoftBank Group, appears to be betting that the AI infrastructure boom is large enough to justify the risk. With demand for data centre hardware surging and AI workloads growing exponentially, there is room for new entrants even one that has historically stayed out of the manufacturing game.
A CPU Play Amid GPU Hype
Arm's decision to launch a CPU rather than a GPU is a deliberate strategic choice. The AI hardware conversation has been overwhelmingly focused on GPUs, driven by Nvidia's dominance in training large language models. But CPUs play an equally essential role in data centre operations, handling the orchestration and data movement that keeps AI systems functioning.
The timing also works in Arm's favour. A growing global CPU shortage has been pushing prices higher and delivery times longer. Earlier this year, Intel and AMD reportedly informed their Chinese customers that wait times for CPU products would increase significantly. Computer prices have also begun to rise as a result of the tightening supply.
Arm's entry into the market with a purpose-built AI inference chip could help ease some of that pressure, particularly for hyperscale customers like Meta that are racing to build out massive AI infrastructure.
What Comes Next
The Arm AGI CPU is clearly a first step rather than a final destination. With Meta as its anchor customer and a roster of launch partners that includes some of the most prominent names in AI, Arm is positioning itself not just as a chip designer but as a full participant in the AI hardware race.
Whether this move alienates existing licensing partners or attracts new ones will be one of the most closely watched dynamics in the semiconductor industry over the coming months. For now, Arm has made its intentions clear: after 35 years of designing chips for everyone else, it is finally building for itself.
