Arm unveils 3nm AGI CPU, powered by Meta’s data center
Arm Takes a Bold Step into the Chip Market
For decades, Arm has been the quiet powerhouse behind most of the world’s chips. It provided designs and collected royalties, but now, the company is stepping out from the shadows to create its own processor. This marks a significant shift in strategy as Arm enters the competitive chip market with its new 3-nanometer AGI CPU, specifically designed for data centers.
The launch of the AGI CPU puts Arm directly up against some of the same companies that have built products using its architecture. At an event in San Francisco, CEO Rene Haas introduced the new processor, highlighting its focus on AI workloads. This move signifies Arm’s transition from being a supplier of foundational designs to offering its own silicon solutions.
Meta Joins Arm in Expanding AI Infrastructure
Meta is the first customer to partner with Arm on this initiative. The social media giant is investing heavily in AI infrastructure, planning to spend up to $135 billion on capital expenditures this year. In February, Meta also secured a large supply of chips from Nvidia and Advanced Micro Devices, making the new Arm deal another source of processors as its AI expansion continues.
Paul Saab, a Meta software engineer who has been involved in the Arm chip project since 2023, noted that “In today’s world, you really only have a couple of players.” He emphasized that the deal offers more flexibility in both the software stack and the supply chain. However, the specific terms of the agreement were not disclosed.
Rising CPU Demand in the Age of Agentic AI
The launch of the AGI CPU comes at a time when CPUs are gaining more attention in the AI sector. Nvidia, which leads the market for AI graphics processors, has previously stated that CPUs are becoming a bottleneck as agentic AI transforms compute needs. Futurum Group described this as a “quiet supply crisis,” predicting that CPU market growth could surpass GPU growth by 2028.
The distinction between GPUs and CPUs is clear. GPUs excel in training and running AI models due to their thousands of cores capable of performing many operations simultaneously. On the other hand, CPUs handle a smaller number of stronger cores designed for general tasks that run sequentially. Agentic AI requires a significant amount of this general compute work, as vast amounts of data move across multiple agents.
At Nvidia’s GTC conference last week, CEO Jensen Huang unveiled an entire rack filled with Vera CPUs. During Arm’s event on Tuesday, Huang appeared in a recorded statement congratulating the company on the launch. Leaders from Google, Amazon, Microsoft, Oracle, Broadcom, Micron, Samsung, SK Hynix, and Marvell also made appearances. Arm reported that around 50 partners had shown support before the launch.
Arm’s Expansion and Reliance on TSMC
To reach this milestone, Arm invested $71 million and spent about 18 months building three new lab rooms at its campus in Austin, Texas. A small team has grown to over 1,000 people, with engineers conducting extensive testing of the chips after they leave the factory.
Like many fabless AI chip companies, Arm relies on Taiwan Semiconductor Manufacturing Company (TSMC) for manufacturing. The AGI CPU is produced on TSMC’s 3-nanometer node, with all production currently based in Taiwan. TSMC is preparing a 3nm fab in Arizona, and while Awad expressed a desire to manufacture there, the final decision depends on customer preferences.
Arm’s Legacy and Future in Data Centers
Arm is best known for powering mobile chips in almost every smartphone. It entered the data center chip market in 2018 with Neoverse, and Amazon helped bring the platform into the mainstream with Graviton. Google and Microsoft also base their AI chips on Arm designs.
Moorhead noted that “If Arm didn’t exist, then all of those companies who have their own processors wouldn’t be able to create their own.” Most server chips still use x86 designs from Intel and AMD, which Moorhead called “tried and true.”
Awad explained that the AGI CPU was “ruthlessly optimized” for artificial general intelligence, which is why it carries that name. One air-cooled rack can hold up to 64 CPUs, or about 8,700 cores. He claimed that the AGI CPU offers “two times the performance-per-watt than you can from an x86 rack.”
Paul highlighted that wattage is a scarce resource, adding that better performance per watt leaves more power available for other parts of the data center.
The smartest crypto minds already read our newsletter. Want in? Join them.
