Meta and Arm are now building CPUs for AI data centers

AI data centers are appearing left and right, and so the demand for the hardware powering them is at an all-time high. To comply with the industry demand, Meta and Arm are now teaming up to create a brand-new class of processors. These new CPUs are designed to handle heavy AI workloads in data centers.

The name of the CPU line is also pretty interesting - Arm AGI CPU. Everybody’s been talking about achieving AGI lately, so Arm choosing this name for data center processors is either a deliberate marketing play, or the company really believes its processors will help AI researches achieve artificial general intelligence.

If you’re accustomed to reading specsheets of consumer-level CPUs, you’re going to be in awe when you see the power of these CPUs:

• Performance: Up to 136 Arm Neoverse V3 cores per CPU, delivering leading performance per core, SoC, blade and rack*, with 6GB/s memory bandwidth per core at sub-100ns latency.
• Scale: 300-watt TDP with a dedicated core per program thread enables deterministic performance under sustained load, eliminating throttling and idle threads.
• Efficiency: Supports high-density 1U server chassis that supports air-cooled deployments with up to 8,160 cores per rack, and liquid-cooled systems delivering 45,000+ cores per rack.

These are high-end specs even by data center-level standards. Meta and Arm are emphasizing that the biggest selling point of AGI CPUs is efficiency. Arm claims the AGI CPU delivers more than double the performance per rack compared to traditional x86 alternatives, all while operating comfortably within existing power limits.

Meta’s role in this deal is providing infrastructure expertise. The company already runs huge data centers just to power its own tools, with a diverse hardware portfolio inside them. The AGI CPU will work alongside Meta's own MTIA (Meta Training and Inference Accelerator) custom silicon.

Both companies have made it clear that this is just the beginning of their partnership, and they are already looking for more ways to collaborate down the road.

Source: Meta | Arm