Arm Enters Merchant Silicon Market With AI Data Center CPU To Challenge x86

Between your smartphone, smart TV, wireless router, and a plethora of other gadgets in your home, there's a near 100% chance at least one of your devices is running silicon based on an Arm design or Arm IP. There's also a 0% chance that Arm actually produced the silicon, because its model has strictly been focused on IP licensing since the early 1990s. That all changes today, however, with the introduction of Arm's first production silicon, co-developed with Meta.

The announcement comes less than a year after reports surfaced of Arm poaching Amazon's AI chip director, to advanced its in-house chips plans. At the time, we noted that the hiring signaled Arm was likely getting more serious about offering its own hardware, and that turned out to be correct.

Arm joined forces with Meta on the chip design to optimize infrastructure for the social network giant's family of apps, as well as work alongside its own Meta Training and Inference Accelerator (MTIA) silicon. According to Arm, the duo will collaborate on several future generations of its AGI CPU.

This is a big deal on a number of levels, not the least of which is the big threat Arm's latest flex poses to x86 incumbents AMD and Intel. Arm isn't pivoting away from its IP licensing model, but expanding its reach by finally selling physical chips to the public.

Why now? The data center has always been a high-margin market, but it's never been more red hot than it is right now as big tech rides the AI wave to unprecedented demand for silicon, including not just processors and graphics cards, but also memory and storage.

"AI has fundamentally redefined how computing is built and deployed. Agentic computing is accelerating that change," said Rene Haas, CEO, Arm. "Today marks the next phase of the Arm compute platform and a defining moment for our company. With the expansion into delivering production silicon with our Arm AGI CPU, we are giving partners more choices all built on Arm’s foundation of high-performance, power-efficient computing, to support agentic AI infrastructure at global scale."

The new AGI CPU is comprised of up to 136 Neoverse V3 cores per chip. Arm is claiming its design delivers leading performance per core, SoC, and rack, with 6GB/s of memory bandwidth per core at sub-100ns latency.

Arm's reference server configuration is a 1OU, 2-node design with dual chips, dedicated memory, and I/O for a total of 272 cores per blade. The company says these blades can fully populate a standard air-cooled 36kW rack, with 30 blades delivering 8,160 cores. Arm also says it's partnered with Super Micro on a liquid-cooled 200kW design that can house 336 AGI CPUs with over 45,000 cores.

"In this configuration, the Arm AGI CPU is capable of delivering more than 2x the performance per rack compared to the latest x86 systems, achieved through the fundamental advantages of the Arm architecture and careful matching of system resources to compute," Arm says.

Other key stats include a 300W TDP with a dedicated core per program thread to enable deterministic performance under sustained load (Arm claims this eliminates throttling and idle threads), dedicated 2MB of L2 cache per core, up to a 3.7GHz clock speed, and support for high-density 1U servers.

Arm is calling the launch of its AGI CPU a new chapter in its data center journey, and it's lined up some heavy hitters beyond just Meta, including "more than 50 leading companies" representing hyperscale, cloud, silicon, memory, networking, and several other categories.