TECH – Arm is making an unusually bold move in the data-center market: instead of simply licensing its processor designs to other companies, the chip designer is now introducing a complete server CPU of its own. According to Jagat Review, the Arm AGI CPU is designed specifically for modern AI data centers, where increasingly complex workloads are putting new demands on conventional server processors.
For more than three decades, Arm has primarily supplied CPU architectures that companies such as Qualcomm and MediaTek turn into their own chips. The AGI CPU changes that business model by giving Arm a finished processor aimed at generative AI and agentic AI infrastructure. The shift also reflects a change in how data centers handle AI workloads. GPUs remain crucial for intensive model computation, but CPUs are increasingly responsible for the surrounding tasks, including database access, API calls, orchestration and business applications.
The AGI CPU is manufactured by TSMC using a 3nm process and can feature as many as 136 Neoverse V3 cores running at up to 3.7GHz. Arm is using a chiplet-based approach to build the processor, allowing multiple computing components to work together rather than relying on one enormous piece of silicon. At Hot Chips 2026, further technical details showed that the AGI design uses two chiplets, with each chiplet containing 70 CPU cores connected through Arm’s CMN fabric.
Read More: Intel Xeon Diamond Rapids Packs 256 Cores and 1.28GB of Cache
The scale becomes even more interesting at the rack level. A server configuration can reportedly accommodate as many as 64 AGI CPUs, creating a huge pool of Arm-based processing power for cloud and AI workloads. Jagat Review says Arm claims the platform can achieve up to twice the efficiency of comparable x86 systems, although real-world performance will depend heavily on software, workload and system configuration.
Meta is reportedly the first customer and co-developer of the processor, while OpenAI, Cloudflare and AWS have also shown interest. The AGI CPU is not intended to operate in isolation, either; it can work alongside accelerators and platforms from companies including NVIDIA, Intel and AMD.
Arm has already outlined plans for future generations of the platform, suggesting this is more than a one-off experiment. With AI infrastructure becoming increasingly heterogeneous, Arm’s decision to build complete server silicon could give data-center operators another route beyond the traditional x86 model. The server CPU battlefield just gained a rather formidable new contestant.
