Arm Holdings, a renowned semiconductor and software company, is beginning to produce its own chips after around 36 years of licensing its designs to firms like Nvidia and Apple.
During an event in San Francisco, the company introduced the Arm AGI CPU, a chip designed for running inference in AI data centers. This chip, developed using Arm Neoverse CPU IP cores, was made in partnership with Meta.
Meta is the first customer for the Arm AGI CPU, which is built to integrate with its training and inference accelerator. Arm has also partnered with companies like OpenAI, Cerebras, and Cloudflare.
The shift towards producing its own silicon has been anticipated. The company began this project in 2023, and the processors are now available to order, as reported by CNBC.
TechCrunch contacted Arm for more details on the timeline of the chip’s development and release.
This move marks a significant shift from Arm’s tradition of solely licensing its designs. Now, the company, largely owned by SoftBank Group, will compete with many of its existing partners.
The production of a CPU instead of a GPU is noteworthy, as GPUs are widely used for training and running AI models. CPUs play a critical role in data center racks.
Arm highlights that CPUs handle numerous tasks like managing memory, scheduling workloads, and moving data, making them crucial in modern infrastructure for efficient AI system operations.
This evolution places new demands on CPUs. Additionally, CPUs are becoming scarce.
In March, Intel and AMD informed Chinese customers about longer wait times because of CPU shortages, leading to increased computer prices as reported by Reuters.