Google launches Axion, its custom Arm CPU for data centers and updated AI chip
Google has announced the launch of its first custom-built Arm processor for data centers, the Axion, based on Arm’s Neoverse V2 designs. This new processor is said to provide a 30% performance improvement over the fastest general-purpose Arm-related instances available in the cloud, with Google already deploying Axion to power several of its first-party services such as BigTable, Spanner, BigQuery, Blobstore, Pub/Sub, and YouTube Ads platform.
In addition to Axion, Google introduced its TPU v5p, the latest iteration of its data center artificial intelligence chips, designed to run in large-scale configurations known as pods, containing up to 8,960 chips. This new TPU generation is reported to have double the raw performance of its predecessors, signaling a significant step in Google's hardware development efforts.
Google's move to custom processors like Axion could increase competition in the data center processor market, particularly with Amazon, which has been a pioneer in the deployment of Arm-based processors for cloud services. The Axion's reported energy efficiency—a 50% improvement over current x86-based virtual machines—aligns with industry concerns about the growing energy demands of AI data centers, which are predicted to account for a significant portion of US power consumption by the end of the decade.
The announcement of Axion and TPU v5p reflects Google's ongoing investment in its in-house chip capabilities, an initiative that complements its existing TPU and mobile chip developments. While specific details on availability dates, pricing, and technical specifications for Axion were not disclosed, Google has signaled its intention to make these processors available through Google Cloud, potentially later this year.
