Google Cloud Touts Axion Processors for Energy Efficient Data Centers, AI Training Performance

Amid growing concerns about generative AI's energy needs, Google Cloud plans to roll out energy efficient Axion processors sometime in 2024, the search giant announced during the Google Cloud Next 2024 conference in Las Vegas.

Exact availability was not disclosed. Also, the company does not plan to sell the processors to third-party data center companies or server makers, multiple reports said.

Axion processors, based on ARM technology, will arrive at a critical time. Indeed, Google Cloud and its major rivals -- namely, Amazon Web Services and Microsoft Azure, among others -- have been exploring various chip alternatives to Nvidia, which dominates the processor market for running AI workloads.

To be clear: Google continues to expand its Nvidia partnership. But in theory, Axion processors could reduce Google's long-term dependence on high-priced Nvidia hardware.

Generative AI Applications, Data Centers and Energy Consumption

Meanwhile, U.S. electric utilities predict a tidal wave of new demand from data centers powering technology like generative AI, with some power companies projecting electricity sales growth several times higher than estimates just months earlier, Reuters reported in April 2024.

Some customers and channel partners, meanwhile, will increasingly pursue energy efficient data centers in order to address Scope 3 (i.e., supply chain) carbon emissions.

Amid that backdrop, Google claims the Axion processors will deliver:

• Instances with up to 30% better performance than the fastest general-purpose Arm-based instances available in the cloud today;
• up to 50% better performance and up to 60% better energy-efficiency than comparable current-generation x86-based instances. 

Google Cloud Services, Data Centers: Leveraging ARM

Ahead of the Axion processor arrival, Google has started to deploy cloud services such as BigTable, Spanner, BigQuery, Blobstore, Pub/Sub, Google Earth Engine, and the YouTube Ads on current generation ARM-based servers, the company said.

Those services will eventually move over to Axion processors -- but here again, the company did not disclose a specific timeframe for the migrations.

Meanwhile, Google Cloud data centers are already 1.5X more efficient than the industry average and deliver 3X more computing power with the same amount of electrical power compared with five years ago, the company said in a blog. "With Axion processors, customers can optimize for even more energy-efficiency," wrote Amin Vahdat, VP and GM for machine learning, systems and Cloud AI at Google.

Microsoft, AWS, Meta Pursue Nvidia Alternatives

Meanwhile, major cloud service providers and social media companies have focused their R&D efforts on Nvidia alternatives. Example developments include:

• April 2024: Facebook parent Meta is developing a new generation of MTIA infrastructyre for AI workloads.
• February 2024: Microsoft is developing a new network card that could improve the performance of its Maia AI server chip and potentially reduce the company's reliance on Nvidia, The Information reported and Reuters further analyzed.
• November 2023: Amazon updated two AWS-designed chip families—AWS Graviton4 and AWS Trainium2—to further boost "price performance and energy efficiency for a broad range of customer workloads, including machine learning (ML) training and generative artificial intelligence (AI) applications."

Meanwhile, we expect more potential chip and sustainability updates at the Microsoft Build 2024 conference in May. Stay tuned.