Urgent.News

What's breaking now, across thousands of outlets.

AI

AI inference servers shrink to fit the air-cooled racks enterprises already run

Power-efficient inference chips are becoming the entry price for enterprises that want to run AI on hardware they own. As AI shifts from building models to running them every day, costly graphics processing unit racks and the liquid cooling they demand keep many companies renting compute instead. That gap is opening room for challengers to […] The post AI inference servers shrink to fit the…

AI inference servers shrink to fit the air-cooled racks enterprises already run

Power-efficient inference chips are becoming more accessible to enterprises looking to run AI on hardware they own. As AI moves from model development to daily operations, expensive GPU racks and their associated liquid cooling costs deter many companies from purchasing compute. This has created an opportunity for competitors to the major chipmakers.

Axelera AI B.V., which specializes in low-power chips and edge AI development tools, is now expanding its technology into data center infrastructure. Manuel Botija, Axelera's vice president and head of product, explained to theCUBE Research's John Furrier and Rebecca Knight that enterprises are often locked out of owning their own AI infrastructure due to the high costs and power requirements of GPU racks.

Axelera's solution, the Europa chip, was introduced in September and has since been followed by Titania, a third-generation architecture that can scale up to a 400-watt chip. This results in a six-kilowatt server that can perform like leading data center GPUs while fitting into the existing 20- to 30-kilowatt air-cooled racks commonly found in enterprise data centers.

Botija emphasized that having a six-kilowatt server with data center-level performance is a significant achievement, as it allows enterprises to add AI inference hardware acceleration to their existing server room or data center without needing to rebuild or invest in new infrastructure. For businesses preferring to own their AI hardware rather than pay recurring cloud bills, the barriers to entry have been significant.

Axelera's approach relies on digital in-memory computing, a technique developed by its founders, which provides a performance-per-watt advantage. Botija emphasized that with Axelera's power-efficient inference chips, enterprises can now own their AI infrastructure, amortize investments, reduce operational expenses, and ultimately increase profitability.

Written by urgent.news from SiliconANGLE's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at siliconangle.com →

More in AI

Anthropic Leveraged ETF Race Heats Up

Bilal Little, global ETF strategist at Direxion, joins Scarlet Fu on a special edition of “Bloomberg Markets.” Providers have filed for more than 20 exchange-traded products to amplify daily moves on…

More from Thursday 19 November →