AI inference servers shrink to fit standard air-cooled racks

Executive Briefing
“Axelera AI introduced low-power inference chips and server designs that allow enterprises...”
Key Takeaways
- 1.According to Manuel Botija of Axelera AI, enterprises are locked out of owning AI infrastructure due to excessive costs and power requirements.
- 2.Axelera unveiled its third-generation Titania architecture featuring a 400-watt chip, resulting in a six-kilowatt server fitting standard 20- to 30-kilowatt air-cooled racks.
- 3.The technology behind the company's power-efficient chips relies on digital in-memory computing.
Axelera AI introduced low-power inference chips and server designs that allow enterprises to own AI hardware without costly liquid cooling systems.
Product Features & Key Highlights
According to Manuel Botija of Axelera AI, enterprises are locked out of owning AI infrastructure due to excessive costs and power requirements.
Axelera unveiled its third-generation Titania architecture featuring a 400-watt chip, resulting in a six-kilowatt server fitting standard 20- to 30-kilowatt air-cooled racks.
The technology behind the company's power-efficient chips relies on digital in-memory computing.
Practical Assessment & Evaluation
Integrating low-power inference machines into existing air-cooled racks lowers upfront investment barriers, allowing companies to avoid renting cloud compute.
Pricing, Availability & Next Steps
The real-world performance and durability of the Titania server under heavy enterprise AI workloads remain independently unverified.







