AWS Adopts Liquid Cooling and Simplified Power for AI Servers
AWS announces a major shift to liquid cooling for its AI servers, including those using Nvidia hardware and its own Trainium2 chips. The company also reveals simplified electrical designs that dramatically boost data center efficiency and reliability.
AWS introduces liquid cooling for its AI servers to handle the massive heat generated by intensive workloads. This new cooling approach applies to both AWS Trainium2 chips and Nvidia's GB200 NVL72 systems. The company designs these updated systems to seamlessly integrate both air and liquid cooling, ensuring that traditional networking and storage servers continue to function efficiently alongside the high-performance AI hardware.
In addition to cooling upgrades, AWS simplifies the electrical and mechanical designs inside its data centers. By reducing the number of power conversions between the electrical grid and the servers, the company significantly cuts energy loss. These streamlined systems achieve an impressive infrastructure availability of 99.9999% and reduce the number of racks affected by electrical issues by 89%.
These infrastructure improvements represent a strategic push to support the growing demands of AI while maintaining sustainability and lowering costs. AWS emphasizes that the new data center capabilities are modular, allowing the company to scale rapidly as emerging workloads evolve. The updates position AWS to deliver maximum performance and resilience for customers running both traditional cloud applications and advanced AI models.