Google just dropped its answer to one of the biggest bottlenecks in the AI infrastructure race: keeping the hardware from melting. The company’s new Brazos liquid cooling system is a rack-mounted, closed-loop solution designed specifically for the kind of power-hungry chips that are becoming standard in AI and high-performance computing workloads.
The system can handle thermal loads up to approximately 60kW per rack. For context, that’s roughly enough to cool chips with a thermal design power exceeding 1,000W, which is the territory occupied by the latest generation of AI accelerators.
What Brazos actually does
Brazos is a liquid-to-air system that circulates either deionized water or a coolant mixture containing 25% propylene glycol through the rack. The liquid absorbs heat directly from the chips, then transfers it to air, which gets exhausted from the system.
Because Brazos operates as a closed-loop system at the rack level, data center operators can deploy it in existing air-cooled facilities without requiring facility-wide retrofits.
