To address the high heat challenges posed by high-performance AI hardware such as Nvidia H100, Meta has deployed a new closed-loop liquid cooling system in its AI infrastructure in Texas. This system uses a mixture of water and ethylene glycol to flow through servers to remove heat, which is then cooled by a heat exchanger before being recycled for reuse, with minimal water consumption. Compared to traditional air cooling, this solution not only significantly reduces annual water usage (even lower than that of several full-service restaurants), but also reduces the space required for data centers. For locations lacking infrastructure, Meta has also deployed compact air-assisted liquid cooling systems to supplement heat dissipation capabilities.
Meta uses closed-loop liquid cooling technology in its AI infrastructure in Texas to cope with the high heat generated by AI hardware. Traditional data centers typically use air cooling, but new AI hardware (such as Nvidia H100) has reduced the efficiency of simple air cooling. Meta’s new generation of AI-optimized data centers primarily utilize closed-loop liquid cooling systems, where a mixture of water and ethylene glycol is used to remove heat from servers, and then re-used after being cooled by heat exchangers, with minimal water consumption. This system not only achieves high resource efficiency, with annual water usage even lower than that of several full-service restaurants, but also reduces the space required for data centers. For locations lacking infrastructure, Meta has also deployed compact air-assisted liquid cooling systems.