Nvidia says its AI data center design runs hotter to use a lot less water - The Verge

NVIDIA has announced that its next-generation AI data center design uses a fully liquid-cooled system and operates servers at higher temperatures to achieve near-zero water usage. This approach reduce... Read original →

NVIDIA’s shift to high-temperature, fully liquid-cooled AI data centers isn’t just about water savings—it’s a strategic redefinition of compute infrastructure. This move pressures upstream thermal interface and chip packaging suppliers to accelerate high-conductivity material innovation, while forcing data center builders to recalibrate CAPEX assumptions as air-cooling’s cost advantage erodes. In water-stressed regions like California, it mitigates regulatory risk, yet high upfront costs may deter smaller cloud operators. Intel is countering by pushing open liquid-cooling standards with ODMs to avoid ecosystem lock-in, while Google and Meta may double down on co-designed accelerators (TPU/MTIA) with integrated thermal management. Within 18 months, liquid cooling will transition from premium feature to baseline requirement for AI clusters, spurring demand for Taiwan, China-based thermal module makers and European heat-exchanger vendors—and exposing a new bottleneck: without local waste-heat utilization, the promised efficiency gains collapse under grid integration constraints.