Cooling copper plates could slash data center energy use by 90%

3 hours ago

Data centers consumed 485 TWh of electricity in 2025, with 30% used for cooling, surpassing Sweden's total annual power consumption.
Scientists developed a 3D-printed copper-plate cooling technology using topology optimization and electrochemical additive manufacturing (ECAM), which can reduce cooling-related electricity use from about 30% to 1.1% in data centers.
AI's high power demand is exacerbated by Joule heating, where chips dissipate nearly all consumed power as heat, making cooling crucial; for example, a single NVIDIA GB200 chip consumes 1,200 watts, equivalent to an average U.S. household's daily use.
Traditional air cooling struggles with modern AI accelerators, leading to a shift toward liquid-based direct-to-chip cooling, but conventional cold plates are limited by simple designs and materials like aluminum alloys.
The new copper cold plates feature complex, optimized internal fin structures that improve cooling performance by up to 32% and reduce pressure drop by 68%, cutting pumping energy requirements.
At scale, the technology could lower cooling overhead in a 1 GW data center from 550 MW with air cooling to around 11 MW, achieving a Power Usage Effectiveness (PUE) close to 1.011, nearing the theoretical ideal of 1.0.
While results are modeled, the approach holds promise for reducing energy costs in AI data centers and could be adapted for broader electronics cooling applications.

Hasty Briefsbeta