As demand for computing power and artificial intelligence grows and computer chips become more advanced, the data centers that house all these electronics are getting hotter. Nearly 40% of the energy consumed by a data center goes toward cooling chips and servers.
Researchers have now created a new liquid-cooling technology that could slash that share to only 1% of a data center’s energy use (Cell Rep. Phys. Sci. 2026, DOI: 10.1016/j.xcrp.2026.1).
Most data centers rely on blowing chilled air over computer chips. But air cannot efficiently siphon away the intense heat today’s advanced AI chips produce. Liquids are better at capturing and removing heat.
As a result, data centers are now starting to switch to a technology called direct-to-chip cooling. This involves placing a copper cold plate with an array of circular, rectangular, or square micropillars, called fins, over a computer chip. Pumping a coolant—typically a mix of water and propylene glycol—through the fins helps channel heat away from the chip. But the performance of cold plates is limited because of their simple design, says Nenad Miljkovic, a mechanical science and engineering professor at the University of Illinois Urbana-Champaign.
Miljkovic and his team used a computational technique to design a fin shape that would let fluid flow through the cold plate most efficiently. “We focused on minimizing pressure drop and maximizing heat transfer from the [cold plate],” which cuts the energy needed to pump liquid through the plate, he says.
The optimized shape resembles the jagged spikes depicted on a fictional monster’s back. “We call them Godzilla fins,” Miljkovic says.
The spikes are about 40 µm wide and 100 µm tall. To make the small, complicated metal features, the team partnered with Fabric8Labs to use the start-up’s new 3D-printing technology, called electrochemical additive manufacturing (ECAM).
Traditional metal 3D-printing processes like laser powder bed fusion require heat, lasers, or expensive metal powders. They work well for making large aerospace and medical parts. But for manufacturing on-chip cold plates, “they have not been successful due to the high temperatures and inability to fabricate fine features,” says Yogendra Joshi, who studies the heat management of electronics at the Georgia Institute of Technology and was not involved in the work.
Instead, ECAM uses electrodeposition. In classical electrodeposition, you apply voltage across two electrodes immersed in an electrolyte containing metal salts, which causes metal to deposit on one side, Miljkovic explains.
In ECAM, the plating electrode is an array of millions of individual pixels, each a few micrometers in size. The machine applies a voltage at each pixel to deposit copper ions at that pixel, and moves the plate up slowly to print ultrahigh-resolution metal parts layer by layer.
The researchers made a 2 cm x 2 cm cold plate of the newly designed copper fins. They then used it, as well as conventional cold plates, for liquid cooling of gallium nitride chips. A comparison of the results shows that the optimized plate transferred heat up to 32% more effectively and reduced the pressure drop across the plate by 68%. The team’s calculations show that for an entire data center, this would translate to significant energy savings over air cooling and commercially available liquid-cooling systems.
Joshi calls it “an innovative and promising approach worthy of further examination.” Whether it will be cost-effective at large scale remains to be seen, he says.
But Miljkovic says that even if the plates are expensive to manufacture, the energy savings for data centers would more than make up for it.
