Every cooling system has a heat exchanger in it, and most of them are built the way they were decades ago: thin metal plates stamped into shape and brazed together. That process limits the shapes a designer can make, and the shape limits how much heat the part can move in a given space. AI data centers need more heat moved in less space than that process was ever meant to deliver.
So we were pleased to see 3D Printing Industry's report on a Korean research consortium taking the same view. LincSolution and a group led by Chung-Ang University, with Kookmin University, the Korea Institute of Machinery and Materials, LG Electronics, and Michigan State University, have started a three-year project on 3D printed heat exchangers for AI data center cooling and electric vehicles. It runs from July 2026 through June 2029 with 3.4 billion won in funding, and the goal is mass production: a printing foundry in Daejeon is meant to be fully running in 2027. Metal 3D printing, as the article puts it, makes it possible to build tiny internal channels and combined structures as a single part, with no seams to braze.
A heat exchanger works by putting the fluid in contact with as much surface as possible. Smaller channels mean more surface in the same volume, so the part moves more heat for its size, and printing makes channels far smaller and more intricate than stamping can. LincSolution's CEO, Choi Geun-sik, ties the project to the energy transition and the growth of AI, and we'd say the same.
We agree with the direction. Our questions are two the article doesn't raise.
Small channels are harder to push fluid through
Smaller channels catch more heat, but the fluid has to work harder to get through them, and that shows up as pump or compressor energy for the life of the unit. A printed exchanger has to be designed for both at once, or it wins on heat transfer and loses on the electricity bill. The article doesn't mention flow resistance, and any operator comparing a printed part to a stamped one should ask about it.
The second question is cost per part. A stamped plate is cheap because the tooling is paid off. A printed plate carries machine time and metal powder on every unit, and no one will know whether that number falls below the value of the better shape until the foundry is running.
Where we come in
We build printed heat exchangers now, at data center scale. Our 1 MW heat exchanger for the Open Compute Project's Deschutes cooling design is built as two 500 kW units, and we'll show it at the OCP Global Summit in San Jose, October 12 to 15, 2026. Its channels are shaped to move heat well and let coolant flow freely at the same time. Printing lets us choose the shape of every channel for heat transfer and flow, instead of accepting the shapes a stamping die can make.
What the industry should ask
The question for every printed heat exchanger, ours included, is when the cost of the part falls below the value of the shape. We think that point has arrived for the densest racks, and those are the racks we build for.