
Solid-State Cooling Uses Waste Heat, Not Electricity
A KIT and Tsukuba prototype cools using only waste heat, hitting a 13 K span in the refrigerant with zero electrical input. Published in Nature Energy.
A Cooling System With No Electricity Going In
Researchers at the Karlsruhe Institute of Technology and the University of Tsukuba have built a solid-state cooling prototype that runs on heat alone. No compressor, no refrigerant gas, and — this is the part worth pausing on — no electrical input at all. The work was published in Nature Energy and reported on August 28, 2026.
- Two ultrathin nickel-titanium shape-memory films do all the work: one converts heat into mechanical motion, the second uses that motion to produce cooling
- Roughly 13 K of cooling in the elastocaloric refrigerant, with about 4 K measured at the device level, driven by an actuator held at 86°C
- The system also ran from a 130°C external heat source, the temperature range where industrial and computing waste heat actually sits
- No compression cycle and no fluorinated refrigerants, the two things that make conventional cooling both power-hungry and environmentally awkward
Why Elastocaloric Cooling Matters for Data Centers
Conventional refrigeration works by compressing and expanding a gas, which requires a motor, which requires electricity. Elastocaloric cooling works on a different physical principle entirely: certain shape-memory alloys change temperature when they are mechanically stressed and released. Stretch the alloy and it warms. Let it relax and it cools below where it started.
The clever move in this design is where the mechanical stress comes from. Previous elastocaloric devices still needed a motor to do the stretching, which put electricity back into the loop. The KIT and Tsukuba team used a second nickel-titanium film as the actuator — a film that deforms when heated. Feed it waste heat and it becomes the motor. The energy that drives the cooling is heat that was already there and already being thrown away.
For anyone following the energy arithmetic behind modern compute, that framing lands immediately. Cooling is one of the largest non-compute line items in a data center's power budget, and it scales with the very thing you are trying to cool. Our AI infrastructure coverage has tracked a steady run of efficiency work on the compute side, from Nvidia's Vera Rubin NVL72 targeting 30x tokens per watt to OpenAI's Jalapeño benchmarks claiming 1.9x more work per kilowatt. A cooling loop that consumes waste heat instead of grid power attacks the problem from the opposite end.
What Can 4 K at the Device Level Actually Do?
Honesty first: a 4 K device-level temperature difference is a laboratory result, not a product. It is not going to chill a rack today, and the researchers do not claim it will. What the number demonstrates is that the physics closes — that a purely heat-driven elastocaloric loop produces real, measurable cooling with the electrical input set to zero.
That is the hard part in this class of research. Getting a proof of principle from "theoretically possible" to "measured on a bench" is where most passive-cooling concepts stop. The team, led by Dr. Jingyuan Xu at KIT's Institute of Microstructure Technology with Yi-Ting Hsiau as lead author, is now working on scaling the architecture into compact systems.
Where a Waste-Heat Cooler Would Go First
The applications KIT names are pleasingly specific. Processor cooling that runs on the processor's own waste heat is the obvious one, and it has a satisfying circularity to it — the hotter the chip runs, the harder its cooler works. The second is automotive electronics cooled by drivetrain heat, which is the same idea applied to a system that already produces enormous quantities of low-grade heat with nowhere useful to put it.
Both share a property that makes this approach attractive well beyond the headline: they are places where a heat source is guaranteed to be present and is currently a liability. Turning that liability into the power supply for the cooling loop is a genuinely elegant piece of engineering, and it is the kind of compounding efficiency gain that tends to show up in edge hardware first — the fanless and thermally constrained designs where every watt of cooling budget is contested.
Scaling remains the open question, and the researchers say so plainly. But the direction of travel here is worth watching, because a cooling technology whose input is the problem it solves is a rare shape for an efficiency story to take.
Sources: Karlsruhe Institute of Technology — August 2026; The Register — August 28, 2026; Nature Energy — August 2026.
More AI Stories

Claude Desktop Now Runs Local Models Through Ollama
Ollama's new Claude Desktop integration is one toggle: local or cloud open models appear in Claude's picker, with the full agent toolset intact.

Qwen3.8-Flash-Next Fits in 75GB With No GPU Needed
Qwen's 125B Flash-Next MoE runs locally in 75GB of RAM with no GPU VRAM required, scoring 62.5 on SWE-bench Pro with just 6B active parameters.

Model Hardware Standard Lets AI Agents Run Lab Gear
Anthropic's Model Hardware Standard cuts lab instrument integration from weeks to minutes, with Genentech, Carnegie Mellon and QuEra among first users.
