Microsoft tests microfluidics to cool AI chips: What it is, how it works, and more

Microsoft is testing microfluidic cooling technology as it seeks to manage the growing heat output of chips powering AI data centres. As demand for AI workloads surges, data centres have faced significant challenges in keeping processors and GPUs cool while maintaining efficiency. Microfluidics, which pushes liquid through microscopic channels etched into chips, may provide a solution.

Traditional liquid cooling systems circulate fluid across the surface of chips, but microfluidics allows liquid to flow much closer to hotspots inside the silicon. This can manage chip temperatures up to 70°C while enabling sustained high performance. Microsoft has already demonstrated prototypes at its Redmond campus, where the approach showed marked improvements compared with existing methods.

The technology also allows for new design possibilities. By cooling more efficiently, chips can be stacked vertically, increasing power density in data centres. This makes it possible to achieve higher performance in the same footprint. Additionally, microfluidics could enable controlled overclocking during temporary demand spikes, such as when millions of users join Teams calls simultaneously.

Microsoft is exploring other hardware innovations in parallel. These include hollow-core fibre for faster networking and next-generation memory, particularly high-bandwidth memory (HBM), which can accelerate AI training and inference. Together, these innovations reflect the company’s attempt to balance performance expansion with energy efficiency.