Microsoft and Corintis Advance Microfluidics Cooling for More Powerful Chips and Efficient Data Centers

Microsoft has made a significant leap in data center cooling technology with a new microfluidic system that could revolutionize how high-performance chips are cooled. The method involves etching microscopic channels directly into the back of silicon chips, allowing coolant—typically a mix of water and propylene glycol—to flow through them and remove heat far more efficiently than traditional cold plates. In lab tests, Microsoft found the system can cool chips up to three times better, reducing the maximum temperature rise in a GPU by 65 percent. This breakthrough could enable more powerful, densely packed chips and reduce the energy needed to cool data centers. Current cooling methods, like air fans and copper cold plates, are reaching their limits. Air is a poor heat conductor—about 23 times less effective than water—and cold plates still rely on multiple insulating layers between the chip and coolant, trapping heat. Microfluidics bypasses these barriers by bringing liquid directly into the silicon, eliminating thermal resistance and allowing coolant to be heated more before needing recooling. This means lower energy use for cooling systems overall. The technology is especially critical for next-generation AI hardware, where chips are becoming more powerful and generate more heat. As data centers push toward 1 megawatt per rack—up from today’s 145 kilowatts—traditional cooling falls short. Microfluidics enables higher performance without overheating, allowing for more aggressive overclocking and reducing the need for excess server capacity. This could cut costs and environmental impact by minimizing the number of servers and data centers needed. Microsoft developed the system in collaboration with Corintis, a Swiss startup founded in 2022. Corintis uses AI-powered design software called Glacierware to map optimal coolant flow paths, mimicking natural patterns like leaf veins. This allows precise cooling of hot spots—areas that typically force chips to slow down—while minimizing flow where heat is lower. The result is a 55 percent reduction in pressure drop and a 13 percent decrease in coolant intake temperature compared to cold plates. While the system shows promise in the lab, real-world deployment faces challenges. Manufacturing must integrate microchannel etching into the chip fabrication process, and supply chains will need to adapt. Microsoft hasn’t set a timeline for commercial rollout, but the company hopes its progress will help drive industry-wide adoption of more efficient cooling. The concept isn’t new. IBM pioneered embedded liquid cooling in the 2000s, including 3D chip stacks with integrated water channels. Research at institutions like Georgia Tech and Binghamton University has also explored microfluidics for chips. Meta, KLA Instruments, and others are actively studying the technology. Despite the hurdles, microfluidics could unlock 3D chip architectures, where layers of compute, memory, and logic are stacked vertically—a design that’s been limited by heat. With direct liquid cooling, such systems become viable. Critically, while efficiency gains are welcome, they may trigger the Jevons paradox: as computing becomes cheaper and more efficient, demand grows, potentially increasing overall energy use. Microsoft and other tech giants are aware of this and are balancing innovation with sustainability. In short, Microsoft’s microfluidic cooling represents a major step forward in managing the heat of AI-era computing. If successfully scaled, it could make data centers faster, smaller, and more energy-efficient—key to powering the next generation of AI.