Microsoft researchers have demonstrated a new microfluidic cooling technology that channels liquid directly into silicon chips, removing heat up to three times more effectively than today’s cold plate systems. The approach etches hair-thin channels into the back of the chip, allowing coolant to flow directly onto the hotspots. Early lab tests showed a 65% reduction in maximum GPU temperature rise compared with cold plates.
The company combined this hardware innovation with AI algorithms that map heat signatures and direct coolant flow for greater efficiency. By integrating cooling at the silicon level, Microsoft believes microfluidics could enable denser datacenter designs, higher overclocking potential, and more sustainable operations. The technology also promises to improve power usage effectiveness (PUE), cut operational costs, and support future 3D chip architectures that would otherwise be constrained by heat.
Microsoft has collaborated with Swiss startup Corintis on bio-inspired microchannel designs and has produced four iterations of prototypes in the past year. The company is now working on packaging, etching, and manufacturing techniques to move toward production deployment. The breakthrough comes as Microsoft invests more than $30 billion in capital expenditures this quarter, including custom chips such as its Cobalt and Maia families.
• Removes heat up to 3x better than cold plates
• 65% lower maximum GPU temperature rise in lab tests
• Bio-inspired channel design resembles veins in leaves
• Potential to enable overclocking without thermal damage
• Supports higher server density and future 3D chip stacking
“Microfluidics would allow for more power-dense designs that will enable more features that customers care about and give better performance in a smaller amount of space,” said Judy Priest, corporate vice president and chief technical officer of Cloud Operations and Innovation at Microsoft.
🌐 Analysis: Thermal management has become one of the biggest constraints for AI infrastructure buildouts, with GPUs such as NVIDIA’s Blackwell and AMD’s MI300X consuming increasing amounts of power. Microsoft’s in-chip microfluidics effort reflects a wider industry trend as hyperscalers test advanced liquid cooling approaches, including immersion and rear-door heat exchangers. If successfully commercialized, microfluidics could reshape datacenter cooling strategies and accelerate the adoption of more compact, power-dense silicon designs.
Original article by Catherine Bolgar on Microsoft.com: https://news.microsoft.com/source/features/ai-innovation/microfluidics-cooling-at-the-micro-level-for-microsofts-datacenters
