Microsoft claims a 'breakthrough' in AI chip cooling (www.engadget.com)

Microsoft announced a prototype microfluidic cooling system it calls a breakthrough for GPUs, claiming up to three times better cooling than traditional cold-plate methods and a reduction in maximum silicon temperature rise of about 65% (workload- and chip-dependent). The company says this could enable safer overclocking, denser server racks (lowering latency), and better-quality waste heat capture — all attractive outcomes for performance-hungry AI workloads and data-center operators — while also offering potential sustainability benefits by cutting cooling overhead and grid stress. Technically, the design routes coolant through thread-like channels etched directly on the back of the chip, bringing liquid much closer to heat sources than conventional cold plates that are separated by several material layers. Microsoft also uses AI to optimize coolant routing through a biomimetic channel layout inspired by leaf and wing venation. The approach addresses long-standing implementation challenges of microfluidics but remains a prototype and claim pending independent validation; effectiveness will vary by chip type and workload. If practical at scale, it could change thermal design trade-offs in AI infrastructure and reduce one meaningful piece of the energy footprint for large models.