On September 23, Microsoft revealed in a blog post that it has successfully overcome the cooling bottleneck faced by AI chips, thanks to the development of an innovative on-chip microfluidic cooling system. This cutting-edge system boasts a cooling efficiency three times that of the most advanced cold plate technology currently available.

Traditional cold plate liquid cooling methods have long been constrained by their limited heat dissipation capabilities. In contrast, Microsoft's new system operates directly on the silicon chip's surface, effectively eliminating the 'thermal insulation layer' that often impedes cooling. By integrating AI technology, the system can precisely direct the cooling flow, thereby tackling the root cause of excessive heat generation.

Experimental results have demonstrated that this system can significantly reduce the maximum temperature rise in GPU chips, while also enhancing the energy efficiency ratio of data centers. Rigorous tests have confirmed its ability to maintain stable temperature control, even under high loads and during sudden task surges. This capability not only supports overclocking in data centers but also contributes to cost reduction. However, the widespread implementation of this technology is not without its engineering hurdles, including the construction of microchannels and ensuring leak-proof packaging.