This startup’s metal stacks could help solve AI’s massive heat problem

The article discusses Alloy Enterprises, a startup addressing the escalating cooling challenges posed by increasingly power-hungry AI data center racks. Nvidia recently revealed that future GPU racks, expected in 2027, could consume up to 600 kilowatts of electricity—nearly double the capacity of some fast EV chargers. This surge in power density creates significant heat dissipation issues, especially for peripheral chips like memory and networking hardware, which account for about 20% of a server’s cooling load but currently lack effective cooling solutions. Alloy Enterprises has developed a novel cooling technology using stacks of copper sheets bonded through a process called diffusion bonding or "stack forging," producing seamless, solid metal cold plates that can fit into tight spaces and withstand high liquid cooling pressures. Unlike conventional machined cold plates, which are assembled from separate halves and prone to leaks at seams, Alloy’s stack forging method creates single-block copper plates with no seams and superior structural integrity. This process also allows for finer features—down to 50 mic