When Cooling Becomes the Constraint: The Thermal Supply Chain
We have spent years treating power as the one hard limit on AI infrastructure. Power is still the wall everyone talks about, but a second wall is rising quietly beside it, and it is made of cold plates, pumps, couplings, and coolant. The thermal supply chain is concentrated in a handful of specialist firms, it draws on the same industrial parts every other sector needs, and it cannot be scaled by simply spending more. A facility can have its power secured, its chips delivered, and still sit dark because the cooling components are months out. The operators who plan for thermal as they plan for power, early, contractually, with second sources, are the ones who will actually turn their megawatts into working compute.
Knowledge check
Why is the thermal supply chain becoming a bottleneck comparable to power, and what makes it harder to relieve quickly than simply ordering more parts?
Show the answer
Because the critical components, cold plates, dripless couplings, distribution units, and specialized coolants, are made by a small number of specialist firms and draw on shared industrial supply bases like pumps and heat exchangers. Capacity is precision-limited and cannot be doubled by adding a shift, so lead times stretch and a fully powered, chip-ready site can still sit idle waiting for cooling.