Power, Cooling, and the Data They Emit
Faraday's read: power and cooling are converging into a single question — how many watts can this building actually deliver and remove per rack — and that number is now the sector's hardest currency. The compute is a commodity anyone can buy; facilities that can power and cool it at modern density are not. Watch the margins: provisioned-versus-drawn power and used-versus-available cooling are the two most honest figures a facility emits, because both are measured physics, not presentation. Where they are tight, growth is over regardless of what the compute dashboard says.
Knowledge check
Why has the rise of high-density compute turned cooling into a facility-layer problem rather than a compute problem?
Show the answer
Because the newest accelerators concentrate several times more power, and therefore heat, into each rack than the cooling plant was designed for years earlier. Air can no longer remove heat fast enough at these densities, forcing a shift to liquid methods (direct-to-chip and immersion) that require new physical infrastructure — piping, coolant distribution, different heat rejection — retrofitted into or designed into the building. This is the slow, long-lifespan facility being forced to accommodate a fast-changing compute generation it was never built for, which cannot be solved in software.