Above ~30–50 kW per rack, air cooling fails — so next-gen GPUs make liquid cooling mandatory, not optional. This is a fast transition more than a hard shortage.
Tightness
+6 this quarter — Demand is surging as Blackwell-class racks force liquid cooling, but a broad supplier base keeps this a scaling transition rather than a sold-out chokepoint.
Next-gen accelerators run too hot for air — an H200 dissipates ~700W, and Blackwell-class racks push past what air can economically remove. Without liquid cooling, operators simply can't run high-density racks, so it caps how much compute fits per site. Unlike the silicon and power chokepoints, the constraint here is integration and skilled deployment, not a single-supplier monopoly.
Many credible suppliers and fast capacity additions mean supply largely keeps pace with demand; the real constraint is integration and skilled labor rather than a monopoly — which is why this reads tight rather than critical.
Get the squeeze before the headlines.
A weekly read on the AI buildout's tightest constraints. Free, sourced.