What is the Thermal Ceiling?
The Thermal Ceiling is the point at which a data centre's ability to remove heat, rather than its power supply or its silicon availability, becomes the binding constraint on useful compute output. It is a system property of the whole thermal path rather than a property of any single component, and it is not the same thing as a rack density limit.
Definition
The Thermal Ceiling is the point at which a data centre's ability to remove heat, rather than its power supply or its silicon availability, becomes the binding constraint on useful compute output.
Why it matters
For most of the history of data centres, capacity planning was a power question. In AI infrastructure it is increasingly a heat question. A facility can hold sufficient grid capacity and sufficient accelerators and still be unable to run them at rated performance, because heat cannot be removed fast enough at an acceptable coolant temperature.
How it is measured
Thermal Yield, the proportion of total facility power that reaches silicon as useful compute after cooling, conversion and distribution losses, is the direct measure. Tokens per Megawatt is the commercial expression of the same quantity.
Common misunderstandings
The Thermal Ceiling is not the same as a rack density limit. A facility can be well below its rated rack density and still be thermally ceiling-bound, because the constraint is the temperature at which heat can be removed, not the quantity.
Coral's position
Coral was founded on the thesis that the Thermal Ceiling arrives before the power wall in most AI deployments, and that it is a system property of the whole thermal path rather than a property of any single component.
Take the vocabulary into a real site conversation.
An architecture review applies these terms to a specific building: its power envelope, its thermal capacity, its density target and its jurisdiction.