Thermal Components & Coolant Supply Chain

Glossary

Cold plate

A flat metal block, usually copper or aluminum, with fine internal channels that presses against a chip through thermal interface material. Coolant flowing through its channels absorbs heat by convection. Its internal geometry is precision-engineered, making it one of the harder cooling components to manufacture at scale.

Coolant distribution unit

The machine at the heart of a liquid-cooling loop that pumps and conditions coolant for the racks and uses a heat exchanger to isolate the clean technology loop touching the chips from the facility water loop. It contains redundant pumps, filtration, controls, and leak detection, and often has long lead times.

Direct-to-chip cooling

A liquid-cooling architecture that bolts sealed metal cold plates directly onto the hottest chips and runs coolant through them, capturing roughly 70 to 80 percent of a server's heat into liquid while a small amount of air handles the rest. It is the approach the major accelerator vendors have effectively standardized on.

Immersion cooling

A liquid-cooling architecture that submerges an entire server board in a tank of electrically non-conductive (dielectric) fluid. In single-phase immersion the fluid is pumped past a heat exchanger; in two-phase immersion the fluid boils at the chip surface and condenses on a cooled coil, capturing essentially all of the heat.

PFAS

Per- and polyfluoroalkyl substances, a large family of fluorinated chemicals prized in cooling for being inert, non-flammable, and tunable in boiling point, but so environmentally persistent they are called forever chemicals. Sweeping regulation of the class threatens the engineered fluids that two-phase immersion relies on.

Power usage effectiveness (PUE)

The ratio of a facility's total power draw to the power that actually reaches the computers. Air-cooled halls historically ran around 1.4 to 1.6; well-designed liquid cooling pushes PUE toward 1.1 or below, converting cooling efficiency directly into usable compute capacity at power-constrained sites.

Quick-disconnect (dripless coupling)

A self-sealing coupling that lets a fluid-filled server be connected to or removed from a rack manifold without draining the loop and without spilling a drop. Large deployments need hundreds of thousands of them, and because only a few specialist firms make trusted ones, they are a supply chokepoint.

Two-loop design

The standard configuration separating a clean, precisely conditioned technology loop that touches the chips from a dirtier facility water loop, with a heat exchanger passing heat between them without mixing the fluids. It trades a small thermal penalty for control and reliability, protecting expensive hardware.