Sub-system
Direct-to-chip cooling bolts a cold plate onto each processor and circulates liquid through it. It is the mainstream answer for AI racks, it removes most of a rack's heat into a loop that needs no refrigeration, and its component supply chain has become a real constraint on deployment.
In one sentence
Direct-to-chip liquid cooling circulates coolant through cold plates mounted on the highest-power components, carrying heat to a coolant distribution unit that transfers it into a facility loop.
The architecture has three parts. Cold plates and the manifolds that feed them sit inside the server. A coolant distribution unit — in the rack or serving a row — circulates the technology loop and transfers its heat to the facility loop through a heat exchanger, keeping the two fluids separate. The facility loop carries the heat to dry coolers or cooling towers outside.
The efficiency gain comes from temperature. Because the plate sits directly on the component, the coolant can be relatively warm and still remove heat effectively — warm enough that the outside loop can usually reject it without compressors for most or all of the year. Removing the chiller is where most of the energy saving is.
Cold plates cover the processors and memory but not everything: power supplies, drives, network adapters and voltage regulators still produce heat. Typically the plates capture the large majority of a rack's heat and the remainder is handled by air, which is why liquid-cooled racks still sit in halls with air handling.
Facilities gain a fluid system: leak detection, filtration, coolant chemistry management, and quick-disconnect couplings that let a server be removed without draining the loop. Data-centre operations teams did not previously need those skills, and building them is part of the cost of adopting the technology.
Cold plates, manifolds, distribution units, couplings and hoses are precision components made by a comparatively small group of suppliers, whose volumes were sized for a niche. Demand arrived faster than capacity, and a rack cannot ship without them — which is how a hose fitting ends up on the critical path for AI capacity.
3 of these are marked as a chokepoint: a handful of qualified suppliers, a multi-year lead time, or a single geography.
Machined plates matched to a specific processor package, from a precision metalwork base whose capacity has repeatedly been the pacing item.
Cold plates and manifoldsThe skid that circulates the technology loop and hands its heat to the building; without one the plates have nowhere to send the heat.
Coolant distribution unitsDripless couplings rated for many connection cycles are a narrow supplier market and a real lead-time item.
Couplings and hosesThe rack-level loop is only useful if the building can accept the heat and reject it outside.
Heat rejection and waterMixed metals and incompatible fluids cause corrosion and fouling; specifications exist because the failure mode is slow and expensive.
The plates, manifolds and couplings are designed into the machine; without a platform qualified for them there is nothing to plumb.
AI server systemsPower supplies, drives and network adapters keep producing heat the plates do not touch, so a liquid-cooled rack still sits in a hall with a working air path.
Air coolingOther pages in this map that name Direct-to-chip liquid as something they cannot do without.
What each company supplies at this step, and — where a public figure exists — its share of this specific market — with what that share measures, the period it covers and who published it. Some rows also show the company’s own reported revenue for the segment covering this step, which is a different thing: it says how much this business matters to that company, not how much of the market it holds. Not a ranking and not a recommendation.
Supplies coolant distribution units and liquid cooling systems for data centres.
Supplies thermal management products including data-centre cooling systems.
Supplies liquid cooling products including cold plates and distribution.
Integrates liquid-cooled server and rack systems.
Supplies cold plates and coolant distribution units, and is qualified on several accelerator reference designs.
Supplies the cold plates and loops that the server makers integrate rather than build.
Sells the liquid-cooling line it acquired alongside the electrical plant the same hall needs.
Supplies pumps, coolant distribution units and thermal modules into the Taiwanese rack builders.
Supplies the pump and cold-plate designs that much of the liquid-cooling supply chain licensed from it.
Whether cooling component supply catches up with accelerator deployment schedules.
Whether standardised interfaces let operators mix suppliers within one hall.
Whether the air-cooled remainder within liquid racks shrinks as more components get plates.
Each page explains one technology in plain language, states what it depends on, and names companies by what they supply at that step. Company roles are described qualitatively and deliberately carry no market shares, revenue figures or rankings — those change faster than an explainer can, and a stale number is worse than none. Ticker links point at company pages on this site and are provided for reference only.
Nothing here is investment advice, a recommendation, or a forecast. A company named on a page about a technology is not thereby a good investment, and the chokepoints described are structural facts about supply chains rather than predictions about prices. Technology moves; where a page describes something as unresolved or in development, that was true when it was written.