Technology
A cold plate is the only part of a liquid cooling system that touches the thing being cooled. It is a machined and brazed metal block, its internal channels are matched to the layout of a specific processor, and the manifolds behind it decide whether every plate in a rack gets the flow it was designed for.
In one sentence
A cold plate is a machined or brazed metal block clamped against a processor, with fine internal channels through which coolant flows; manifolds distribute that coolant to every plate in a server and every server in a rack.
The performance comes from surface area in a small volume. Channels are formed by skiving fins from solid copper, by folding and brazing thin fin stock into the cavity, or by machining a microchannel pattern directly, and the pattern is laid out to put the densest structure over the hottest part of the die. Because that layout follows a specific package, a plate is designed for one processor generation and redesigned for the next.
Manufacturing is precision metalwork at volume, done largely by the Taiwanese thermal module makers that already supplied heat sinks and vapour chambers, and by a smaller group of specialists selling complete loops. The limit on output is machining hours, brazing capacity and leak testing rather than design, and a plate only ships in quantity once it has been designed into a server maker's platform — which puts a component supplier's schedule inside somebody else's product cycle.
Die layout, hot spot location, permitted mounting pressure and the thermal interface material between plate and lid are all inputs to the channel design. A plate that performs well on one accelerator is not transferable to the next one, which is why the component market moves in step with accelerator generations and why second sources take a full cycle to appear.
Channels a fraction of a millimetre across are intolerant of particulates and of corrosion products from mismatched metals in the loop. The consequence is a slow rise in temperature over months rather than an immediate failure, which is exactly why the technology loop is kept separate from facility water and why filtration and inhibitor chemistry are specified with the hardware.
Plates in parallel do not share flow equally unless the manifold is designed to make them. Get it wrong and one server in the rack runs hot while the loop as a whole looks healthy. That is a hydraulic design problem carrying real consequences, and it is why manifolds are qualified as part of the rack rather than bought as plumbing.
2 of these are marked as a chokepoint: a handful of qualified suppliers, a multi-year lead time, or a single geography.
Output is limited by machining hours, brazing furnaces and leak test benches, none of which expands on the timescale of accelerator demand.
Bearings and precision machiningPlates, manifolds and their mounting hardware are designed into the machine; without a qualified platform there is nothing to fit them to.
AI server systemsThe channel layout follows the die and the power it dissipates, so each accelerator generation obsoletes the previous plate design.
Merchant GPUsThe inhibitor package has to match every metal the fluid touches, or corrosion products block the finest channels first.
Coolants and fluidsPlates are mostly copper, chosen for conductivity, and the quantity per rack rises with every increase in the power being removed.
Other pages in this map that name Cold plates and manifolds 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 cold plates and complete liquid-cooling assemblies to the server makers rather than to operators.
Supplies heat sinks and cold plates into the racks its Taiwanese customers assemble.
Supplies thermal modules and cold plates from the same heat-sink manufacturing base.
Supplies cold plates and manifolds designed into several accelerator reference platforms.
Supplies cold plates and thermal modules, moving from consumer cooling into rack-scale work.
Supplies machined heat spreaders and cold plates as a precision metalwork subcontractor.
Supplies cold plates and the connection and protection products around them.
Supplies plates and manifolds as part of the complete thermal chain it sells.
Whether plate designs converge enough for operators to second-source within one accelerator generation.
Whether machining and brazing capacity catches up with accelerator deployment schedules.
Whether more components inside the server gain plates, shrinking the air-cooled remainder.
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.