Industry explainer · U.S. view · Data centers · AI · Infrastructure

Data Center Cooling and Power Industry Explained: The U.S. View

AI accelerators turn almost all the electricity they use into heat, and racks of them now draw far more power than the servers data centers were designed for. That has created a boom for the equipment around the chips: switchgear and transformers that bring power in, uninterruptible power supplies (UPS) and busways that distribute it, and cooling systems that take the heat out. Air cooling is reaching its limits, so the newest AI racks use liquid cooling, piping coolant directly to cold plates on the chips; even network switches now come in liquid-cooled versions. Vertiv, a leading U.S. supplier of data center power and cooling, reported second-quarter 2026 sales of $3.27 billion, up 24%, with Americas sales up 29%, and raised its 2026 forecast to $13.8–14.2 billion, or 30–32% organic growth, guiding 34–36% organic growth for the third quarter. GE Vernova said its data center orders for grid equipment passed $5 billion in the first half of 2026, more than double all of 2025. The limits are factory capacity, supply-chain congestion and, above all, how much grid power is available.

中文版（中国视角）：一文看懂数据中心液冷与电源产业链 →

Vertiv: Q2 2026 sales $3.27B, up 24%; 2026 organic growth forecast 30–32%
Grid gear: GE Vernova data center orders over $5B in H1 2026, double all of 2025
Liquid: Arista introduced liquid-cooled 1.6T AI switch platforms in 2026

IN ONE LINEAI racks are too hot for air; liquid cooling and power gear boom.

How the chain works

Power arrives from the grid or on-site plants through switchgear and transformers, then UPS systems and busways feed the racks. AI racks are too dense for air alone, so liquid cooling (dashed line) carries heat to chillers and heat-rejection equipment.Swipe sideways on a phone to see the full chart.

Power equipment, liquid cooling and heat rejection

	Power distribution	Liquid cooling	Heat rejection
Job	Bring power in and keep it uninterrupted	Remove heat at the chip	Move heat out of the building
Equipment	Switchgear, transformers, UPS, busways	Cold plates, coolant distribution units, pipes	Chillers, dry coolers
2026 signal	GE Vernova data center orders over $5B in H1	Liquid-cooled switches launched by Arista	Vertiv raised 2026 guidance
What to watch	Lead times, grid connections	Share of new racks liquid-cooled	Water and power use

Why AI changes the building

A data center is mostly a machine for delivering power and removing heat. AI accelerators pack much more power into each rack, so the same building needs heavier power distribution and much stronger cooling. Moving air through a hot rack becomes inefficient, and liquid, which carries far more heat, takes over at the chip.

The change reaches beyond servers. Arista's second-quarter 2026 release introduced 1.6 Tbps AI networking platforms including liquid-cooled options, and switch makers are cutting power use with new optics. Every watt saved on networking or cooling is a watt available for accelerators.

The suppliers' numbers

Vertiv reported second-quarter 2026 net sales of $3,274 million, up 24% from a year earlier, including 18% organic growth; Americas sales rose 29.2%. Its adjusted operating margin was 22.6%, up 4.1 percentage points. Chief executive Giordano Albertazzi said demand for AI and general compute continues to intensify. Vertiv raised its 2026 sales forecast to $13.8–14.2 billion and guided third-quarter organic growth of 34–36%.

Upstream, grid equipment makers are seeing the same demand. GE Vernova said its Electrification segment's data center orders exceeded $5 billion in the first half of 2026, more than double the 2025 total, and that segment's equipment backlog reached $40.6 billion.

Limits: factories, supply chains and power

Growth is constrained by how fast equipment can be built and delivered. Vertiv said second-quarter revenue reflected minor timing shifts, mainly from temporary supply chain congestion and multi-phase project schedules, which it expected to resolve in the second half.

The bigger limit is power itself. PJM said data centers accounted for nearly 5,100 MW of the 5,250 MW increase in its forecast peak load for 2027/28, and its capacity auction fell short of the reliability target. On-site generation is one response, which adds to demand for gas turbines.

How the U.S. view differs from the China view

The Chinese version of this page looks at Chinese makers of liquid cooling parts, power supplies and server components, many of which supply global server and data center builders, and at China's own computing centers.

From the U.S. side, the largest AI data center build-out is at home, and the binding constraint is less the cooling hardware than the power: grid connections, transformers and generation. Equipment makers with U.S. factories and service networks benefit most.

2026: the infrastructure behind AI

As hyperscalers raised capital spending in 2026, orders for data center power and cooling equipment rose with it.

Vertiv raised its full-year outlook to 30–32% organic growth, and GE Vernova's data center orders for grid equipment more than doubled.

Even with strong demand, deliveries were limited by supply chain congestion and by how much grid power new sites could get.

What this shows

AI data centers need new power and cooling designs, not just more servers.
Liquid cooling is becoming standard for the densest AI racks.
Grid power is the ultimate limit on how fast data centers can grow.

Common misreadings

MythData center spending is mostly chips.

RealityPower and cooling suppliers like Vertiv expect 30–32% organic growth in 2026.

MythAir cooling can handle AI racks.

RealityEven switches now come in liquid-cooled versions for AI clusters.

MythEquipment shortages are the main limit.

RealityGrid capacity is tighter: PJM's 2027/28 auction fell short of its reliability target.

MythOnly the data center owner benefits.

RealityGrid equipment makers also gain: GE Vernova's data center orders more than doubled.

Rack density, power and spending: scenarios

Scenario	Main effect	How it passes through
Racks get denser	Cooling demand	More liquid cooling per rack
Grid connections delayed	New sites	More on-site power, slower openings
Hyperscaler capex rises	Equipment orders	Higher backlogs
Supply chain congestion eases	Deliveries	Revenue catches up with orders
Water restrictions tighten	Cooling design	Shift to closed-loop systems
AI spending slows	New projects	Order growth slows

Supply and demand by segment

Segment	Supply rigidity	Concentration	New demand	Policy role	What to watch
Switchgear and transformers	Very high (lead times)	High	Yes	Medium	Lead times
UPS and power distribution	Medium	High	Yes	Low	Supplier guidance
Liquid cooling	Medium (new lines)	Medium	Yes (AI racks)	Low	Adoption rate
Heat rejection	Medium	Medium	Yes	Medium (water rules)	Project pipelines

Supply rigidity: how long new capacity takes and whether money can speed it up. Concentration: whether output sits in a few countries or companies. The more rigid and concentrated, the more prone to shortages.

How to track it

Indicator	What it tells you	Where to find it
Power and cooling suppliers' guidance	Equipment demand	Vertiv quarterly results, free
Data center orders at grid equipment makers	Upstream demand	GE Vernova quarterly results, free
Grid capacity and data center load	Power constraint	PJM releases, EIA outlook, free
Hyperscaler capex	Budget for new sites	U.S. view: AI compute
Power grid	The biggest constraint	U.S. view: power grid

Sources

Last updated 2026-10-18.

This page explains the structure and supply-demand logic of an industry for educational purposes. It does not refer to any specific security and is not investment advice. Figures are dated and sourced; they may have changed since, so check the latest official data.