Layer 6 of 8 · 18 companies
Power, cooling and data centers
Is power the real bottleneck, and who gets paid for relieving it?
AI data centers need power, cooling, buildings and grid connections, and those take years to supply. This layer grew +37% in two years, slower than chips, because physical infrastructure can’t scale as fast. That slowness is the point: it sets how fast everything else can grow, and it is why its backlogs are among the longest in the stack.
- combined revenue, latest four quarters
- $229B
- combined revenue, latest four quarters
- revenue change over two years
- +37%
- revenue change over two years
- GE Vernova’s backlog, 4.5× its annual revenue
- $197B
- GE Vernova’s backlog, 4.5× its annual revenue
- of the two-year revenue gain kept as gross profit
- 31%
- of the two-year revenue gain kept as gross profit
The bottleneck nobody can buy their way out of
A GPU can be manufactured in months. A gas turbine takes years to deliver, a transmission line longer, and a new nuclear plant longer still. As the buyers’ spending has grown, the binding constraint on new AI capacity has shifted from chips toward power and the equipment that delivers it. The filings show it: buyers sign 20-year power contracts, utilities and generators report data-center demand as their main source of growth, and equipment makers report backlogs measured in years.
This layer is wide. It includes equipment makers (Vertiv for rack power and liquid cooling, Eaton for electrical distribution, GE Vernova for turbines and grid gear, Modine for chillers and cooling, Powell for switchgear, nVent and Hubbell for components), builders (Quanta for transmission and grid connections, EMCOR for electrical and mechanical work inside the buildings), power generators (Constellation, Vistra, Talen, Bloom Energy’s fuel cells) and landlords (Equinix, Digital Realty, Applied Digital). Combined, they earned $229B over their latest four quarters.
Quarterly revenue: power, cooling and data centers
The layer’s 16 US-listed companies, five largest shown separately; Other is Vistra, Vertiv, Equinix, Hubbell, Digital Realty, nVent, Talen, Bloom Energy, Modine, Powell Industries, Applied Digital
- GE Vernova
- Quanta Services
- Eaton
- Constellation
- EMCOR
- Other
Source: company 10-K and 10-Q filings (SEC XBRL); fiscal quarters aligned to the nearest calendar quarter. Companies appear from their first reported quarter, so newly listed ones (such as CoreWeave, Sandisk or Astera Labs) add to the total partway through.
Show the numbers
| Quarter | GE Vernova | Quanta Services | Eaton | Constellation | EMCOR | Other |
|---|---|---|---|---|---|---|
| Q1 2022 | — | $4.0B | $4.8B | $5.6B | $2.6B | $10.2B |
| Q2 2022 | — | $4.2B | $5.2B | $5.5B | $2.7B | $10.6B |
| Q3 2022 | — | $4.5B | $5.3B | $6.1B | $2.8B | $12.4B |
| Q4 2022 | — | $4.4B | $5.4B | $7.3B | $2.9B | $11.3B |
| Q1 2023 | $6.8B | $4.4B | $5.5B | $5.8B | $2.9B | $11.3B |
| Q2 2023 | $8.1B | $5.0B | $5.9B | $4.7B | $3.0B | $11.4B |
| Q3 2023 | $8.3B | $5.6B | $5.9B | $5.3B | $3.2B | $14.0B |
| Q4 2023 | $10.0B | $5.8B | $6.0B | $5.0B | $3.4B | $11.7B |
| Q1 2024 | $7.3B | $5.0B | $5.9B | $5.3B | $3.4B | $11.9B |
| Q2 2024 | $8.2B | $5.6B | $6.3B | $4.4B | $3.7B | $12.9B |
| Q3 2024 | $8.9B | $6.5B | $6.3B | $4.8B | $3.7B | $14.0B |
| Q4 2024 | $10.6B | $6.6B | $6.2B | $4.5B | $3.8B | $13.6B |
| Q1 2025 | $8.0B | $6.2B | $6.4B | $6.1B | $3.9B | $14.0B |
| Q2 2025 | $9.1B | $6.8B | $7.0B | $5.2B | $4.3B | $14.3B |
| Q3 2025 | $10.0B | $7.6B | $7.0B | $5.7B | $4.3B | $16.2B |
| Q4 2025 | $11.0B | $7.8B | $7.1B | $5.7B | $4.5B | $17.0B |
| Q1 2026 | $9.3B | $7.9B | $7.5B | $7.5B | $4.6B | $18.2B |
Backlog is the story
GE Vernova reports $197B of remaining performance obligations, 4.5× its revenue over the latest four quarters. Quanta reports $41.8B and EMCOR $18.6B. Backlogs this long give unusual visibility: much of the next two or three years’ revenue is already contracted. They also show where pricing power is building, because a supplier with years of orders can choose which new ones to take.
Growth is fastest where the product is newest or scarcest. Over their latest four quarters Applied Digital grew +559%, Bloom Energy grew +112%, Talen grew +109% and nVent grew +46%. Bloom Energy’s fuel cells and the nuclear generators sell something AI data centers increasingly want: power that doesn’t wait for a grid connection, or carbon-free power sold under long contracts.
Backlog in power and data-center infrastructure
Remaining performance obligations, with how many years of current revenue each represents
Source: company 10-K and 10-Q filings (SEC XBRL); fiscal quarters aligned to the nearest calendar quarter. Companies that report remaining performance obligations above $1 billion.
Show the numbers
| Company | Backlog | Revenue | Backlog ÷ revenue |
|---|---|---|---|
| GE Vernova | $197B | $44.0B | 4.5× |
| Quanta Services | $41.8B | $37.3B | 1.1× |
| EMCOR | $18.6B | $19.8B | 0.9× |
| Equinix | $15.9B | $10.1B | 1.6× |
| Constellation | $11.9B | $24.8B | 0.5× |
| Vistra | $7.1B | $19.0B | 0.4× |
| Talen | $4.0B | $3.9B | 1.0× |
| Powell Industries | $2.4B | $1.2B | 2.1× |
Long contracts with the buyers
The generators’ filings show the buyers locking in supply for decades. Constellation reports that the restart of a nuclear plant is supported by a 20-year power purchase agreement with Microsoft. Vistra reports 20-year agreements to supply Meta with 2,609 megawatts of nuclear power and capacity. Contracts like these let generators finance new or restarted capacity on the strength of a creditworthy buyer, and they turn a commodity business, selling power at market prices, into something closer to a utility with long-term contracted revenue.
The data-center landlords sit between the two. Equinix spent $5.8B on capex over its latest four quarters against $10.1B of revenue, and Applied Digital $3.7B against $597M. Landlords take construction and leasing risk in exchange for long leases with the largest tenants in the world; their returns depend on building at a cost those leases can cover.
The nuclear start-ups: options, not earnings
Small modular reactors are among the most popular AI-power stories with individual investors, so it is worth being clear about what the filings show. Oklo has no revenue yet. Over its latest four quarters it reported an operating loss of $218M and spent $159M on capital projects, and it held $1.6B of cash and short-term investments at June 30, 2026. NuScale, whose reactor design has US regulatory approval, reported $11M of revenue and an operating loss of $733M over the same span.
These companies can’t be analyzed like the rest of the layer, because there are no earnings or margins to compare. Their value rests on getting reactors licensed, built and delivered on budget, years from now, and on customers signing binding contracts rather than announcing intentions. That is why we list them on this page but leave them out of the layer’s totals and growth rankings. The things to read in their filings are cash on hand against the rate it is being spent, and any firm purchase agreements disclosed as commitments.
Margins: steady rather than spectacular
The layer’s gross margin moved from 25% to 27% over two years, and it kept 31% of its extra revenue as gross profit. That is a far smaller share than chips or memory, because much of the work is construction and manufacturing with real labor and materials costs. But the margins are rising and the visibility is long, a different kind of quality from the toll booths.
The risk in this layer is timing rather than demand. Equipment and generation ordered today for data centers due in 2028 or later is a commitment made on today’s forecasts. If the buyers slowed their plans, the order books would shrink before revenue did, and the long backlogs that look like safety now would be renegotiated. Watch new orders, not just backlog.
The companies, side by side
Latest four reported quarters for US filers; latest fiscal year for 20-F filers (marked). Underlined names have their own company page; the others link to the company’s filings on SEC EDGAR. Margins are left blank where a company doesn’t report gross profit or cost of revenue.
| Company | Revenue | vs year before | Gross margin | Operating margin | Free cash flow | Backlog (RPO) |
|---|---|---|---|---|---|---|
| GE Vernova GEVGas turbines, grid equipment and electrification | $44.0B | +17% | 22% | 6% | $17.1B | $197B |
| Quanta Services PWRBuilds transmission lines and grid connections | $37.3B | +37% | 16% | 7% | $3.2B | $41.8B |
| Eaton ETNElectrical equipment from grid to rack | $30.0B | +16% | 36% | — | $3.9B | — |
| Constellation CEGNuclear power sold under long contracts to data centers | $24.8B | +20% | — | 19% | $309M | $11.9B |
| EMCOR EMEElectrical and mechanical construction inside data centers | $19.8B | +22% | 20% | 10% | $1.5B | $18.6B |
| Vistra VSTNuclear and gas generation | $19.0B | +19% | — | 19% | $2.3B | $7.1B |
| Vertiv VRTPower distribution and liquid cooling for racks | $13.1B | +35% | 38% | 20% | $3.1B | — |
| Equinix EQIXColocation data centers | $10.1B | +11% | 52% | 23% | −$1.7B | $15.9B |
| Hubbell HUBBUtility and electrical components | $6.6B | +16% | 35% | 20% | $928M | — |
| Digital Realty DLRWholesale and colocation data centers | $6.3B | +13% | — | 12% | −$718M | — |
| nVent NVTEnclosures, power and liquid-cooling hardware | $4.8B | +46% | 37% | 17% | $578M | — |
| Talen TLNNuclear and gas generation, including power sold to Amazon | $3.9B | +109% | — | 2% | $568M | $4.0B |
| Bloom Energy BEFuel cells for on-site power | $3.8B | +112% | 32% | 15% | $437M | — |
| Modine MODData-center cooling: chillers, air handlers and liquid cooling | $3.4B | +29% | 22% | 10% | $100M | — |
| Powell Industries POWLCustom switchgear and power-distribution equipment | $1.2B | +7% | 30% | 20% | $244M | $2.4B |
| Applied Digital APLDBuilds and leases AI data-center campuses | $597M | +559% | 20% | −32% | −$3.6B | — |
| NuScale SMRSmall modular reactor designs, early revenue | $11M | −81% | 24% | −6854% | −$779M | — |
| Oklo OKLOSmall nuclear reactors in development, pre-revenue | — | — | — | — | −$276M | — |
Revenue and growth. Total revenue over the four quarters, and its change from the four quarters before. Fast growth with a flat margin usually means volume; fast growth with a rising margin usually means pricing power.
Gross margin. The share of revenue left after the direct cost of what was sold. It is the best single measure of how much of each dollar a company keeps rather than passes on to its own suppliers.
Operating margin. What is left after research, sales and overhead too. The gap between gross and operating margin shows how much the company spends to stay competitive.
Free cash flow. Operating cash flow minus capital spending: the cash the business generated after paying to maintain and grow itself. Negative free cash flow during fast growth means the growth is consuming cash.
Where the money comes from and goes
This layer is paid by AI clouds. Its own costs are mostly labor, materials, fuel and its own capital spending, outside the stack. For the dollar amounts and who keeps them, see following the money.
Financing links that involve this layer
Quoted from the filings. Numbers match the money-loop map.
24Microsoft → ConstellationPower contract
20-year PPA for a restarted nuclear plant
“The restart is supported by a 20-year PPA with Microsoft to purchase the output generated from the renewed plant.”
Constellation 10-Q for the period ended June 30, 2026, filed August 6, 2026
25Meta → VistraPower contract
2,609 MW of nuclear power under 20-year PPAs
“In January 2026, Vistra announced it had entered into 20-year PPAs with Meta, pursuant to which the Company has agreed to supply Meta with a total of 2,609 MW of carbon-free power and capacity from the Company's PJM nuclear power plants as follows: 1,268 MW of energy and capacity from Perry and 908 MW of energy and capacity from Davis-Besse; and 213 MW of uprate energy and capacity from Perry, 80 MW of uprate energy and capacity from Davis-Besse, and 140 MW of uprate energy and capacity from Beaver Valley.”
Vistra 10-Q for the period ended June 30, 2026, filed August 10, 2026
What to watch in the next filings
- Orders and backlog at GE Vernova, Quanta, Vertiv and EMCOR: new orders signal demand before revenue does.
- New long-term power purchase agreements between generators and the buyers.
- Gross margins at the equipment makers: rising margins on long backlogs mean pricing power is building.
- Data-center landlords’ capex against signed leases.
- Interconnection queues and grid-connection timelines, which set how fast new capacity can switch on.
Free · quarterly · from SEC filings
Get the AI Stack Report every quarter
After each Big Tech earnings season we update the map from the new filings and publish what changed: who spent, who got paid, who kept it and what the filings reveal about the loop. Free, in Rob May’s Investing in AI newsletter.
Form not showing? Subscribe on Substack. No spam; unsubscribe any time.
Questions about power and DCs
Is power the bottleneck for AI data centers?
Increasingly, yes. Chips can be manufactured in months, but turbines, transformers, transmission lines and power plants take years. The buyers now sign long-term power contracts, and equipment makers report backlogs of several years.
How big is GE Vernova’s backlog?
GE Vernova reports $197B of remaining performance obligations, about 4.5× its revenue over its latest four quarters.
Why are tech companies signing nuclear power deals?
Nuclear plants produce steady, carbon-free power around the clock, which suits data centers. Constellation’s filings report a 20-year agreement with Microsoft supporting a plant restart, and Vistra’s report 20-year agreements to supply Meta with 2,609 megawatts.
What does Vertiv do for AI data centers?
It makes the power distribution and cooling equipment for racks, including liquid cooling, which high-density AI racks increasingly need. It grew revenue +35% over its latest four quarters.
Research and education, not investment advice: nothing here is a recommendation to buy or sell any security. Figures come from companies’ SEC filings as reported; companies named were chosen because they show how the stack works, not as a view on their shares.