When a hyperscaler announces a new AI data center, the headline usually focuses on the AI company.
But the spending spreads much farther.
Someone buys the land.
Someone builds the halls.
Someone supplies transformers, cables, UPS systems, cooling equipment, generators, and electricity.
Someone rents space and power capacity.
And many of those companies keep getting paid long after the first server turns on.
The AI data-center boom is not one investment theme. It is a chain of contracts, equipment orders, leases, and recurring services.
By the end of this article, you will be able to follow a data-center dollar through the value chain, distinguish one-time project revenue from recurring revenue, and read unfamiliar investor terms such as backlog, book-to-bill, margin, and free cash flow without treating every “AI infrastructure” company as the same business.
Start with the Money Flow, Not the Ticker
A large AI data center begins with capital spending—money used to build long-lived assets such as buildings, electrical systems, and cooling equipment.
Before the site opens, the owner may pay for:
- land and buildings
- grid access
- transformers and switchgear
- power and fiber cables
- UPS systems and batteries
- cooling equipment
- backup generation
- construction and commissioning
After the site opens, a different stream of payments continues:
- electricity
- rent
- network connections
- maintenance
- replacement equipment
- software and controls
- service contracts
This gives us two broad revenue types.
Project revenue is money tied to building or expanding a site. It can arrive in large bursts.
Recurring revenue is money that repeats after the site opens—such as rent, power sales, interconnection fees, maintenance, or service contracts.
The Contexta Revenue Timing Map
| When | Typical spending | Main question |
|---|---|---|
| Before opening | construction, grid, transformers, cables, cooling, backup power | Can orders turn into revenue and profit on time? |
| After opening | rent, electricity, interconnection, maintenance, service | How durable is the recurring cash flow? |
| During upgrades | denser racks, liquid cooling, electrical upgrades, replacement equipment | Does AI create repeat spending after the first build? |
1. Data Center Platforms: Equinix and Digital Realty
Equinix and Digital Realty operate large data-center platforms.
Both are REITs, or real estate investment trusts—companies built around owning income-producing real estate and distributing much of their taxable income to shareholders under REIT rules.
They sell secure space, power capacity, and network connections.
A common model is colocation: a customer rents space and power in a professionally operated data center instead of owning the whole facility.
Another important product is interconnection: a direct private connection between customers, networks, cloud providers, or platforms inside the data-center ecosystem.
Equinix reported record net interconnection additions in Q2 2026 and said annualized gross bookings rose 23% year over year, contributing to record backlog.[1]
Digital Realty reported a record $1.9 billion of annualized GAAP base-rent backlog at 100% share in Q2 2026.[2]
Here, backlog means contracted future business that has been signed but has not yet fully started generating reported revenue.
The appeal is recurring contractual revenue.
The risk is capital intensity—the business must spend a lot on land, buildings, power, and financing before those assets are fully occupied.
Watch: leasing, occupancy, power availability, development spending, debt, and cash flow.
2. Power and Cooling: Vertiv and Schneider Electric
Servers cannot run directly from a normal office-building electrical system.
They need systems that convert, protect, distribute, and cool large amounts of power.
A UPS, or uninterruptible power supply, is a battery-backed system that keeps critical IT equipment running through short power interruptions.
Switchgear is equipment that safely controls, protects, and isolates electrical circuits.
Vertiv and Schneider Electric sell these kinds of systems, along with power distribution, controls, batteries, and cooling.
AI increases the importance of this layer because rack power density—the amount of electrical power concentrated in one server rack—is rising.
More power in the same rack creates more heat in the same space.
That pushes demand toward liquid cooling, which moves heat using a liquid loop rather than relying only on room air.
Vertiv reported Q2 2026 net sales up 24% year over year and adjusted operating margin of 22.6%.[3]
A margin is the share of revenue left after a defined set of costs. Higher margins generally mean the company keeps more profit from each dollar of sales, though the exact margin definition matters.
Schneider Electric said data-center demand remained very high in the first half of 2026 and that Data Center & Networks continued to lead demand across its end markets.[4]
Watch: orders, cooling exposure, margins, factory capacity, and service revenue.
3. Grid Equipment: GE Vernova and Siemens Energy
The data-center building may be ready before the power system is ready.
New large loads can require substations, transformers, high-voltage equipment, grid controls, and new generation.
GE Vernova and Siemens Energy sit closer to that grid bottleneck than to the server rack.
GE Vernova said its Electrification segment booked $2.4 billion of equipment orders to support data centers in Q1 2026—more than in all of 2025.[5]
Siemens Energy reported €17.6 billion of total Q1 FY2026 orders, a €146 billion order backlog, and 22 GW of gas-turbine commitments related to data centers.[6]
Siemens also reported a book-to-bill ratio of 1.82.
Book-to-bill means new orders divided by revenue recognized during the same period. Above 1.0 means new orders arrived faster than revenue was delivered in that period.
That can signal strong demand.
But a large backlog is not the same as guaranteed profit.
Delivery delays, raw-material inflation, contract terms, and execution problems can reduce the value of that backlog.
Watch: book-to-bill, backlog, transformer and turbine capacity, delivery times, margins, and cancellations.
4. Cables and Construction: Prysmian and Quanta Services
Equipment does not install itself.
A data center needs both power cables and fiber-optic cables—the thin glass strands that carry digital data as light.
Prysmian supplies both energy and digital connectivity products.
In Q2 2026, Prysmian said its Digital Solutions organic growth accelerated to 18%, helped by optical-cable demand, while its Industrial & Construction business also cited North American data-center demand.[7]
Quanta Services designs, builds, upgrades, and maintains transmission, substations, generation, and other large-load infrastructure.
Quanta reported a record $48.5 billion total backlog at the end of Q1 2026.[8]
Again, backlog is future contracted or expected work, not profit already earned.
Construction companies must still manage labor, materials, contract pricing, schedule, and cost overruns.
Watch: backlog quality, skilled labor, contract type, project margins, raw-material costs, and free cash flow.
5. Electricity Supply: Constellation and NextEra Energy
A data center buys transformers once.
It buys electricity every hour it operates.
This creates a recurring revenue opportunity for power producers and utilities.
Constellation’s Calpine business signed a 380 MW agreement with CyrusOne in Texas in February 2026 and said its Texas agreements with CyrusOne totaled more than 1,100 MW.[9]
NextEra Energy is developing large data-center hubs together with generation and grid infrastructure. Its 2026 investor materials describe a base case of about 15 GW of data-center hubs by 2035, with a larger upside case.[10]
A regulated utility is a power company whose rates and investment recovery are overseen by a government regulator. That means large new investments do not automatically translate into shareholder profit; regulators decide how and when certain costs can be recovered from customers.
Power demand can be strong while returns are weak if projects are overpriced, delayed, poorly financed, or subject to unfavorable regulation.
Watch: contracted MW, power prices, generation cost, regulatory recovery, capital spending, debt, and project timing.
6. Backup Power and Thermal Systems: Caterpillar and Trane Technologies
Data centers need power when the grid fails and cooling every minute the servers are running.
Caterpillar supplies generators and turbine-based power systems.
It is also working with Vertiv on integrated onsite power and cooling designs intended to reduce deployment time and grid dependence.[11]
Trane Technologies sells large cooling and thermal-management systems.
Thermal management simply means controlling and removing heat so equipment stays within safe operating temperatures.
Trane completed its acquisition of LiquidStack in March 2026, adding direct-to-chip and immersion liquid-cooling technology.[12]
Direct-to-chip cooling sends liquid to a cold plate attached to the processor.
Immersion cooling places computing hardware in a non-conductive cooling liquid.
These technologies become more relevant as rack power density rises.
Watch: data-center revenue exposure, equipment orders, service contracts, cooling adoption, and segment margins.
The Contexta Contract Chain
The value chain becomes easier to remember if we follow the contract:
Land / Platform → Power & Cooling → Grid Equipment → Cables & Construction → Electricity → Backup & Service
Different layers get paid at different times.
Different layers also face different risks.
12 Companies Across the Value Chain
| Layer | Examples | How revenue arrives | What can go wrong |
|---|---|---|---|
| Platforms | Equinix, Digital Realty | rent, power capacity, interconnection | high debt/capex, slow leasing |
| Power & cooling | Vertiv, Schneider Electric | electrical/cooling equipment and service | competition, margin pressure |
| Grid equipment | GE Vernova, Siemens Energy | generation, transformer, high-voltage orders | long projects, execution risk |
| Cables & construction | Prysmian, Quanta Services | cable sales and infrastructure contracts | labor, material, fixed-price risk |
| Electricity supply | Constellation, NextEra Energy | power sales and long-term contracts | regulation, debt, construction cost |
| Backup & thermal | Caterpillar, Trane Technologies | generators, cooling, controls, service | data centers may be only part of company-wide revenue |
These are representative examples, not a ranking.
The table is useful only if the reader keeps the business models separate.
Why “AI Exposure” Is Not Enough
One of the most common investor questions is:
Which company benefits most from AI data-center spending?
That question is too early.
First ask:
- Exposure: How much of the company’s revenue is actually tied to this market?
- Contract evidence: Are customer announcements becoming signed orders or leases?
- Revenue conversion: How fast does backlog become reported revenue?
- Margin: Does demand improve profit per dollar of sales?
- Cash conversion: Does accounting profit become real cash after capital spending?
- Balance sheet: How much debt or new capital is needed to fund growth?
- Valuation: How much future success is already reflected in the stock price?
Free cash flow is the cash left after a business pays for operating needs and capital spending under the company’s chosen calculation. It matters because reported profit can rise while cash remains weak.
The Contexta Bottleneck-to-Cash Test
Instead of starting with a ticker, trace four steps:
Bottleneck → Contract → Revenue → Cash
Bottleneck: Is this product or service genuinely hard to replace or slow to supply?
Contract: Are customers signing real orders, leases, or power agreements?
Revenue: When will those contracts show up in reported sales?
Cash: After factories, construction, debt, and working capital, how much cash remains?
This test helps explain why a company can have excellent AI exposure and still produce disappointing shareholder returns.
A Good Company Is Not Automatically a Good Stock
A company may have:
- strong orders
- a useful product
- a growing backlog
- high-profile AI customers
and still disappoint investors.
Possible reasons include:
- the expected growth is already reflected in the share price
- revenue grows but margins fall
- the company needs too much debt or capital spending
- backlog is delayed or canceled
- cash flow lags accounting profit
- data centers are only a small part of the company
Industry growth answers:
Is demand increasing?
Investment analysis asks:
How much of that demand becomes durable cash flow for this company, and what price are investors already paying for it?
What Should You Watch Next?
- Backlog conversion: Are record order books becoming revenue on schedule?
- Cooling shift: How quickly does liquid cooling move from premium deployments to standard AI infrastructure?
- Grid bottlenecks: Do transformer and turbine shortages improve supplier pricing or create delivery failures?
- Recurring mix: Which companies add service, rent, or power revenue after the build?
- Capital intensity: Which business models need the most debt and upfront investment?
- AI exposure: Is data-center growth large enough to change the whole company, or only one segment?
The Main Idea
AI data centers move money through many industries.
Some companies sell one-time equipment.
Some build projects.
Some collect rent, power revenue, interconnection fees, and service revenue for years.
Do not start with “Which AI stock?” Start with “Where is the bottleneck, who holds the contract, when does revenue arrive, and how much cash remains?”
Continue Reading
- What Must Be Built to Power the AI Data Center Boom? — understand the physical systems before following the money.
- Cheap Power Is Not Always Cheap — see how grid timing changes economics.
- Why Power, Not Chips, May Limit the AI Data Center Boom — see why grid bottlenecks can redirect spending.
Key Terms
- capital spending (capex): money used to build or buy long-lived assets
- recurring revenue: revenue that repeats through rent, power, service, or ongoing contracts
- REIT: real estate investment trust, a structure for owning income-producing real estate
- colocation: renting data-center space and power instead of owning the whole facility
- interconnection: direct private links among customers, networks, and cloud platforms
- UPS: battery-backed system that bridges short power interruptions
- switchgear: equipment that controls, protects, and isolates electrical circuits
- rack power density: electrical power concentrated in one server rack
- backlog: contracted future business that has not yet fully become reported revenue
- book-to-bill: new orders divided by revenue in the same period
- margin: the share of revenue remaining after specified costs
- free cash flow: cash remaining after operating needs and capital spending under the stated calculation
Sources
- Equinix — Q2 2026 results.
- Digital Realty — Q2 2026 results.
- Vertiv — Q2 2026 results.
- Schneider Electric — 2026 financial results.
- GE Vernova — Q1 2026 results.
- Siemens Energy — Q1 FY2026 analyst presentation.
- Prysmian — Q2 2026 results.
- Quanta Services — Q1 2026 results.
- Constellation / Calpine — CyrusOne Texas agreement.
- NextEra Energy — 2026 investor presentation.
- Caterpillar / Vertiv — onsite power and cooling collaboration.
- Trane Technologies — LiquidStack acquisition.
Status checked September 30, 2026. The Revenue Timing Map, Contract Chain, and Bottleneck-to-Cash Test are The Contexta analytical frameworks. The companies are representative examples across the value chain, not a ranking or recommendation. Company-level exposure to data centers varies and should be checked against current financial disclosures.