Where Does the Money Go When a $10 Billion Data Center Is Built?

Where does $10 billion go? Follow the data center money.

What does a “$10 billion data center investment” actually buy?

It sounds like one giant purchase.

It usually is not.

The money may be spread across land, buildings, electrical systems, cooling, servers, GPUs, networking, grid connections, professional services and financing. Some assets may be owned by the data-center operator. Others may be owned by a utility, landlord or outside investor.

A large data-center announcement is better understood as a network of money flows than as one construction bill.

Start With One Rule: Do Not Ask for the Pie Chart Too Early

There is no universal rule saying that every $10 billion data-center project spends the same percentage on GPUs, cooling, buildings or electricity.

Before asking “what percentage goes where?”, first ask:

What does the $10 billion number include?

Original Asset 1: The Four Boundaries Test

Program boundary — everything included in the public investment announcement.
Construction boundary — the cost of building and fitting out the physical facility.
Ownership boundary — which assets the data-center company actually owns.
Contract boundary — capacity the company secures through leases or long-term contracts rather than owning directly.

Two projects can both be called “$10 billion data centers” while using very different boundaries.

1. Construction Cost Is Not Total Program Cost

A construction-cost benchmark measures a defined set of building-related costs. It does not automatically include every dollar needed to make the project operate.

Turner & Townsend's 2025–2026 methodology includes items such as:

shell and core — the basic building structure,
architectural fit-out,
mechanical infrastructure — mainly cooling and related building systems,
electrical infrastructure — equipment that receives, protects and distributes power,
general-contractor costs,
and mechanical/electrical equipment.

But the same benchmark explicitly excludes land purchase, utility works, active IT equipment, professional fees and several site-related costs.^[1]

Construction cost
≠
Total data-center program cost

2. Inside the Construction Budget, Power and Cooling Are Huge

A data center is not just a warehouse that happens to contain computers.

It has to deliver large amounts of electricity safely and remove large amounts of heat continuously.

For its current U.S. liquid-cooled construction benchmark, Turner & Townsend shows this indicative construction-cost mix:^[2]

Construction-cost category	Indicative share
Electrical systems and equipment	48%
Mechanical systems and equipment	33%
Shell/core and architectural	9%
General-contractor requirements and fees	10%

Those percentages are useful—but only if we keep the boundary clear.

48% electrical does not mean 48% of every total data-center investment goes to electrical equipment. It means 48% inside this specific construction-cost benchmark.

The same source says liquid-cooled U.S. facilities currently show an average construction-cost premium of roughly 7% to 10% compared with similar-capacity air-cooled facilities.^[2]

3. Then Comes Active IT—the Equipment That Actually Computes

Active IT equipment means the hardware that performs or supports computing: servers, GPUs, CPUs, storage and networking.

That layer can be excluded from a construction index even though it may be one of the largest parts of the total economic program.

Power + cooling ready
        ↓
Servers
        ↓
GPUs / CPUs
        ↓
Networking + storage
        ↓
Usable AI compute

This is why the GPU is not the data center.

An expensive accelerator only becomes useful when the building can power it, cool it, connect it and keep it operating.

Microsoft's fiscal 2026 fourth quarter illustrates the asset mix. The company reported $41 billion of CAPEX, with roughly two thirds going to shorter-lived assets, primarily CPUs and GPUs. The rest went to longer-lived assets. Cash paid for property, plant and equipment was $35.8 billion.^[3]

So even inside one company's “AI CAPEX,” some assets may last much longer than others.

Original Asset 2: The Physical Money Flow

Land / site
     ↓
Building
     ↓
Electrical + cooling
     ↓
Active IT
     ↓
Grid / energy support
     ↓
Operations
     ↓
AI / cloud revenue
     ↓
Cash return

4. The Project Boundary Extends Outside the Fence

A finished data center cannot operate without enough electricity.

That can require assets outside the campus:

substations,
transmission or distribution upgrades,
new generation,
batteries,
and other grid-support systems.

Turner & Townsend's current industry survey found that 48% of respondents saw power availability as the most prominent obstacle to delivering data-center projects on schedule, with long connection lead times the main constraint.^[4]

But the data-center operator does not necessarily own or directly pay for every outside asset.

Original Asset 3: Inside the Fence / Outside the Fence

INSIDE THE FENCE
building
electrical distribution
cooling
IT equipment

OUTSIDE THE FENCE
grid connection
substation
transmission
generation
battery / flexibility

Economic dependence
does not require
direct ownership

5. A Long-Term Contract Can Cause Someone Else to Invest

A PPA, or power purchase agreement, is a contract under which a customer agrees to buy electricity under defined terms for a long period.

The customer may not own the power plant.

But the contract can make future revenue more predictable for the generator, which can help justify new investment.

Original Asset 4: Contract-to-CAPEX Chain

Data-center electricity need
        ↓
Long-term PPA
        ↓
Revenue certainty for generator
        ↓
Generator can justify CAPEX
        ↓
More power capacity / longer asset life
        ↓
Electricity available for data-center growth

Google and Fortum provide a current example.

Google announced a €13 billion Finland program for 2027–2028 covering digital infrastructure, clean-energy projects and partnerships—not simply one construction budget. It also signed a 22-year agreement connected to Fortum's Loviisa nuclear plant, added new wind commitments and contracted for a 94 MW battery system.^[5]

Fortum says the agreement can cover up to 50% of Loviisa's generation capacity during 2030–2049. Fortum also said it has an approximately €1 billion investment program for extending the plant's life, with about €700 million still pending investment decisions at the time of the announcement. The PPA provides long-term revenue certainty supporting those investments.^[6]

Google does not have to “buy the nuclear plant” for its demand to influence nuclear CAPEX.

6. Who Pays, Who Owns and Who Uses Can Be Three Different Parties

This is one of the most useful ideas in modern infrastructure finance.

Original Asset 5: Who Pays / Who Owns / Who Uses

Question	Possible answer
Who provides capital?	Hyperscaler, bank, bond market, infrastructure fund, private credit
Who owns the asset?	Hyperscaler, landlord, utility, generator, joint venture
Who uses the capacity?	Cloud company, AI customer, colocation customer

A lease is a contract that gives one party the right to use an asset owned by another party in exchange for payments.

A joint venture is a business structure in which two or more parties share ownership of a project or company.

In July 2026, Meta and BlackRock announced a venture for a data-center campus in El Paso, Texas. Funds managed by BlackRock will own 80% and Meta 20%. The parties described approximately $14 billion of development costs for buildings and long-lived power, cooling and connectivity infrastructure. Meta will lease the entire campus, while part of BlackRock's investment is supported by a $12.5 billion debt financing.^[7]

Debt financing means raising money that must later be repaid, usually with interest.

This example makes the separation visible:

Capital providers
        ↓
Joint venture owns asset
        ↓
Meta leases and uses campus
        ↓
Lease economics return cash
to capital providers

7. The First Company Paid Is Not the Last Company That Benefits

Suppose the data-center developer pays a general contractor.

A general contractor manages a large part of the construction work but usually hires specialist subcontractors and buys equipment from suppliers.

Developer
   ↓
General contractor
   ↓
Electrical / mechanical / civil contractors
   ↓
Transformer / switchgear / cooling / cable suppliers
   ↓
Components / materials / labor

The original payment continues through the industrial chain.

That does not mean the same economic value should be counted repeatedly. Each company records only its own revenue, costs and margin.

Margin is the portion of revenue left after specified costs. A company can receive a large order and still earn a weak margin.

8. An Order Is Not the Same as Revenue—and Revenue Is Not the Same as Cash

A supplier may announce a large data-center order long before it receives all the cash.

Backlog is contracted or committed work that has not yet been fully recognized as revenue.

Original Asset 6: Supplier Cash Ladder

Data-center CAPEX plan
        ↓
Supplier order
        ↓
Backlog
        ↓
Production / delivery
        ↓
Revenue
        ↓
Cash collection
        ↓
Profit / free cash flow

Cash conversion means how effectively accounting revenue or profit becomes actual cash.

That is why “this company has AI exposure” is not enough.

You still have to ask:

Did it actually win orders?
At what margin?
When does backlog turn into revenue?
When does revenue turn into cash?

9. Why a Flow Chart Is Usually Better Than a Pie Chart

A pie chart is useful only when the project has published a reliable, comparable cost breakdown.

Without that, a pie chart can create false precision.

A flow chart asks a more durable set of questions:

Who provides capital?
        ↓
Who receives cash?
        ↓
What asset is created?
        ↓
Who owns it?
        ↓
Who uses it?
        ↓
Who receives recurring revenue?
        ↓
Who carries the risk?
        ↓
What return comes back?

Original Asset 7: The 10-Question Money Flow Test

When you see a giant data-center investment announcement, ask:

What is the boundary of the headline number?
Whole program, construction only, one phase or several years?
Does it include active IT?
Servers, GPUs, CPUs, networking and storage?
Does it include land and utility works?
What is inside the fence and what is outside?
Who provides the financing?
Who owns each major asset?
Who actually uses the capacity?
Which suppliers receive purchase orders?
When do orders become revenue and cash?
What utilization and cash return are needed to justify the investment?

What This Money-Flow View Reveals

The bottleneck can capture value. If transformers, grid connections, cooling systems or networking are scarce, those constrained layers may gain pricing power or strong backlog.

The largest spending bucket may not earn the highest return. Large revenue does not automatically mean large profit.

Risk can move between balance sheets. A balance sheet is a snapshot of what a company owns and owes. Leases, joint ventures, PPAs and outside capital can shift ownership, debt and utilization risk from one balance sheet to another.

CAPEX is not finished when the asset is built. It creates value only when capacity is used and cash eventually comes back.

That connects directly to: What Is Capex? Why Spending More Can Make a Company Stronger—or Weaker.

Money committed
      ↓
Asset built
      ↓
Capacity available
      ↓
Capacity used
      ↓
Revenue
      ↓
Cash flow
      ↓
Return

The Takeaway

A $10 billion data-center program is not one transaction.

It is a network connecting finance, land, construction, power, cooling, compute, utilities, suppliers, contracts and eventual customer revenue.

To follow the money, ask three things at every layer: who pays, who owns, and who gets paid next.

Next in the Money Flow series: Revenue Is Not Cash Flow: Why Fast-Growing Companies Can Still Run Out of Money.

Key Vocabulary

Active IT equipment — servers, GPUs, CPUs, storage and networking that perform or support computing.
Shell and core — the basic structural building before specialized fit-out.
Mechanical infrastructure — cooling and related systems that manage heat and facility operation.
Electrical infrastructure — equipment that receives, protects, transforms and distributes power.
PPA — a long-term power purchase agreement.
Lease — a contract allowing one party to use an asset owned by another.
Joint venture — a project or company jointly owned by multiple parties.
Backlog — contracted work not yet fully recognized as revenue.
Cash conversion — the process by which revenue or profit becomes actual cash.

Sources

The $10 billion figure is an illustrative project size, not a universal cost split. Project boundaries, ownership, leases, utility investments and supplier contracts differ widely.

The Contexta

Search This Blog