<

The Barn · Saline Township, Michigan

Understanding AI Data Centers

A Caracal Global briefing on how a $16 billion campus gets built, paid for, and explained.

A $16 billion AI campus broke ground in a Michigan farm township in June 2026. It cleared every legal and regulatory hurdle and still arrived carrying a trust deficit no groundbreaking could retire.

That gap — legally unstoppable, socially contested — is not an engineering problem. It's a communications one.

This page is a working reference on how a gigawatt-scale AI data center actually gets built, paid for, and explained: who pays whom, where the $16 billion really goes, and why the region that powered the last industrial century holds exactly what the AI century needs. We built it around The Barn because it's the first large instance of a pattern the Great Lakes will see repeatedly.

It is also a demonstration. This is the discipline Caracal brings to infrastructure that has to win a room before it can break ground.


01 · The Visual

The money flow, on a napkin

A quick visual of how a deal like this is structured and where the dollars go. The headline is $16 billion; the point is that nearly six of every ten of those dollars buys chips, not building. Rough strokes, not contracts — but enough to reset what the number actually means.

The Barn — Following the Money
The Barn  /  Saline Township, Michigan  ·  Napkin math, not contracts

Following the money

A back-of-the-napkin sketch of how a gigawatt-scale AI campus gets paid for. The $16 billion headline is real — but the point here is simpler: nearly 60% of the spend is chips, not the building. Here is who pays whom, where the dollars go, and how a deal like this earns a return. Rough strokes, not audited figures.

Verified from public reporting and developer disclosures
Illustrative — industry benchmarks, not Barn-specific terms
01
Who pays whom

A build-to-suit arrangement. The developer owns the building and its systems; the tenant runs it; the tenant's customer ultimately funds the demand.

Developer / Owner
Related Digital
Builds and owns the campus: structure, power infrastructure, cooling systems. Carries the construction cost.
leases to
Tenant / Operator
Oracle
Leases the campus, operates the compute, sells cloud capacity onward. Pays rent to occupy.
sells to
End Customer
OpenAI
Buys the computing power to train and run AI products. The demand that funds the whole chain.
Money flows back ←
OpenAI pays Oracle for compute. Oracle pays to occupy the campus. The developer earns a return on what it spent to build.
Stays local in Michigan
The building, the jobs, the energy investment, the tax base. The durable value, not the sticker price.
Not disclosed
Rent, the compute price, lease term, and who owns the chips are private commercial terms.
02
Where the $16 billion actually goes

It is not a warehouse with servers. The computers are the expense. The building, land, and even the substantial power systems are the minority of the spend.

AI chips & servers Nvidia GPUs, networking
~60%
Power infrastructure Substation, battery, electrical
~18%
Building & cooling Shell, mechanical systems
~15%
Land & site Acquisition, prep, open space
~7%
The honest headline: roughly six of every ten dollars buys silicon that is replaced every three to six years and is bought from Nvidia regardless of where the building sits. The durable Michigan value is the infrastructure, the energy investment, and the recurring reinvestment. Shares are illustrative, drawn from AI data-center cost studies, not Barn-specific figures.
03
How a deal like this earns out

An illustrative model, not The Barn's real terms, which are private. This shows how the math works in general for a gigawatt-class campus.

Verified
$16B
Total project figure cited at groundbreaking. Earlier estimates ran $7–10B. Largest economic investment in Michigan history.
Verified
1+ GW
Gross capacity across three 550,000 sq ft buildings. Among the most significant data-center projects in the U.S.
Illustrative
~$40k / kW
Industry benchmark all-in capex for AI-heavy capacity. Over half is the GPUs. Implies the chip-dominant split shown above.
Illustrative
3–6 yrs
Useful life of the chips before refresh. The building lasts 20–30+ years and gets re-fitted several times over.
Illustrative
Long lease
A multi-year tenant commitment is what justifies billions upfront. The developer's return rides on the lease holding.
Illustrative
Phased
Built in stages so capacity scales up with demand. Scaling down is hard: the risk is an underused campus, not a smaller one.
04
Why here, and why not all in one place

"Data center" hides two opposite jobs with opposite location rules. Untangling them explains why a Michigan farm township ends up hosting a hyperscale campus.

Training
Goes where the inputs are
Building the AI model: tens of thousands of chips running for weeks. Nobody is waiting in real time, so location is free to chase cheap power, water, land, and a willing community. This is mostly what The Barn does.
Serving
Goes where the people are
Answering you when you type: every mile adds lag you can feel. Has to sit near population centers, spread across many cities. This is why you cannot put everything in one state.
Why Michigan, specifically: a utility able to supply a gigawatt, water, a cool climate that helps with cooling, land, and a skilled trades workforce. The region that powered the last industrial era has the exact inputs the AI era is short on. Why not one giant campus: a single site can only pull so much power before it strains the grid; one failure cannot be allowed to take everything down; and each region's affordable power and water run out, pushing the next campus somewhere new. The pattern is concentrated campuses, distributed across many regions.
Sources: Project figures from groundbreaking coverage and Related Digital disclosures (June 2026). Cost-share and economic benchmarks from public AI data-center cost analyses, presented as illustrative ranges. Commercial terms between Related Digital, Oracle, and OpenAI are not public. Verify all figures before publication.
Caracal Global
Always Be Communicating

02 · The Explainer

The Barn, explained

A plain-language guide to what the project is, how it works, what it costs, and what it means for the community — written to answer the questions people are actually asking, from “why does it cost $16 billion” to “will it raise my electric bill.” Honest on both columns: real benefits, real impacts, named directly.


03 · The Discipline

The communications discipline behind the project

Two briefings on why infrastructure that’s legally approved still has to be socially earned. Read them in either order — both argue the same thing: do the reading before the engineering.


Building the next one?

If you’re developing, financing, or operating AI infrastructure that has to win, influence, and neutralize stakeholders before it can win a permit, that’s the communications Caracal Global does.

Book a conversation →

Caracal Global helps leaders understand the world and how to talk about it.