InfraComputeThe capital behind compute.
Structure guide

Colocation versus build-to-suit: data centre models and how each is financed

Retail colocation, wholesale, hyperscale leases, build-to-suit and powered shell compared: who the customer is, how long the contract runs and which financing tools fit each model, with public deal examples.

data centres are run on a spectrum of business models, from retail colocation (many small customers, short contracts) to build-to-suit (one hyperscaler, long lease, signed before construction). Where a facility sits on that spectrum decides who takes the risk, how long the cash flow is contracted and which financing fits.

The five models

Model Customer Typical contract Operator's role Typical financing
Retail colocation Many enterprises, a few racks each 1 to 5 years Runs everything: power, cooling, security, connectivity Corporate debt, ABS once stabilised
Wholesale colocation Fewer, larger tenants taking halls or floors 5 to 15 years Runs the facility, tenant fits out its space ABS, CMBS, bank loans
Hyperscale lease Hyperscalers taking whole buildings 10 to 15+ years Landlord and operator of a dedicated building Construction loans, ABS, master indentures
Build-to-suit One hyperscaler or AI customer, built to spec Long lease signed before build Developer and landlord Construction loan, then lease-backed bond or JV
Powered shell Tenant fits out everything inside Long ground or shell lease Real estate landlord Real estate-style financing

Retail and wholesale colocation

The traditional data centre business. The operator owns a building, buys power and rents capacity to many customers. Revenue is diversified but contracts are shorter, so lenders look at the whole portfolio rather than any single lease.

These portfolios suit securitisation once they are stabilised (built, leased and operating). CyrusOne's 2025-1 ABS, for example, was reported as backed by 13 stabilised data centres with 98 tenants, 83.5% of them investment grade. Diversity of tenants and buildings is what supports the rating.

Hyperscale leases

As cloud demand grew, operators began leasing entire buildings to single hyperscalers on long terms. Revenue is concentrated but the credit is strong. The QTS Thunder master indenture, rated A+ by KBRA in 2026, is backed by 12 data centres with 493.5 MW, 100% leased to investment-grade hyperscalers. Concentration risk is offset by tenant quality.

Build-to-suit

In a build-to-suit, the customer commits before a shovel goes in the ground. The developer builds to the customer's design, often on a campus planned around a large power connection.

The financing usually runs in two stages:

  1. Construction. A bank construction loan sized on loan-to-cost, drawn as the building goes up. Galaxy's Helios loan for capacity leased to CoreWeave, for example, was reported at up to 80% loan-to-cost with a 1.40x DSCR test after stabilisation.
  2. Take-out. Once the lease starts, the construction debt is refinanced with longer-term capital: a lease-backed project bond, an ABS issue or a joint venture.

The largest public example is Meta's Hyperion campus in Louisiana: a joint venture with funds managed by Blue Owl, financed with US$27.3bn of amortising notes and leased to Meta under 11 triple-net leases, with a residual value guarantee covering the risk of non-renewal. See hyperscaler backstops and residual value guarantees.

AI-specific leases at former bitcoin mining sites

A newer variant: former bitcoin miners with large power connections lease purpose-built AI halls to neoclouds or hyperscalers. Applied Digital (leased to CoreWeave and, at Polaris Forge 2, Oracle), Core Scientific (colocation licences with CoreWeave), Cipher, TeraWulf and Hut 8 (leases to Fluidstack, backstopped by Google) have all financed these sites with secured high-yield or investment-grade bonds. Pricing follows tenant credit: Applied Digital's Polaris Forge 2 bond, with Oracle (rated BBB) as tenant, priced at 6.75%, against 9.25% for Polaris Forge 1, leased to CoreWeave. See lease-backed project bonds.

Powered shell and powered land

At the real estate end of the spectrum, the landlord provides land, a building and a power connection, and the tenant installs all the mechanical and electrical equipment. This shifts most of the technical and cost risk to the tenant, so the landlord's financing looks like real estate lending. As power has become the main constraint, "powered land" (land with a secured grid connection) has become a financeable asset in its own right.

How the model changes the term sheet

Term Colocation portfolio Build-to-suit / single tenant
Sizing Portfolio DSCR and LTV Lease cash flow and tenant credit
Key diligence Tenant mix, churn, renewal history Lease terms, termination rights, construction risk
Construction risk Low (stabilised assets) High until completion; covered by completion guarantees
Main protection Cash trap and amortisation triggers Lease, backstops, RVGs, completion support
Refinancing Recurring ABS issuance at the ARD Construction take-out, then amortising or bullet bond
Common questions

Questions clients ask

What is the difference between colocation and build-to-suit?
In colocation, an operator builds a facility and rents space, power and cooling to many customers. In build-to-suit, a developer builds a facility to one customer's specification, usually leased to that customer for a long term before construction starts.
What is a powered shell?
A building with power and connectivity delivered to it but without the internal mechanical and electrical fit-out, which the tenant installs itself. The landlord's role, and so its financing, is closer to real estate.
Which financing suits a colocation portfolio?
Stabilised multi-tenant portfolios are typically financed with data centre ABS or CMBS. Build-to-suit projects are financed with construction loans that are later refinanced with lease-backed bonds or securitisation.

For in-house counsel

Take it to the documents

The filed wording behind this piece, for the lender’s or the borrower’s side. Previews are open; the full text is free with InfraCompute Weekly.

Evidence

Sources

Every figure is taken from the public document linked. Links open the filing or release itself.

  1. CyrusOne Data Centers Issuer I 2025-1, Asset Securitisation Report ↗
  2. KBRA: QTS Thunder master indenture (preliminary rating, 2026) ↗
  3. Meta press release: Hyperion joint venture (October 2025) ↗
  4. Galaxy Helios (CoreWeave) credit agreement (S-1 Ex 10.23) ↗
  5. Core Scientific 7.75% senior secured notes indenture ↗
  6. Applied Digital Polaris Forge 2 6.75% senior secured notes indenture ↗

General information only, drawn from public documents and press reports, which are cited. It is not legal, financial or investment advice. Mention of a company or transaction does not mean the author or any firm acted on it, and nothing here is a view on any company's credit. Views are personal and are not those of any employer or client.

Randeep Bubbra

InfraCompute Weekly

The week in compute finance, read from the filings.

Deals, terms and the clauses behind them, for the people financing data centres, GPU clouds and the power behind them. By Randeep Bubbra.

Working on a GPU or data centre financing? Book 20 minutes with Randeep to compare your terms with the public deals.

Fridays, one email. Work email only. Unsubscribe any time. Privacy.