Projects RH

All insights

Capital raising

AI Data Centre Offtake: What Makes a Contract Bankable?

AI Data Centre Offtake can determine whether a compute project is financeable. Learn what makes reservations, capacity agreements, take-or-pay structures and hyperscaler leases bankable.

Paul Raftery, 14 min readReviewed by Carmenza Hoyos Llano

AI infrastructure is creating extraordinary demand for data centres, power and high-performance computing capacity. But strong demand alone does not make an AI data centre financeable.

For lenders and infrastructure investors, the critical question is whether future demand has been converted into contracted, durable and enforceable cash flow.

That is where AI Data Centre Offtake becomes fundamental.

A reservation, a capacity agreement, a take-or-pay commitment and a long-term hyperscaler lease can all demonstrate customer demand, but they do not provide the same revenue certainty. Understanding the difference can determine whether a project remains speculative or develops into an asset capable of supporting project debt.

A lender does not finance headline demand. It finances a credible project's ability to generate and protect cash flow.

What Is AI Data Centre Offtake?

AI Data Centre Offtake is a contractual arrangement under which a customer commits to reserve, consume or pay for defined data centre or computing capacity under specified commercial and technical conditions.

Depending on the business model, the contracted product may include:

  • Critical IT load measured in MW
  • Data hall capacity
  • Rack capacity
  • GPU or accelerator capacity
  • Compute hours
  • Dedicated hosted infrastructure
  • Cloud computing services
  • Power, cooling and connectivity
  • A combination of infrastructure and compute capacity

The distinction matters because AI infrastructure can require substantial capital expenditure before meaningful revenue begins.

The International Energy Agency estimates that data centres consumed around 415 TWh of electricity in 2024, approximately 1.5% of global electricity consumption. This illustrates the increasingly infrastructure-intensive nature of digital and AI growth.

But increasing demand for electricity and compute does not automatically create bankable revenue.

The customer commitment must first be translated into a contractual structure that investors and lenders can underwrite.

Why AI Data Centre Offtake Matters for Project Finance

A developer may control attractive land, obtain access to power, design advanced cooling infrastructure and identify potential AI customers.

None of those achievements, individually, guarantees financing.

A lender needs to understand how the project will generate the cash required to:

  • Pay operating expenses
  • Maintain infrastructure
  • Service debt
  • Fund lifecycle expenditure
  • Meet financial covenants
  • Provide an acceptable return to equity

This means examining not only how much capacity customers want, but how much capacity they are legally committed to paying for.

The distinction can be expressed as:

Customer Interest → Reservation → Contracted Capacity → Accepted Capacity → Minimum Payment → Collected Cash

The further the project moves through that sequence, the stronger its potential revenue visibility becomes.

Reservation Agreements: Evidence of Demand

A reservation is generally the weakest of the four structures from a project-finance perspective.

A customer might reserve 20 MW, 50 MW or 100 MW of future capacity.

Commercially, this can be valuable. It demonstrates that there is genuine interest in the facility.

Financially, however, the reservation may have limited value if the customer can cancel without a material financial consequence.

Investors and lenders therefore need to establish:

  • Is the reservation legally binding?
  • Is there a reservation payment or deposit?
  • Is that payment refundable?
  • When does the reservation convert into committed capacity?
  • What conditions must be satisfied first?
  • Can the customer withdraw?
  • What happens if the development is delayed?
  • Are termination payments sufficient to protect committed capital?

The fundamental lesson is simple:

Reserved capacity is not automatically contracted revenue.

A project might advertise hundreds of megawatts of customer reservations while still having relatively little revenue that can confidently be incorporated into a debt case.

Capacity Agreements: Defining What Will Be Delivered

A capacity agreement moves the relationship further by defining what the developer or compute provider intends to make available.

For an AI facility, the agreement may establish:

  • Critical IT capacity
  • GPU type and quantity
  • Rack density
  • Power requirements
  • Cooling specifications
  • Network performance
  • Availability requirements
  • Delivery milestones
  • Expansion capacity
  • Service commencement

This improves technical certainty, but it does not necessarily resolve the financing question.

Consider two customers.

Customer A receives an option to use up to 100 MW.

Customer B commits to pay for a minimum 80 MW irrespective of whether it ultimately uses all that capacity.

Both may have signed capacity agreements, but their value to a lender could be fundamentally different.

Customer A represents potential demand.

Customer B may represent a contractual revenue floor.

Take-or-Pay Structures: Converting Capacity Into Revenue

A take-or-pay structure can substantially improve AI Data Centre Offtake because it requires the customer to pay for a defined minimum amount of capacity even when actual usage is lower.

Consider a simplified example.

A customer contracts 80 MW for ten years. During one year it only uses 60 MW.

Under an effective take-or-pay structure, the customer may nevertheless remain obligated to pay for the agreed minimum 80 MW.

The economic logic is straightforward.

The developer has built or reserved expensive infrastructure for that customer and may no longer be able to sell that capacity elsewhere.

A take-or-pay obligation therefore shifts part of the utilisation risk away from the infrastructure owner.

Instead of depending entirely on:

Available Capacity × Actual Utilisation × Variable Price

the financing case may contain a contractual floor based on:

Minimum Committed Capacity × Contracted Price

That distinction can materially influence bankability.

However, inserting the words "take-or-pay" into a contract does not automatically make the revenue bankable.

The lender must still examine the counterparty, payment mechanics, conditions precedent, termination rights, service obligations and enforceability of the agreement.

Hyperscaler Leases: Long-Term Revenue Visibility

A long-term lease with a financially strong hyperscaler can potentially provide one of the strongest foundations for AI data centre financing.

These structures may combine:

  • Defined critical IT capacity
  • Long contractual terms
  • Fixed or minimum base rent
  • Take-or-pay obligations
  • Capacity ramp schedules
  • Power commitments
  • Rent escalation
  • Energy pass-through mechanisms
  • Service-level requirements
  • Expansion options
  • Termination protections

The importance of these structures is already visible in publicly announced AI infrastructure transactions.

Applied Digital announced in May 2026 a 15-year take-or-pay lease with a U.S.-based investment-grade hyperscaler for 300 MW of critical IT load at its Polaris Forge 3 campus. The company reported approximately US$7.5 billion of base-term contracted revenue for the arrangement.

The transaction illustrates several characteristics that project-finance investors tend to value: defined capacity, long duration, minimum contractual economics and a strong counterparty.

The bigger lesson is not that every project needs a hyperscaler.

It is that scale alone does not substitute for contract quality.

A smaller project with durable contracted revenues could present a stronger financing proposition than a substantially larger project supported primarily by expressions of interest.

What Makes an AI Data Centre Offtake Contract Bankable?

There is no single clause that makes an offtake agreement bankable.

Bankability emerges from several contractual, technical and financial elements working together.

1. Clearly Defined Capacity

The agreement should identify precisely what the customer is purchasing.

For example:

"40 MW of critical IT load"

is materially clearer than:

"AI computing capacity."

For GPU infrastructure, further detail may be required around:

  • Accelerator technology
  • Cluster configuration
  • Network architecture
  • Storage
  • Cooling
  • Performance
  • Upgrade rights
  • Hardware substitution

The contracted service must be sufficiently clear to connect the commercial agreement to the project's technical design and financial model.

2. Objective Acceptance Tests

A bankable contract should establish when capacity is deemed ready and when the customer's payment obligation begins.

Important provisions can include:

  • Mechanical completion
  • Energisation
  • Commissioning
  • Performance testing
  • Customer acceptance
  • Acceptance deadlines
  • Deemed acceptance
  • Delay responsibility
  • Remedies for non-performance

Ambiguous acceptance can postpone revenue.

That creates a direct risk for a project whose debt service starts shortly after completion.

3. Minimum Payment Obligations

This is one of the most important elements of AI Data Centre Offtake.

The lender needs to know:

What is the minimum amount the customer is contractually obligated to pay?

The revenue floor may come from:

  • Take-or-pay obligations
  • Minimum capacity charges
  • Fixed base rent
  • Minimum committed spend
  • Minimum monthly payments
  • Minimum annual payments

Without an effective minimum commitment, the project may retain substantial utilisation risk.

4. Sufficient Contract Tenor

AI infrastructure is capital intensive, so contract duration matters.

But the headline term should never be analysed in isolation.

A 15-year agreement that the customer can terminate after two years at minimal cost could offer considerably less financing protection than its headline maturity suggests.

Lenders therefore consider:

Contract tenor + termination rights + payment obligations + credit quality

as an integrated risk.

5. Creditworthy Counterparty

A long-term commitment is only as dependable as the party responsible for paying it.

Due diligence should therefore determine:

  • The exact legal counterparty
  • Its financial strength
  • Whether the entity is investment-grade
  • Whether a parent guarantee exists
  • Whether letters of credit are available
  • Whether cash collateral is required
  • Whether additional credit support applies

This is particularly important in AI infrastructure.

A globally recognised technology brand may be associated with the project, but the entity that actually signs the agreement could be a different subsidiary or special-purpose entity.

Lenders underwrite the contractual obligor, not the headline brand.

6. Termination Protection

Termination provisions can fundamentally alter contract value.

Suppose an offtaker signs a ten-year agreement but can terminate after two years with only a small payment.

The legal headline may say ten years.

The lender may see something much closer to a two-year firm commitment.

A bankability review should therefore test:

  • Termination for convenience
  • Termination for cause
  • Early termination payments
  • Customer default
  • Provider default
  • Stranded equipment costs
  • Remaining contracted revenue
  • Remarketing rights
  • Cure periods

Strong contracts align termination economics with the capital invested to serve the customer.

7. Pricing and Indexation

The price mechanism must also be sufficiently predictable to support financial modelling.

AI infrastructure pricing can use metrics such as:

  • $/kW/month
  • $/MW/month
  • Rack/month
  • GPU/hour
  • Reserved GPU/month
  • Fixed capacity charge
  • Base rent plus energy costs

Contracts should also address:

  • Inflation
  • Escalation
  • Electricity costs
  • Demand charges
  • Cooling expenses
  • Taxes
  • Network charges
  • Hardware refresh
  • Price caps and floors

There is no universally superior pricing metric.

The essential issue is whether the economics can be translated reliably into forecast cash flow.

8. Availability and Service Levels

Data centre customers require high levels of availability.

Contracts may therefore include service-level agreements covering:

  • Power availability
  • Compute availability
  • Uptime
  • Cooling performance
  • Network performance
  • Maintenance windows
  • Service credits
  • Performance remedies

The Uptime Institute's Tier Classification System provides internationally recognised criteria for comparing data centre infrastructure performance, including power, cooling, maintenance and fault capability.

For lenders, however, the important financial question is:

How much revenue survives if availability falls below the contractual target?

Service credits and performance penalties should therefore be modelled under downside scenarios rather than treated solely as technical provisions.

9. Power and Compute Must Be Aligned

The bankability of compute offtake ultimately depends on power.

A project with a ten-year customer contract but only five years of secured electricity supply has a structural mismatch.

Developers should therefore align the critical chain:

Land → Power → Data Centre → Compute → Customer → Cash Flow → Debt

The International Energy Agency estimates that servers alone account for around 60% of electricity consumption in modern data centres on average, while cooling consumption varies materially by facility type.

AI Data Centre Offtake therefore cannot be assessed separately from grid availability, power procurement and cooling infrastructure.

10. Lender Rights and Security

If an offtake agreement supports project debt, lenders may require protections over the resulting contractual cash flow.

Depending on the transaction, these can include:

  • Assignment of receivables
  • Security over material contracts
  • Direct agreements
  • Notice of default
  • Cure rights
  • Step-in rights
  • Restrictions on material amendments
  • Restrictions on termination
  • Controlled accounts or cash waterfalls

The underlying principle is simple: a finance provider needs adequate protection over the contractual revenue on which its repayment depends.

Reservation vs Capacity Agreement vs Take-or-Pay vs Hyperscaler Lease

The distinctions can be summarised simply.

Reservation

"We want this capacity."

Useful as evidence of demand, but potentially weak for financing if it remains refundable or easily terminable.

Capacity Agreement

"We have agreed what capacity will be provided."

Greater technical definition, but payment commitments still need to be tested.

Take-or-Pay

"We will pay for a minimum amount of capacity whether or not we use it."

Potentially much stronger because the structure can create a contractual revenue floor.

Hyperscaler Lease

"We commit to defined infrastructure capacity over a long period under agreed lease and payment obligations."

Potentially highly financeable where the arrangement combines sufficient term, minimum payment obligations, termination protection and a strong counterparty.

Real-World Evidence: Contracted AI Compute Is Becoming Material

The market is already producing large contracts that demonstrate how strategic compute capacity has become.

CoreWeave disclosed that a September 2025 order form with OpenAI related to reserved cloud computing capacity included commitments of up to approximately US$6.5 billion through May 2031, subject to contractual termination provisions and delivery and availability requirements.

Separately, CoreWeave announced an earlier agreement with OpenAI in March 2025 with contract value of up to US$11.9 billion for AI infrastructure.

For project developers, these transactions illustrate an important point.

The AI infrastructure market is not merely discussing future demand. Major participants are entering substantial contractual commitments for dedicated computing capacity.

But lenders still need to examine the actual contractual mechanics behind those headline values.

The AI Data Centre Offtake Bankability Test

Before approaching lenders or infrastructure investors, a developer should be able to answer the following questions:

  1. Who is legally obligated to pay?
  2. What exact capacity is contracted?
  3. When does the payment obligation start?
  4. What is the minimum contractual payment?
  5. How long is that payment committed?
  6. Can the customer terminate early?
  7. What compensation applies following termination?
  8. How much can SLA penalties reduce revenue?
  9. Does secured power match the compute commitment?
  10. Can lenders obtain appropriate rights over the contract and cash flow?

If several of these answers remain uncertain, a project may have genuine customer demand without yet having bankable AI Data Centre Offtake.

From Customer Demand to Bankable Cash Flow

AI infrastructure development naturally attracts attention through megawatts, GPUs, cooling technologies and projected demand.

Project finance looks at the opportunity differently.

It asks whether the infrastructure can generate sufficiently predictable cash flow to support the capital required to build and operate it.

The progression matters:

Customer Interest → Reservation → Contracted Capacity → Accepted Capacity → Minimum Payment → Collected Cash

This distinction is becoming increasingly important as AI infrastructure moves deeper into institutional capital markets.

The most important question for a developer is therefore no longer:

"How much demand can we demonstrate?"

It is:

"How much revenue can an investor or lender responsibly underwrite?"

How Projects RH Approaches AI Data Centre Bankability

At Projects RH, we assess infrastructure opportunities from the perspective of capital readiness rather than headline project potential.

For an AI data centre, that means connecting:

Project Definition → Site → Power → Technical Capacity → Offtake → Financial Model → Capital Structure → Investor Documentation → Financing

A strong site and large customer pipeline may establish potential.

A strong investment case must show how that potential converts into defensible revenues, debt-service capacity and returns.

Before approaching the market, project sponsors should therefore reconcile the technical, contractual and financial assumptions supporting the investment case.

This includes ensuring that contracted capacity, commissioning dates, pricing, power availability, service obligations and financial-model assumptions tell the same story.

An AI data centre does not become bankable simply because customers want its capacity. It becomes bankable when the contracts supporting that capacity can credibly support the capital required to build it.

If you are developing an AI data centre or digital infrastructure project, start your investment readiness assessment to identify the technical, commercial, financial and documentation gaps that should be addressed before approaching capital providers.

Frequently Asked Questions

What is AI Data Centre Offtake?

AI Data Centre Offtake is a contractual arrangement under which a customer commits to reserve, consume or pay for defined data centre or computing capacity under agreed commercial and technical conditions.

Is a data centre reservation bankable?

Not necessarily. A reservation can demonstrate demand, but a lender will examine whether it creates binding payment obligations, how easily it can be cancelled and whether the developer receives adequate compensation if the customer withdraws.

What is the difference between a reservation and take-or-pay?

A reservation generally secures access to future capacity. A take-or-pay arrangement typically goes further by requiring payment for a minimum amount of contracted capacity even when actual utilisation is lower.

Why are hyperscaler leases attractive to lenders?

Long-term hyperscaler leases can provide defined capacity, extended revenue visibility and, where the tenant is financially strong, attractive counterparty credit. Bankability still depends on the detailed payment, acceptance and termination provisions.

Does a long contract automatically make AI Data Centre Offtake bankable?

No. A long stated tenor can be weakened by termination rights, conditions precedent, service-level deductions or weak counterparty credit. Lenders assess the enforceable economics, not simply the headline duration.

What should an AI data centre developer prepare before approaching capital?

At minimum, developers should establish a credible project definition, site and power position, technical design, contracting strategy, financial model, capital requirement, development schedule, risk analysis and investor documentation.

Is Your AI Data Centre Ready for Capital?

Strong demand for compute does not automatically make an infrastructure project investment-ready. Projects RH helps project sponsors assess bankability, strengthen financial models, prepare investor documentation and structure a credible path to capital.

If you are developing an AI data centre or digital infrastructure project, start your investment readiness assessment and identify the gaps that should be resolved before approaching investors or lenders.

Sources

What to do next

Projects RH prepares the company and the project first, then takes them to investors. The fastest way to find out where you stand is a short call.

Book a free 20-minute discovery callSend your documents