When problems arise, however, that system can quickly break down into its individual contracts. A construction delay may be excused under one agreement but trigger termination rights under another. A power purchase agreement (PPA) may permit interruptions at the same time as a customer contract requires continuous service. Lenders may have security over key assets but lack the rights needed to keep them operating.
As a result, disputes often arise not from individual contracts, but from gaps between them.
Construction relief may not protect project revenue
An AI data centre is not economically complete when its building is finished. It still requires electricity, cooling, connectivity and computing equipment, usually within an agreed delivery timetable. Delay in one workstream can prevent the project from meeting completion requirements elsewhere.
If customer payments depend on completion, delay can postpone the revenue expected to service debt while simultaneously triggering contractor claims, longstop dates or termination rights.
The project experiences one event, but different contracts can assign different legal consequences to it. For sponsors and lenders, completion tests, extensions of time, force majeure provisions, termination rights and compensation mechanisms therefore need to be assessed across the contractual structure rather than on an agreement-by-agreement basis.
A related issue is the alignment between the construction contract and the wider project agreements it is intended to support. What a contractor accepts by way of relief may not correspond with what the customer or lender requires under the project documents.
Where a construction contract allows a longer delay than the project agreement permits, or excuses an event that a customer contract treats as a service failure, the project company bears the risk. These gaps often arise because the agreements are negotiated separately, at different times and with different counterparties.
Power must be bankable, not merely available
For financing purposes, the key question is not simply whether sufficient electricity exists in a market. It is whether the project has durable rights to access that electricity.
A site may be close to substantial generation capacity but lack firm connection rights, allocated capacity or long-term supply arrangements capable of supporting project finance. A data centre may also be completed before the transmission infrastructure needed to operate it at the intended scale.
A grid arrangement may permit delay or curtailment while the customer contract requires availability and the financing assumes continuing revenue. Where those positions are not aligned, the project company may bear the consequences.
Curtailment illustrates the issue clearly. A grid connection agreement will typically allow the network operator to curtail supply in defined circumstances. That may be a contractually authorised event rather than a breach. However, the customer contract may not adequately address curtailment risk, while the financing assumptions may underestimate its impact.
The result is that an event permitted under one contract can trigger problems under another. Unless the customer contract and financing documents properly address curtailment risk, the project company bears the loss.
Contracted revenue can become contested revenue
Compute offtake and colocation arrangements can provide the cashflow foundation for AI infrastructure financing. However, a take-or-pay commitment is only the starting point.
The commercial value of compute depends on more than the existence of physical capacity. Performance, electricity, cooling, connectivity, software and cybersecurity can all affect whether the contracted service is actually being delivered.
In practice, cashflows are influenced by acceptance testing, service levels and ongoing performance obligations. Service credits may apply to some failures, while more serious failures may give customers rights to suspend or terminate agreements. Liability caps and exclusions may also determine how much loss can ultimately be recovered.
Where projected customer revenues are used to size debt facilities, the provider's ability to deliver the contracted service is just as important as the customer's willingness and ability to pay.
A contract can improve visibility of future cashflows without making those cashflows unconditional.
Credit support must match the exposure
Guarantees, backstops and other forms of credit support can improve bankability, but the strength of the support provider and the scope of the support obligation are different issues.
AI infrastructure financing arrangements may involve a developer entering into long-term arrangements with a compute customer, supported by a third party that guarantees specified customer obligations. Such arrangements can improve the financing value of a customer contract without transferring all project risks.
Lenders therefore need to understand exactly what obligations are covered, when support applies, what triggers payment and how the support can be enforced.
Completion, commencement of customer payments and the operation of credit support may all be governed by different provisions. As a result, the allocation of risk can change as the project moves from construction into operation.
That is not necessarily a weakness. Different risks may deliberately be allocated to different parties at different stages. However, lenders need to understand precisely when those transitions occur.
Security does not necessarily preserve value
AI infrastructure presents a challenge that is less common in traditional infrastructure projects.
Computing equipment may lose economic value much faster than the building, power connection or debt supporting it. New hardware can quickly reduce the competitiveness and residual value of previous generations even while those assets remain operational.
A lender may hold valid security over computing equipment but find that the equipment has lost value or depends on power, cooling, software, maintenance and connectivity arrangements that do not transfer with it.
The same issue applies to the facility itself. A specialised data centre may have substantial value when supported by a strong customer relationship and significantly less value if that customer departs.
Lenders therefore need to identify not only the assets covered by their security package but also the operating rights that must survive enforcement. Step-in rights, assignment provisions and change-of-control protections should be tested against realistic distress scenarios involving replacement owners, operators or customers rather than only the project's intended operating model.
A security package that captures the assets but not the wider operating ecosystem may preserve legal rights while losing much of its economic value.
Government action can fragment the business
AI data centres depend on planning approvals, grid access, electricity regulation, data governance rules, technology availability and, in some jurisdictions, foreign investment approvals. Environmental requirements relating to energy, water and land use can also affect construction and operation.
Regulatory action may constitute a change in law under one contract but not another. It may excuse a technology supplier while leaving the operator exposed to claims from its customers. Ownership or control restrictions may also affect whether a purchaser or lender can exercise rights that appeared available when the financing was signed.
The key question is whether its consequences have been allocated consistently across the project documents.
Distress can expose the gaps
In AI infrastructure projects, property, computing equipment, customer contracts, power arrangements and operating rights may support a single business while being held by different entities. However, special purpose vehicles may separate legal liabilities, but they cannot eliminate commercial dependencies.
Certain financing structures may give rise to heightened concentration and contagion risks. Changes in market sentiment, bondholder enforcement action, regulatory investigations or significant defaults could have wider implications where financing arrangements are linked through cross-default provisions, guarantees or other contractual dependencies. These risks may be amplified where circular financing arrangements create mutual financial exposures, potentially allowing financial distress affecting one entity to spread across the wider structure.
Cross-border AI infrastructure projects present another challenge. Financing, land, equipment, power and customer agreements may be governed by different laws and subject to different dispute resolution processes. Export controls and sanctions can also affect projects far beyond the jurisdiction that introduced them.
These factors can make restructuring particularly difficult because insolvency and administration processes typically focus on individual legal entities rather than interconnected commercial arrangements.
For sponsors and lenders, that means testing projects against failure as well as completion. They need to understand where losses may fall, which contract provides relief, whether that relief matches exposures elsewhere in the structure, and what remains capable of generating cash following enforcement.
In a distressed scenario, lenders will need to determine quickly whether key contracts can be assigned or stepped into, whether power and technology rights survive enforcement, and whether a replacement operator can retain the licenses and customer relationships needed to continue the business.
The key bankability challenge is not whether each contract works independently. It is whether the contracts still work together when one or more do not.