GPU financing requires more than a valuation of the hardware—it demands a clear understanding of repayment and the protections available if things go wrong
Artificial intelligence (AI) and data center growth have accelerated since 2025. According to the International Energy Agency’s 2026 report, “Key Questions on Energy and AI,” capital expenditure by the largest technology companies exceeded US$400 billion in 2025 and is expected to increase by 75 percent in 2026. These figures reflect broader capital expenditure, not just purchases of graphics processing units (GPUs). Strong demand for AI is driving major investment in GPUs, which help train and run AI models once they are deployed. GPUs are specialized processors capable of parallel processing, allowing them to perform many calculations simultaneously and making them central to AI performance.
The capital required to acquire GPUs in a highly competitive market creates significant opportunities for financiers. However, GPUs cannot function or generate revenue in isolation. Their utility and revenue-generating capacity depend on the systems into which they are integrated. For example, in data centers, GPUs operate within servers supported by memory, storage, high-speed networking, power and cooling.
The central questions for a financier are whether the transaction supports repayment as expected and if repayment fails, whether the transaction provides effective remedies. Answering those questions requires assessing the hardware, cash flows and practicalities of recovery. Against this backdrop, five considerations should guide the structure of GPU financing.
Assess the economic life of GPUs
The first question is how long GPUs will remain economically useful for the workloads expected to generate the revenue needed for repayment, not simply how long they will remain technically functional. As technology evolves and computational demands increase, hardware limitations may make GPUs less suitable for those workloads. Accordingly, financiers should distinguish technical functionality from commercial competitiveness rather than assume both decline at the same rate. The financier’s objective is not to predict the next generation of technology, but to avoid repayment structures that depend on optimistic assumptions about the current generation’s future value.
When determining financing terms, financiers should assess how newer hardware, lower computing-capacity prices and changing customer requirements could affect revenue and repayment capacity over the financing term. These assessments should then inform advance rates, amortization, tenor and any balloon payment. In addition to financial modeling, financiers should consider periodic appraisals and loan-to-value covenants specifying the valuation methodology, testing frequency and consequences of a shortfall, including any obligation to provide additional collateral or make a partial prepayment. The financing should be structured so that the outstanding balance declines at a pace consistent with a conservative assessment of the GPUs’ economic life and recoverable value.
Plan for repossession and recovery
While a financier may have contractual rights to recover GPUs, it is important to consider the practical difficulties in doing so. As part of their due diligence, financiers should work through what recovery would require and what value would remain after the associated costs and delays. Financiers should also prepare a credible recovery model that accounts for downtime, unpaid facility charges, professional fees, deinstallation and transport costs, and any work required before resale. A valuation based on an orderly sale is less useful if the financier has no practical route to conduct one.
Specifically, financiers should consider who will locate, access, disconnect, package, transport, test and remarket the GPUs. Can the GPUs be sold in place, or must they be removed? Where are the GPUs located? Are they identifiable? How can they be accessed in order to be removed? Answers to these questions will help formulate a recovery plan to minimize risk.
To facilitate recovery, the transaction documentation should restrict the borrower’s ability to relocate the GPUs without the financier’s consent and require details of any replacements, including make, model and serial numbers. Without reliable records, the financed GPUs may be difficult to locate and repossess if the borrower defaults. Financiers should also determine whether arrangements with the hosting provider are needed to facilitate notice, access, continued operation or removal following a default.
Additionally, financiers should confirm the conditions under which warranties, support and other arrangements can be transferred to a replacement owner or operator before funding, rather than assume they accompany ownership of the hardware. Where applicable, any required consents and assignment documentation should be obtained in advance to facilitate a transfer following recovery. Restrictions on transferability may adversely affect recovery value.
Understand the borrower’s revenue stream
A financier should establish early whether the GPUs will support its borrower’s operations or generate revenue through the sale of computing capacity to third parties. The source of repayment and the diligence required will differ based upon the answer.
If the borrower is a capacity provider, its projected utilization and the revenue stated in its contract should be closely examined. How much revenue is committed rather than forecast? Are customers required to pay for reserved capacity, or only for what they use? Can they terminate, reduce commitments or withhold payment if deployment is delayed or performance falls short? The borrower’s reliance on and the financial strength of major customers should be assessed alongside the borrower’s own credit.
Financiers should also compare the financing repayment schedule with installation milestones, customer acceptance, revenue commencement and the contract expiration date. A long-term financing supported by shorter-term customer commitments should explicitly address the resulting renewal and repricing risk during the term of the financing.
Where customer contracts are central to repayment, diligence should include an examination of payment directions, assignments, any required consent and continuity if a borrower defaults. The aim is to identify the cash flows available for repayment and structure appropriate protections, including (if the credit requires) lockbox arrangements and financial coverage covenants, effective from closing or upon specified triggers. The documentation should define the applicable ratios, calculation methods and testing frequency.
Know exactly what is owned and financed
The transaction documents should identify whether the financed property consists of individual cards, servers, networking equipment or a wider integrated system. They should also distinguish financed assets from other equipment owned by the borrower or lessee, the hosting provider or another financier.
This approach requires a documented ownership chain and an asset register supported by serial numbers or other reliable identifiers. Delivery, installation, acceptance and payment evidence should be reconciled with the assets actually deployed. The documents should address relocation, substitution and upgrades so that the financier can track what happens after closing, not merely establish what existed on day one.
The legal analysis should reflect the transaction structure and relevant locations. For a lease financing, obtain advice on characterization and the treatment of the lessor’s ownership rights. For a secured financing, address the steps required to establish and protect security and priority. In either case, investigate competing claims, the contractual rights of suppliers, facility operators and other creditors, as well as the local rules governing landlord rights.
The practical test is straightforward: If the borrower or lessee defaults, can the financier identify the financed assets, establish its rights and distinguish those assets from third-party property?
Confirm insurance
As part of the diligence for the transaction, financiers should determine whether the insurance available for GPUs at the relevant data center adequately addresses the transaction’s risks, including the potential exposure following a total loss. Insurance is an important protection supporting repayment if the GPUs are lost or damaged, and the transaction documents should require the borrower to maintain coverage on agreed terms, including an appropriate valuation basis and policy limits.
A contractual insurance requirement, however, does not establish that the required coverage is available in the market. Concentrated GPU values can constrain insurance capacity, as insurers may be unwilling to cover the full value of GPUs housed in a single data hall or in nearby halls exposed to the same incident. A single event, such as a fire, could result in substantial aggregate claims. Before funding, financiers should confirm that the required coverage is in place and review its valuation basis, limits, deductibles, exclusions and treatment of proceeds. Any material uninsured exposure should inform the financing decision and whether additional protection, such as a funded cash reserve, is appropriate.
Looking beyond GPU value
GPU financing demands the same discipline in downside protection as in assessing the commercial opportunity. Strong structures connect GPU value, credible repayment cash flows and a workable recovery strategy.
For financiers, the right questions are therefore broader than “What are these GPUs worth?” They are: “What supports their value and repayment, and which protections will remain available if the transaction goes wrong?” Addressing these questions before funding helps structure more resilient financing.
White & Case means the international legal practice comprising White & Case LLP, a New York State registered limited liability partnership, White & Case LLP, a limited liability partnership incorporated under English law and all other affiliated partnerships, companies and entities.
This article is prepared for the general information of interested persons. It is not, and does not attempt to be, comprehensive in nature. Due to the general nature of its content, it should not be regarded as legal advice.
© 2026 White & Case LLP
