How GPU Financing Works: The Structure of USD.AI GPU Secured Loans

Let’s assume a company has entered into a contract to provide computing resources to a customer. The customer has been secured, the data center is ready, and a GPU purchase order has been prepared.

The problem is simple. Purchasing GPUs requires millions of dollars in funding. This is why companies turn to USD.AI.

On the surface, the transaction structure appears straightforward. When a company wants to purchase GPUs, USD.AI provides the loan, and the operating company repays the principal and interest over a specified period.

However, in reality, GPUs are just part of the transaction. The core lies in assessing a much broader range of risks. This includes the borrower’s financial condition, operational risks, customer contracts supporting the GPU deployment, data center and operational contracts, expected revenues and cash flows, and legal and structural mechanisms that connect and protect these elements during the loan period.

This article examines the loan assessment, structuring, and execution methods of USD.AI, designed to support the expansion of AI capital expenditures (CapEx).

1. The Complete Process of USD.AI Loans

Loans executed through USD.AI generally follow these steps:

1. The GPU operating company, OpCo (Operating Company), applies for financing from USD.AI to purchase GPUs. The first step is to submit the necessary documents for loan assessment and confirm that the transaction meets USD.AI’s requirements. This includes an approved hardware purchase order, required equity contributions, completed colocation contracts, and qualified materials that can prove signed off-take agreements or on-demand revenue.
2. A loan-specific SPV (Special Purpose Vehicle) is established under a newly formed Parent HoldCo (Parent Holding Company). Both entities are 100% subsidiaries. The reason for having a Parent HoldCo is to isolate the SPV from the operating company in the event of bankruptcy. This ensures that even if OpCo goes bankrupt, the enforcement of the collateral rights on the SPV shares held by USD.AI is not interrupted.
3. The GPUs purchased through financing, along with off-take contracts, colocation contracts, and revenue accounts, are transferred into the SPV structure.
4. The SPV orders approved equipment from an OEM (Original Equipment Manufacturer) or official supplier.
5. USD.AI establishes a first-priority claim on the GPUs and the key contracts held by the SPV.
6. USD.AI deposits the agreed loan amount into an escrow account at Wilmington Trust.
7. The OEM manufactures the approved servers and ships them to the designated data center.
8. After installing the servers, independent verification is conducted, and actual operations begin.
9. Once all conditions for disbursement of the loan are met, funds are released from escrow to the OEM or official supplier.
10. The off-taker, or customer purchasing the computing resources, makes payments into the revenue accounts managed by the SPV as per the contract.
11. The SPV uses this revenue to repay the principal and interest to USD.AI.
12. Any remaining revenue after loan repayment can be distributed to OpCo.

This structure is designed to prevent any single entity from arbitrarily controlling the entire transaction. The borrower cannot use the GPU purchase funds for other purposes, and it is difficult to change the collateral to other assets or move the equipment arbitrarily. Diverting revenue generated from GPU operations elsewhere is also restricted.

Now, let’s take a closer look at how each element combines to create a single financial structure.

1-1. SPV

First, let’s consider the initial issue.

USD.AI must clearly understand exactly what it is lending against and what it can recover in the event of borrower default. The GPU operating company cannot simply receive $50 million from USD.AI as a general unsecured corporate loan.

If the loan funds are mixed with the operating company’s other funds and the GPUs are purchased alongside other assets of the company, USD.AI will have to compete with other creditors for those assets if the operating company goes bankrupt. In this case, it becomes challenging for the lender to clearly recover the GPUs and the related contracts necessary for GPU operations.

When a company defaults, the situation becomes even more complicated. A financially distressed company typically has multiple creditors. The data center may be owed back rent, and external contractors may not have been paid for services already rendered. If the company goes bankrupt, all of these parties become creditors.

Thus, from USD.AI’s perspective as the lender, it is essential to isolate the collateral supporting the loan from other business areas to create an overall structure that is bankruptcy-remote. Even if the operating company goes bankrupt, the loan, GPU operations, revenue from off-takers, and data center usage rights should not be affected. By separating the collateral from the operating company’s general business, it becomes easier to enforce collateral rights even in default situations.

The first step in this collateral isolation is to establish a dedicated Special Purpose Vehicle (SPV).

USD.AI holds a first-priority claim on the SPV’s assets and shares. The SPV is also restricted from incurring additional debts unrelated to the transaction or voluntarily filing for bankruptcy without independent consent.

As a result, all revenues generated from the GPUs flow through the SPV, and USD.AI has the first right to this revenue for loan repayment. Even if the borrower defaults, the lender can secure clearer control over the economic value generated by the GPUs.

1-2. Off-taker

The off-taker is the customer who has contracted to purchase the computing resources generated by the GPUs financed through USD.AI.

Depending on the deployment form, off-takers can include hyper-scalers, AI inference service providers, AI model companies, general businesses, and other cloud platforms requiring dedicated computing capacity.

The off-taker is the most critical credit element supporting the senior loan. Ultimately, they are the entity that actually pays the fees used to repay the loan.

USD.AI reviews financial health indicators such as financial statements, key investors and sponsors, cash runway, and leverage levels to assess whether the off-taker can fulfill its contractual obligations.

It also reviews the off-take contract itself to check for clauses that allow the customer to easily exit the contract, such as an 'out' clause.

According to USD.AI, the creditworthiness of the off-taker and the legal enforceability of the off-take contract account for about 70% of the entire loan assessment and approval process.

1-3. Data Center

GPUs need a physical location to be installed and operated.

GPUs purchased through USD.AI financing must be installed in a Tier 3 or higher data center. This refers to data centers with high levels of security, system redundancy, and power stability. These criteria are also requirements set by property and casualty insurers providing insurance for GPU clusters.

Modern data centers are designed to store and operate GPUs and servers worth millions of dollars. Typically, physical security systems include biometric access, mantrap-style dual entry systems, continuous monitoring, and restricted area access control.

In the event of default, USD.AI can exercise step-in rights under the contracts signed with the data center operator. This means they can take over the obligations under the colocation contract and secure control over the GPUs.

The data center operator also waives any separate lien on the hardware to ensure that no rights exist that would supersede USD.AI’s collateral rights on the GPUs.

It is also crucial to ensure that payments to the data center operator are made regularly to secure the GPU equipment itself.

In the expansion of AI infrastructure, server rack space with power supply is a very limited resource. GPUs that lose their place in the data center and become inactive will inevitably be less valuable than those that continue to operate.

1-4. OEM

Even if USD.AI lends millions of dollars for GPU purchases, the actual operating company, OpCo, does not directly receive those funds.

Instead, the funds are paid directly to OEMs (Original Equipment Manufacturers) such as Dell, Lenovo, and Supermicro to purchase the completed servers.

USD.AI does not only finance the GPUs but also provides financing for the entire server rack. This includes all components necessary for server operation, such as memory, storage, cables, and switches.

1-5. Wilmington Trust: A Structure to Hold Funds Until GPUs Are Operational

USD.AI uses Wilmington Trust, a subsidiary of M&T Bank, as an institutional escrow agent.

For large GPU purchase orders, OEMs require proof that the purchase funds are securely in place before they manufacture and deliver the actual systems.

When ordering equipment from the OEM, the operating company directly pays 20-30% of the total purchase amount as a deposit.

Once the GPUs and servers are ready for use, USD.AI pays the remaining 70-80% to the OEM from escrow.

In USD.AI’s structure, if the transaction is not executed within the specified final deadline (longstop period) of 120-150 days, the funds held in escrow and any interest accrued during that period are returned as per protocol.

1-6. Independent Verification and Monitoring

USD.AI continuously monitors the status throughout the entire lifecycle of the loan even after it has been executed.

To do this, telemetry equipment installed in the data center is utilized.

This equipment provides real-time data on the uptime and health status of the servers. This allows USD.AI to independently verify whether the servers are actually operational and functioning normally.

2. USD.AI Growing Alongside the AI Infrastructure Market

Ultimately, USD.AI is a financial structure that bundles GPUs, off-take contracts, data centers, escrow, and SPVs into a single collateral structure to channel on-chain capital into AI capital expenditures.

It is not merely about lending funds secured by GPUs; the key is to structure and independently verify the entire process from equipment purchase to installation, operation, revenue generation, and loan repayment, thereby managing credit risk related to physical AI infrastructure.

As the scale of AI models grows and inference demand expands, investments in AI infrastructure centered around GPUs and data centers are likely to increase as well. Structures like USD.AI are significant in that they clearly separate and manage the use of funds, collateral, and cash flows, thereby enhancing transparency and stability while ensuring the efficiency of fund execution. As the AI infrastructure market grows in the future, the utilization of financial structures that connect physical infrastructure with on-chain capital is also expected to expand.

Future attention will be on how quickly USD.AI can scale up its loan execution and how much actual demand for AI infrastructure financing it can absorb.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.