GPU Financing Equity Requirement: How Much Your Equity Buys

How Much GPU Capacity Your Equity Buys

At an 80% maximum loan-to-value, the largest GPU deployment your equity supports is your equity divided by 0.20. That means $10M of equity supports up to $50M of eligible hardware. Loan-to-value is the loan as a share of the deployment cost, and the GPU financing equity requirement is the 20% you fund yourself.

Equity is usually the scarcest capital in a deployment, so it sets the ceiling on how many GPUs you can put into service. This guide is for operators sizing a first or next deployment. It walks through the formula, what counts toward deployment cost, what sits outside the loan, and where operators get the math wrong.

Step 1: Turn Your Equity into a Maximum Deployment Cost

GPU financing through USD.AI covers up to 80% of the deployment cost, and the operator contributes at least 20% as equity. Because the equity is the input you already know, divide it by 0.20 to get the largest deployment it supports. Multiply that deployment by 0.80 to get the largest loan.

With $10M of equity, the deployment can reach $50M and the loan $40M. The same math works in reverse when you start from a quote: a $30M purchase order needs at least $6M of equity. Financing starts from about $5M, with larger facilities available, so very small deployments fall below the floor.

Running the formula before your first conversation with any capital provider turns a vague raise into a hardware budget. It also shows what each extra dollar of equity is worth. At 80% loan-to-value, every $1 of equity carries up to $5 of eligible hardware.

Calculator: equity in, deployment out

The calculator takes one input, your equity, and returns three outputs. The server count uses an 8-GPU B300 server at about $600,000, from USD.AI pricing research for Q4 2026. The NVIDIA B300 reference page covers the hardware itself.

Illustrative. Q4 2026 pricing, USD.AI pricing research. Server-only pricing; figures show principal only.

Both server-only rows assume the deployment cost equals the server cost. Treat every server count as an upper bound, because networking, racks, installation and tax all reduce it.

The middle row shows why. At server-only pricing, $50M would buy about 83 servers, but part of that $50M goes to switches, cabling, racks and PDUs. Installation and tax then draw on the same cash, so the real count lands below 83. The next two steps show where that money goes.

Step 2: Count What Goes into the Deployment Cost

In financing through USD.AI, the 80% applies to eligible hardware on the purchase order: servers plus networking (switches, optics and cabling), owned racks and PDUs. Anything off that list is a soft cost you pay from your own cash.

USD.AI pricing research puts an 8-GPU B300 server at about $600,000 in Q4 2026, or about $75,000 per GPU. That figure is server-only and pre-tax, so the purchase order carries more hardware on top of it. NVIDIA's HGX enterprise reference architecture lists what an 8-GPU deployment carries beyond the server:

• Network cards. Eight ConnectX-8 800 GbE SuperNICs sit in each node.
• Switches. Spectrum-X SN5600 switches run the compute fabric, and SN2201 switches run management.

Rack-scale systems follow the same pattern. The GB300 NVL72 reference architecture (Mar 2026) specifies separate SN5600 fabric and SN2201 management switches, AI Enterprise and Mission Control software, and implementation services. The switches and cabling belong in the deployment cost, while the software and services fall outside it.

No public filing splits deployment capex numerically between servers, networking and installation, so we don't estimate the split here. Build the purchase order line by line from your vendor quotes. Our guide to financing a GPU cluster covers how servers and networking come together in one deployment.

Step 3: Budget the Costs That Sit Outside the Loan

Your equity covers 20% of the eligible hardware, and you fund the soft costs and the debt-service reserve on top of it. These items draw on the same cash as your equity, so plan them in one budget.

• Soft costs. Installation, labor, shipping, taxes, software subscriptions and data-center build-out all sit outside the loan.
• Debt-service reserve. This three-month cash buffer covers loan payments if collections are delayed, and you fund it at closing on top of your equity.
• Colocation and power deposits. Data center providers may ask for deposits before the cluster goes live, and those come from your cash.
• Working capital. The cluster has operating costs before it earns, and you need cash to carry them until revenue starts.

In the $10M example, equity supports up to $50M of eligible hardware, and the soft costs and reserve come on top. Your total cash need is therefore higher than $10M. The reserve is due at closing, so you need it in hand alongside the equity, even though it never buys hardware. USD.AI's underwriting standards set out the loan terms behind both the equity contribution and the reserve.

Large facilities can draw the line elsewhere. CoreWeave's Q2 2026 10-Q covers the quarter ended Jun 2026 and describes its DDTL 4.0 facility, a delayed-draw term loan. It says borrowings are "primarily used to finance the acquisition and installation of computing infrastructure and related fees and expenses." That wording reflects market practice at one large facility, and financing through USD.AI follows the hardware line in Step 2.

The order of work matters too. You line up the offtake contract, the hardware purchase order, colocation and the equity contribution before you apply, as the GPU financing process steps show. Budgeting the costs outside the loan at that stage keeps the equity you raise from falling short at closing.

Step 4: Decide Where the Equity Comes From

Equity is cash you bring, and an offtake prepayment can fund part of it where the offtake contract allows, reviewed deal by deal. An offtake contract is the agreement under which a customer buys your compute. When that customer pays part of the contract upfront, the prepayment can sit alongside your cash in the equity contribution.

The rest of the equity comes from your balance sheet or your investors. Plan for operator cash to make up the balance, because the prepayment's share depends on what each offtake contract permits. A prepayment term agreed early in the offtake negotiation also lowers the cash you need to raise before closing.

Investment-grade structures show how far prepayments can stretch the capital stack. IREN's FY2026 10-K says its financing covers "a portion of the acquisition cost of GPUs and related infrastructure." Its second-quarter FY26 results (Feb 2026) say that financing, together with a Microsoft prepayment, covers 95% of GPU-related capex. In Jun 2026, IREN closed an investment-grade GPU financing, rated A by Fitch, that funds about 96% of GPU capex together with customer prepayments.

Each percentage combines debt with an investment-grade customer's prepayment, so it describes IREN's whole capital stack. The loan's advance rate is only one part of that stack. Structures like these are atypical for a mid-sized operator, whose customers rarely carry an investment-grade rating.

Common Problems with the GPU Financing Equity Requirement

The equity math breaks in five places, and each has a direct fix. Check your plan against each one before you start raising.

• Treating 20% as the whole cash need. Add soft costs, the debt-service reserve, deposits and working capital to the equity before you raise.
• Sizing from server price alone. Build the purchase order with networking, racks, PDUs and cabling, then apply the 80%.
• Planning below the floor. Financing starts from about $5M of loan, so plan a deployment of $6.25M or more. The 20-server, $12.0M example clears it.
• Expecting a prepayment to replace all equity. A prepayment funds part of it, where the offtake contract allows. Plan operator cash for the rest.
• Reading investment-grade deals as advance rates. Separate the loan from the customer's prepayment before you compare any deal with your own.

Where USD.AI Fits

Your equity sets the maximum deployment, and the purchase order sets what the loan covers. Your cash funds what sits outside the loan, and an offtake prepayment can fund part of the equity. Work through those four steps and you have the number to bring to a first conversation.

Financing through USD.AI covers up to 80% of the deployment cost at fixed or floating rates. Terms run 3–5 years with straight-line monthly amortization. Capital is held in escrow until the servers are installed, burned in and independently verified.

Every deal is scored by Compute IQ against published underwriting standards on a 100-point scale. The score sets financeability, the pricing tier, and the loan's size and structure. Every deal closes on the same structure, so the equity math in this article is the math your deal closes on.

Bring your purchase order, offtake contract and equity plan. Talk to USD.AI

FAQ

How much equity do you need for GPU financing through USD.AI?

You need at least 20% of the deployment cost as equity, and the loan covers up to 80%.

What counts toward GPU deployment cost?

Eligible hardware on the purchase order counts: servers, networking, owned racks, PDUs and cabling. Soft costs such as installation, shipping, taxes, software and data-center build-out are excluded.

How many B300 servers can $10M of equity buy?

Fewer than 83 B300 servers at about $600,000 each (Q4 2026, server-only), once networking, racks, installation and tax are added. The figure is illustrative.

Can a customer prepayment count as equity?

Part of it can, where the offtake contract allows, and each case is reviewed deal by deal.

Is the debt-service reserve part of the equity?

No. The operator funds the debt-service reserve at closing, on top of its equity.