Who lends against GPUs, and which kind of lender fits your cluster
Quick answer, Nov 1, 2026As of 1 November 2026, GPU clusters are financed by banks, private credit funds, equipment lessors, small-cluster programs, bridge lenders and onchain credit. Order size and whether a customer has signed decide which one will look at you. Under about $20 million, start with the small-cluster programs, the equipment lessors, and banks that run a smaller equipment book.
The lenders, by kind
Each kind of lender starts at a different deal size. The names below are the lenders on the Frontier Desk map, each linked to its own page, with public deals where the borrower has announced one.
| Kind of lender | Who does it | What it looks like in public |
|---|---|---|
| Bank | JPMorgan Chase, Goldman Sachs, Morgan Stanley, MUFG, Macquarie Group, Wells Fargo, Citizens Bank, Crédit Agricole, Silicon Valley Bank, a division of First Citizens Bank | Macquarie Group led a GPU-backed facility of up to $500 million for Lambda, with Industrial Development Funding alongside it, announced 9 April 2024 (Latham & Watkins) |
| Private credit fund | Blue Owl, Magnetar, Blackstone, BlackRock, Carlyle, Coatue, Apollo, Brookfield, Upper90 | Funds managed by Blue Owl led a $2.4 billion compute equipment financing for IREN on 28 August 2026 (Blue Owl). Magnetar and Blackstone led CoreWeave's $2.3 billion facility, with Coatue, DigitalBridge, BlackRock, PIMCO and Carlyle alongside them (CoreWeave, 3 August 2023) |
| Equipment lessor | Post Road Equipment Finance, Wingspire Equipment Finance, Dell Financial Services, Data Sales, Sentry Financial, Tech Finance Co. | A lessor buys the servers and rents them to you, so it asks for less cash up front than a loan (/compute) |
| Small-cluster program | American Compute, Hydra Capital (both arrange; their lenders decide) | Some arrangers run programs with their lending partners for orders under about $20 million: fixed criteria, standard documents, a term sheet within days (/compute) |
| Bridge lender | Forum Markets | Pays for the hardware order for 90 to 120 days, until the main lender takes over once the cluster is installed (/compute) |
| Onchain credit | USD.AI | Asset-backed loans over three to five years, up to 80 percent of hardware value against at least 20 percent of your own (USD.AI) |
What decides which one will look at you
Two things, in this order.
Order size. Above a few hundred million dollars, the digital-infrastructure lenders and asset-backed notes open up to operators with a record (/compute). A first cluster under about $20 million belongs with a small-cluster program, an equipment lessor, or a bank that runs a smaller equipment book. A widely cited guide published by American Compute, which runs one of the small-cluster programs above, puts the largest private credit funds out of reach for a first project and sends new operators to smaller firms instead (GPU Financing 101, updated June 2026).
A signed customer. Every kind of lender above reads the customer contract first. On-demand and marketplace rentals do not count, because no lender can underwrite rental rates that move with the market. A multi-year take-or-pay commitment from a customer with a balance sheet is something it can. That same guide calls contracted revenue "the single most important thing you can bring to a financing partner" (GPU Financing 101).
What they quote once they are interested
The ranges that guide publishes: an advance of 50 to 80 percent of the equipment cost, a debt-service coverage covenant of 1.15 to 1.25 times the loan payments, and a rate of 7 to 10 percent a year behind an investment-grade customer, 10 to 12 percent behind a strong corporate one, and 12 to 15 percent with a weak contract or none (GPU Financing 101). Every lender sets its own terms. The order in which they read a project is covered in how lenders underwrite a GPU cluster loan.
Where to start
Run several lenders of the right kind in a short burst rather than one at a time, and bring the same package to each: the signed contract, your cash position, a purchase quote with the full bill of materials, and a colocation quote from a Tier III data center.
If your cash is short, an operating lease asks for less up front than a loan. If the senior lender will only fund once the cluster is installed and earning, a bridge covers the order until then.
What Frontier Desk does
The GPU financing tool on /compute reads a project the way a lender would, names the kinds of financing that fit it, and lists the lenders that do each kind with the deals behind them, each profile linked to the lender's own page. Answer the questions and it names who lends on projects like yours.
Questions
Will a lender finance a cluster with no customer contract? Rarely on a term loan. Without one, the American Compute guide says you fund 50 to 60 percent of the equipment cost yourself and show about 150 percent of the equipment value in cash and funding, and it puts the rate with a weak contract or none at 12 to 15 percent (GPU Financing 101).
My customer is another GPU cloud. Does that change who will lend? It narrows the list. A customer whose own revenue comes from the same market your cluster earns in counts for less, and the guide tells operators to line up a well-funded customer that is not another GPU cloud (GPU Financing 101).
Do I have to pick one kind? No. A project can carry a senior loan or lease for most of the hardware and a bridge for the order until the cluster is installed and the customer is paying (/compute).