Contact us

How we finance GPU servers.

The fund we manage owns NVIDIA GPU servers and leases them to AI cloud providers, or lends against them with a first or second lien. Each deployment starts with an appraisal and the customer contract that pays for it.

Apply for financingRequest access

In brief

Structures
Full-stack ownership, first-lien loans and second-lien loans
Deployment size
$1M to $5M1
Platforms
NVIDIA HGX B300, HGX B200 and HGX H200, and 8-GPU RTX PRO 6000 Server Edition systems
Useful life
We underwrite a six-year useful life, with lower rental rates assumed at re-lease. Actual life may be shorter.
Credit decisions
Made by our Managing Partner. You work with him directly.

Own the servers, or lend against them

The fund either owns the servers or lends against them. That choice decides who keeps the residual value, who carries utilization risk, and in what order cash comes back.

Ownership

Full-stack ownership

The fund buys the servers and owns them. The operator runs them for its customer under a multi-year contract.

  1. Fund investment
  2. Profits, shared
Title
The fund owns the servers.
How cash comes back
After data center costs, the customer’s prepayment and monthly fees go to the fund until it has recovered what it paid for the servers. Profits after that are shared with the operator. This depends on the customer paying as agreed.
Credit support
Title to the servers. Where available, a customer prepayment and personal or company guarantees.
What the fund carries
Utilization, re-leasing and residual value risk, because it owns the hardware.
End of term
The servers are re-leased, sold, or bought by the operator under a salvage buyout option.
Size
Up to about $5M per deployment1

Lending

First-lien and second-lien loans

The operator buys and owns the servers. The fund lends against them, and each loan is repaid inside the customer contract.

  1. The fund
  2. Operator equity
  1. Senior lender
  2. The fund
  3. Operator equity
Title
The operator owns the servers. The fund holds a first or second lien on them.
How cash comes back
Monthly interest and principal from the operator, paid out of its customer’s payments. A second-lien loan is paid after senior debt service.
Credit support
A lien on the servers and an assignment of the customer contract. Where available, personal or company guarantees.
What the fund carries
The operator’s credit, secured by the servers. A second-lien loan is smaller, priced higher, and ranks behind the senior lender.
End of term
The loan is repaid in full and the operator keeps the servers.
Size
From about $1M per deployment1

Where the fund sits in each structure. Illustrative. Layer sizes are not to scale and vary by deal.

GPU-backed debt at large providers

Large GPU cloud providers, such as CoreWeave, have raised debt secured by GPU equipment. Those financings differ greatly in size, structure and credit quality from ours.

BorrowerFacilityDateAmount
CoreWeaveDelayed draw term loan rated A3 by Moody’s and A (low) by DBRS, the first investment-grade GPU-backed financing. One tranche is fixed at about 5.9%.March 31, 2026$8.5B2
CoreWeaveLoan at SOFR plus 5.50%, with a five-year maturity against customer contracts averaging about three years.August 10, 2026$2.6B3
NscaleGPU-backed facility at Term SOFR plus 5.00%.February 11, 2026$1.4B4

Quoted for market context. These companies are not affiliated with GPU Capital and do not endorse it.

Full-stack ownership

How money moves

The customer’s payments are applied in a set order: data center costs first, then the fund, until it has recovered what it paid for the servers, then profits shared with the operator. Each step depends on the customer paying as agreed.

  1. The customer pays

    A prepayment at signing, then monthly fees for compute.

    End customer

  2. Site costs come out

    The operator collects the cash and pays for the data center and power.

    Operator platform
    • Data center and power

  3. The fund recovers its cost

    What is left goes to the fund until it has recovered what it paid for the servers.

    The fund’s costRecovered

  4. Then profits are shared

    After that, the fund and the operator share the profit.

    Profit share
    • Fund
    • Operator

  5. At the end of the term

    The servers are re-leased or sold, or the operator buys them out.

    Servers
    • Re-leased
    • Sold
    • Operator buyoutSalvage option
In a first-lien loan
The operator collects from its customer and pays the fund monthly interest and principal. If the operator defaults, the fund has the first claim on the servers, though recovering and selling them can be slow and costly.
In a second-lien loan
The operator pays the senior lender first and the fund second. If the operator defaults, the fund’s claim on the servers ranks behind the senior lender’s.
When the contract ends
Owned servers are re-leased, sold, or bought by the operator. What they fetch depends on GPU prices at the time.

Underwriting in five steps

Each deployment goes through the same five steps, whether the fund owns the servers or lends against them.

You work directly with our Managing Partner, who makes credit decisions.

  1. Origination

    Deals come from direct relationships with operators, relationships with server makers (OEMs), and referrals from investors and industry contacts.What we checkA customer contract, power and colocation for the deployment, a server platform we know, and a financing need of $1M to $5M.

  2. Appraisal

    Each server is valued on market comparables and on the income it can earn. The appraisal comes from GPU Appraisals, which is affiliated with GPU Capital, so its valuations of equipment we own or finance are not independent (conflicts of interest).What we checkThe generation, age and exact specification of each server, recent comparable prices, the contract rate against market rates, and the value expected when the contract ends.

  3. Structuring

    Tenor is matched to the customer contract and kept inside the hardware’s useful life. We underwrite a six-year useful life, with lower rental rates assumed at re-lease. Actual life may be shorter.What we checkThat the lease or loan ends no later than the customer contract, the take-or-pay and termination terms, any prepayment and reserves, and that title or the lien can be perfected before funding.

  4. Approval

    Our Managing Partner makes the credit decision on each deployment.What we checkThe customer’s ability to pay for the full term, the operator’s record of running GPUs for customers, and power and colocation agreements that run at least as long as the deal.

  5. Monitoring

    Each deployment is tracked for its whole term. If a lessee fails, we work out whether to re-lease or sell the servers.What we checkPayments, utilization and hardware condition each month, and the end dates of the customer contract, the power agreement and the colocation agreement.

How we value a server

An example structure

A hypothetical lease, to show who owns what and who pays whom. It is not an actual investment result and shows no returns. The fund we manage may never make a similar investment.

Hypothetical example

Not an actual investment result

8 NVIDIA H100 servers leased to an AI cloud provider

Equipment
8 NVIDIA H100 servers, owned by the fund
Lessee
An AI cloud provider
Agreement
A multi-year rental agreement, with rent paid monthly
Purchase financing
The server manufacturer finances part of the purchase for the first 12 months
End of the agreement
The fund still owns the servers, and can re-lease or sell them at prices that depend on the market then
Eight GPU servers in one rackLine drawing of a rack holding eight GPU servers.8 servers
How payments flow
  1. AI cloud providerLessee
  2. Monthly rent
  3. The fundOwner of the servers
  4. Financing payments, first 12 months
  5. Server manufacturerFinances part of the purchase

NVIDIA and H100 are trademarks of NVIDIA Corporation. GPU Capital is not affiliated with or endorsed by NVIDIA.

Risk management

The main structural control is time. Each lease or loan ends no later than the customer contract that pays for it, and that contract ends inside the server’s useful life.

Illustrative. Lengths vary by deal. We underwrite a six-year useful life, with lower rental rates assumed at re-lease. Actual life may be shorter.
Diversification
We aim to spread exposure across operators, GPU generations and sites. Early on, exposure may be concentrated.
Collateral
The fund owns the servers it leases, or takes a first or second lien on servers it finances. Each server is appraised before capital is committed.
Tenor inside useful life
Each lease or loan ends no later than its customer contract. We underwrite a six-year useful life, with lower rental rates assumed at re-lease. Actual life may be shorter.
Monitoring
Payments, utilization and hardware condition are tracked monthly, with the end date of each contract, power agreement and colocation agreement.
Remarketing
If a lessee defaults, we can try to recover the servers and re-lease or sell them. That can be slow and costly, and what we recover depends on GPU prices at the time.
Insurance
Before funding, we review the insurance on the servers against physical loss and damage, and who would be paid on a claim.

What can go wrong

GPU financing has real risks. Here they are, with what the structure does about each one, and where it does nothing.

RiskWhat can go wrongHow the structure responds
Rental rate compressionMarket rates fall as new chip generations ship.The contracted rate is fixed for the term. Underwriting assumes lower rates after the contract ends.
ObsolescenceHardware loses value faster than planned.We underwrite a six-year useful life, with lower rental rates assumed at re-lease. Actual life may be shorter. Residual values are uncertain.
Customer defaultThe end customer stops paying.A prepayment is collected at signing where available. The fund holds title or a lien and can try to redeploy the servers. Recovery can be slow and costly, and may be much less than is owed.
Operator failureInstallation runs late, uptime is weak, or the operator fails.We diligence the operator before funding. In full-stack ownership the operator shares profits only after the fund’s investment is recovered. Where available, personal or company guarantees.
Power and sitePower or cooling is not ready on time, or the site is lost.Power and colocation agreements are confirmed before capital is committed, and must run at least as long as the deal.
RepossessionThe servers sit in a third-party data center, and getting them back takes time and money.We confirm the site and the jurisdiction before funding, and that the servers could be repossessed and re-leased there. It can still take months.
A broad AI downturnRental rates, customer credit and hardware values fall together, at the same time as equity prices.No structural fix. Spreading exposure across operators, generations and sites may reduce it. It does not remove it.
IlliquidityCapital is committed for the life of each deal.No structural fix. Investor liquidity is governed by a fund’s terms, not by the timing of any deal.

Valuation and conflicts

We value each server the fund owns or lends against. The appraisals come from GPU Appraisals, a practice our Managing Partner leads. Its valuations of equipment we own or finance are not independent.

Who appraises
GPU Appraisals, a practice of Value Prism Consulting led by Marek Omilian, CFA, our Managing Partner. It is affiliated with GPU Capital, not independent.
What it values
Fair market value, orderly liquidation value and going-concern value, from market comparables and the income each server can earn.
Why it matters
The person who values the collateral also makes the credit decisions. An appraisal that runs high could let the fund advance too much against a server, or carry its holdings at values it cannot realize.

Deal sourcing

Operators reach us through three channels. The ones we can help have a customer contract and a data center, and need capital for the servers.

Operators
Direct relationships with AI cloud operators, from first deployments to expansions.
Server makers
Relationships with server makers (OEMs), who see financing needs when an order is placed.
Referrals
From investors and industry contacts.

70+ cloud providers5

In discussion with us as of September 2026. Discussions may not lead to financings.

Talk to the person who makes the credit decision.

Operators get a straight answer on whether a deployment fits and how we would structure it. Investors can request access to learn more about how we work.

Apply for financingRequest access