Credit strategy · 02In development
AI infrastructure credit — the second strategy.
Senior secured lending to mid-market compute operators and enterprise buyers. Same underwriting, with the equipment held as security.
£1–20M
Ticket size
senior secured
≤65%
Advance at drawdown
stepping down to 55%
≤36m
Tenor
monthly amortisation
1.25x
Minimum cover
debt service, tested monthly
The financing gap
We underwrite the borrower, and hold the asset
Specialist lenders finance rental contracts in tickets of $25M+. Own-use buyers and mid-market operators have few secured lenders.
- Hyperscale GPU financing is concentrated in very large facilities
- Mid-market operators buy one to twenty servers at a time
- Equipment is financeable: it has serial numbers, a resale market and measurable income
Illustrative market positioning.
Contract-led underwriting
Business-led underwriting
Who we lend to
Three kinds of borrower
GPU cloud operators
Mid-market clouds renting capacity to AI companies, with contracted or recurring revenue.
AI platforms buying own compute
Model and API businesses replacing third-party inference with their own servers.
Enterprise and research buyers
Companies installing on-premise or colocated clusters for their own workloads.
What we finance
Current-generation, enterprise hardware
| Eligible | Condition |
|---|---|
| NVIDIA B300 / B200 HGX and DGX servers | New, from an authorised reseller |
| GB200 / GB300 systems | Case by case |
| H200 and earlier | Case by case, lower advance |
| Networking, storage and installation | Part of the same purchase only |
| Hosting | Tier III data centre in the UK or EU, with lender access rights |
Indicative loan termsINDICATIVE
- Ticket
- £1–20M per borrower
- Advance
- Up to 65% of cost, stepping down to 55%
- Tenor
- Up to 36 months
- Repayment
- Monthly, amortising
- Security
- First-ranking debenture; equipment serials listed
- Cash
- Revenue through a controlled account
- Covenant
- Debt-service cover of at least 1.25x
- Pricing
- Risk-based, fixed for the term
Staged release
Capital follows milestones, never a purchase order alone
Paid to the supplier
In stages, direct to the manufacturer or reseller.
First-ranking charge
Over the equipment and the company, registered at Companies House.
Controlled collections
Revenue services the loan first; the balance goes to the borrower.
Dynamic LTV
65% at drawdown, stepping to 55% as the hardware ages.
Underwriting
Five questions before any loan
01Borrower
→
Ownership, KYC/KYB, financials, equity paid in.
02Demand
→
Contracts, usage history and the utilisation needed to repay.
03Hardware
→
Specification, supplier, delivery date, warranty.
04Hosting
→
Power, cooling, access agreement, location.
05Exit value
→
Invoice, market and stressed-sale values.
Worked example
One 8-GPU B300 server
Illustrative, from our internal feasibility model at UK public prices, September 2026.
| Amount | |
|---|---|
| All-in cost (server, installation) | £441,000 |
| Loan at 60% | £265,000 |
| Borrower equity | £176,000 |
| Term / repayment | 36 months, monthly |
| Monthly debt service (illustrative rate) | ≈ £9,000 |
| Utilisation needed to break even | ≈ 40% |
| Debt-service cover at 50% utilisation | ≈ 1.45x |
Illustrative only. Actual performance may differ materially; terms depend on the borrower, hardware and market prices.
Illustrative. Assumes 18% annual depreciation and straight-line amortisation.
Equipment valueLoan balance
0mLoan begins below asset value
12mCollateral cover grows
24mPrincipal amortises faster
36mLoan fully repaid
Illustrative. Assumes 18% annual depreciation and straight-line amortisation.
Depreciation discipline
Principal falls faster than the hardware
GPUs hold real resale value, but they depreciate as new generations arrive. We underwrite 17–20% a year and amortise faster.
- Value = lower of invoice and verified market value
- Stressed case: quick sale after removal and resale costs
- Revalued at least every six months
Risk and controls
Each risk matched to a control
01Purchase & delivery
→
Staged draws, direct payment to the supplier, acceptance test.
02Title & security
→
Title transfer, serial registry, first-ranking charge, insurance.
03Utilisation
→
Monthly usage and revenue reports; cash sweep below 1.25x cover.
04Obsolescence
→
Short tenor, amortisation and a stepping-down advance.
05Residual value
→
Lower of invoice and market value; stressed case.
06Recovery
→
Data-centre access rights and a resale route agreed up front.
Status
Rulebook and structure being finalised; first transaction in preparation.
Funding
BridgingFi capital and co-investment, with bank facility funding subject to lender approval.
Timeline
About 8–10 weeks from term sheet to a live, installed server.
Discuss AI infrastructure credit
Investors can follow GPUL. Operators and buyers can discuss financing.