Why neoclouds need their own approach to financing
A neocloud is a cloud provider built around GPU capacity for AI training and inference. Most own or lease their GPUs, rent space and power from data center operators, and sell capacity to AI companies, enterprises and other clouds. The business is capital-intensive in a way few young companies have been before: a single cluster can cost more than the company has raised in equity, and the hardware must be bought before the revenue that pays for it arrives.
That makes the question of how a neocloud finances its GPUs central to how fast it can grow. Equity alone is too expensive and too slow. Vendor and bridge financing help at the start but rarely scale. What scales is debt that lenders are comfortable holding in size, and lenders are comfortable when they can see exactly which cash flows will repay them.
Contracted and uncontracted revenue are financed differently
Most neoclouds earn two kinds of revenue from the same hardware. Reserved capacity is sold under multi-year contracts with fixed monthly payments. On-demand capacity is sold by the hour or the month to whoever needs it. Both can be profitable, but a lender reads them very differently.
Reserved capacity under a non-cancellable, take-or-pay contract with a creditworthy customer can support senior debt sized to the contract’s payments. On-demand revenue depends on utilization and pricing that can change quickly, so senior lenders usually give it little or no credit when sizing the loan. The practical consequence is that the share of a cluster sold under qualifying contracts sets how much senior debt the whole cluster can carry.
The remaining gap is funded by the neocloud’s own equity or by separately priced junior capital that is paid for taking utilization and residual value risk. Separating the two is what allows each to be priced for what it is, rather than pricing the whole cluster at the level of its riskiest revenue.
How much of a cluster the contracts can carry
The example below shows a neocloud cluster where about two thirds of capacity is sold under three-year reserved contracts and the rest is sold on demand. The senior debt is sized on the reserved contracts alone.
An illustrative example
| Cluster cost, GPUs, networking and installation | $120 million |
| Reserved contracts over 36 months | $130 million |
| Net cash from those contracts after allocated costs of 30% | $2.5 million a month |
| Debt service supported at 1.25x coverage | $2.0 million a month |
| Senior debt repaid in full over 36 months at an assumed 9% rate | $64 million |
| Remaining cost, funded by equity or junior capital | $56 million |
| On-demand revenue from the remaining capacity | Upside to equity |
Illustrative only, with figures rounded. Not an offer or an indication of terms for any transaction.
Selling a larger share of the cluster under qualifying contracts before the hardware is ordered is the most direct way for a neocloud to reduce the equity each cluster requires. So is negotiating the contract terms lenders need at the outset, as described in the guide to GPU financing for contracted AI infrastructure.
How neocloud financing grows with the company
Neoclouds rarely use one form of financing for long. The first clusters are often funded with equity, vendor terms or short bridge loans. Once the company signs its first substantial reserved contracts, each deployment can be financed on its own, in a project company that holds the GPUs and the contract.
As the number of deployments grows, financing them one at a time becomes slow and expensive. The next step is a program: a facility that finances successive deployments on agreed terms as contracts are signed, or a pool that combines several deployments and customers so that no single contract dominates. Pools and programs spread customer concentration, lower the cost of each new financing and give institutional investors a larger, repeatable position to buy.
What lenders look for in a neocloud
- Contracted share. The proportion of capacity sold under non-cancellable, take-or-pay contracts, and their remaining terms.
- Customer quality and concentration. Who the largest customers are, how much of revenue each represents and what credit support stands behind them.
- Operating record. Uptime, utilization, delivery against past contracts and the team’s experience running GPU capacity at scale.
- Site, power and colocation. Contracted power and space for at least the term of the debt, and the landlord’s willingness to recognize the lenders.
- Hardware and vendor terms. Widely deployed NVIDIA or AMD systems, delivery schedules, warranties and support.
- Reporting. The ability to report customer payments, utilization and uptime to lenders monthly.
Where Park Street Global fits
Park Street Global structures financing for neoclouds and arranges it with institutional credit investors in the United States and Europe. We work on the credit itself: separating contracted from uncontracted revenue, sizing the senior debt to what the contracts can repay, and designing the structure so that the senior investors hold clean, contract-backed paper while other risks are placed with capital that is paid to hold them.
For a neocloud with its first large contract, that means financing a single deployment. For one signing contracts regularly, it means building a program that finances each new deployment on terms agreed in advance. We do not lend our own money, so the size of a financing follows the transaction rather than our balance sheet.
Common questions
What is a neocloud?
A cloud provider built specifically around GPU capacity for AI, as distinct from the large general-purpose cloud platforms. Neoclouds typically own or lease the GPUs, rent data center space and power, and sell capacity under reserved contracts and on demand.
Can on-demand revenue be financed?
Senior lenders usually give on-demand revenue little or no weight when sizing debt, because utilization and pricing can change quickly. It can support junior capital priced for that risk, and it improves the equity return once the senior debt is sized on the contracted revenue.
Do lenders prefer one hardware vendor?
Lenders focus less on the vendor than on how widely a system is deployed, the strength of its warranties and support, and the depth of the market for it. Current NVIDIA and AMD systems both qualify.
Can several sites or customers be financed in one facility?
Yes. A program or pool can finance deployments across several sites and customers on common terms, which spreads concentration and lowers the cost of each additional financing.
What reporting will lenders expect?
Typically monthly reporting of customer payments, utilization and uptime, with notice of any customer default, service credit or change at the site.