Your stablecoins finance GPU operators: the companies, often called neoclouds, that own GPUs and rent them to AI companies. The loans are repaid from the cash flow those GPUs earn running compute and inference, and secured by the hardware itself.
Every loan has two layers of support: the cash flow from customers paying for compute, and the GPUs that generate it.
Operators repay from what AI companies pay to train and run models on their GPUs, whether under contract, on demand or through marketplaces.
Equipment sits in a dedicated SPV. We hold first-priority security over the hardware and a pledge of the SPV's equity.
Revenue flows through a controlled account, and data center access and recovery rights mean the hardware can be redeployed if needed.
AI needs more GPUs than anyone can fund out of pocket, and financing for them is only beginning to form. That gap is what makes the yield available, and the underlying asset is what makes it durable.
Three ideas sit behind why we lend against compute. Each links to the research we lean on.
Compute demand no longer grows with users alone. Pre-training, inference and post-training each follow their own curve. Agents, long contexts and video multiply the load, and cheaper models invite more usage, not less (Jevons paradox).
GPUs depend on a handful of chokepoints: EUV lithography, advanced packaging, high-bandwidth memory and grid power, where connections can take five or more years. New supply takes years and billions to bring online.
Public daily indices now track what GPUs rent for. Even the H100, a design that is several years old, still earns close to $3 per hour on neoclouds, which matters for a lender's recovery value.
Neocloud and marketplace rates. Snapshot only; prices move daily. Sources: Silicon Data and Ornn Compute Price Index.
We are working directly with operators on an initial set of GPU loans. Early capital partners get the earliest look and help shape the product.
Amortizing or bridge structures sized to each operator's cash flow.
12% target net of fees, rising to 15% for capital held over a year. Paid from borrower interest at 14–20%. Indicative and not guaranteed.
Smaller tickets diversified across operators, GPU models and locations.
Minimum investment, with $100,000 or more preferred.
After 90 days, withdrawals settle monthly from available cash. Requests that can't be filled roll to the next month.
Income is distributed monthly from interest collected, in stablecoins.
As the portfolio grows, we plan to package these loans into structured, more liquid products, with senior and junior positions against a diversified pool and onchain transferability.
As the portfolio grows, Perena will turn the pool of loans into structured credit, with senior and junior positions issued against the same portfolio. Treasuries can take the protected seat, and credit funds can take first loss for more yield.
Discuss an allocationGPU deals move on vendor timelines, not bank committee calendars. Stablecoins let us fund them at that speed, with transparency built in.
Capital settles in minutes, around the clock, so operators can lock in hardware allocations when they open.
A deep, borderless pool of dollar liquidity looking for real-world yield, not limited to one balance sheet.
Positions, payments and portfolio composition can be tracked onchain rather than in quarterly PDFs.
Tokenized positions can be transferred and built on, which is what lets us grow from single loans into more liquid products.
Borrowers put their own cash in alongside ours.
We verify utilization, realized rates and customer concentration from operating history.
Every deal is modeled against 30–50% revenue declines and commissioning delays.
We focus on US data centers, where title, liens and recovery rights hold up.
Data center acknowledgments allow inspection, cure, removal and replacement operators.
Equipment insurance is required before funding.
Operators pay 14–20% annual interest on loans secured by their GPUs, out of revenue from customers renting that compute for training and inference. After Perena's fee and fund expenses, investors target 12%, rising to 15% for capital held over a year.
We enforce on the security package: the GPUs, the SPV equity and the controlled revenue account. Facility access rights let us bring in a replacement operator to keep the hardware earning, or remove and sell it. In a structured pool, junior capital absorbs losses before senior.
Spot and on-demand compute is how much of the market buys, and those operators are underserved by traditional lenders. We underwrite their observed revenue history and stress it, rather than relying on a single contract.
This is private credit. Operators can default, GPU values decline over time, and recovering hardware takes time and money. Funding with stablecoins adds stablecoin and smart-contract risk. Early deals are individual loans with limited liquidity before maturity. You could lose some or all of your principal.
The minimum is $50,000, with $100,000 or more preferred. Each subscription is locked for 90 days. After that, withdrawals settle monthly from available cash, so they are not guaranteed on a specific date: unfilled requests roll forward, and capital stays exposed to the portfolio until it is paid out. Full terms and risk disclosures are shared with qualified investors.
Tell us about your mandate. We share deal materials with qualified investors.
Running GPUs? See GPU financing →