Analysis · New markets · 4 min read
On-chain GPU financing in plain English
Alongside the bank-led loans, a small on-chain market lends against tokenised receipts for GPUs. The headline numbers are large. The commitments are not.
- 01USD.AI has announced frameworks of up to US$500m each with QumulusAI and Sharon AI, both uncommitted.
- 02QumulusAI's framework advances up to 70% against tokenised GPU receipts.
- 03Sharon AI later signed a US$356m conventional GPU loan led by Goldman Sachs.
USD.AI is a protocol that lends stablecoin liquidity against GPUs. The borrower's hardware is represented by a token, a digital receipt recorded on a blockchain, and the protocol advances funds against it. Investors in the protocol earn the interest.
Three public deals show the model. In October 2025 QumulusAI announced a framework of up to US$500m, advancing up to 70% against tokenised GPU receipts. Sharon AI announced a framework of the same headline size in January 2026. NexGen Cloud's US$34m facility in June 2026 financed B200 GPUs in Sweden, secured on the hardware and contracted cash flows.
Public USD.AI GPU financings
| Date | Borrower | Size | Collateral |
|---|---|---|---|
| Oct 2025 | QumulusAI | Up to US$500m, uncommitted | Tokenised GPU receipts; up to 70% advance |
| Jan 2026 | Sharon AI | Up to US$500m, uncommitted | Tokenised GPU collateral; Australia and Asia-Pacific |
| Jun 2026 | NexGen Cloud | US$34m | B200 GPUs in Sweden and contracted cash flows |
Read the word "uncommitted"
The two US$500m frameworks are uncommitted. The lender is not obliged to fund any particular amount, so the headline is a ceiling, not a facility. That is why the GPU Financing Handbook leaves them out of its volume totals.
The model has a place. It can fund smaller fleets quickly, and an advance rate of 70% is conservative next to the 90% to 95% seen in contract-backed bank deals. But borrowers who can access bank markets appear to move there: in October 2026 Sharon AI signed a US$356m SPV loan led by Goldman Sachs at 9.95% fixed.
Questions a traditional lender would ask
- What is the legal relationship between the token and the physical GPUs, and under which law?
- How is security over the hardware perfected, and who holds it on enforcement?
- Who has access to the data centre if the borrower fails?
- What happens to the token if the GPUs are moved, damaged or replaced?
For in-house counsel
Take it to the documents
The filed wording behind this piece, for the lender’s or the borrower’s side. Previews are open; the full text is free with InfraCompute Weekly.
- GPU loan summariesFiled GPU credit agreements side by side: advance rates, covenants and the collateral package.
- Financing clause libraryCash traps, coverage ratios, cures, step-in and GPU security, quoted from credit agreements and indentures.
- Issues list: GPU credit agreementBuild it from the lender’s or the borrower’s side, with where filed deals land.
- The legal desk for in-house counselContract library, issues lists, checklists and English law notes.
Sources
- QumulusAI: newswire.com
- Sharon AI (USD.AI): pulse2.com
- NexGen Cloud: nexgencloud.com
- Sharon AI SPV (Goldman Sachs): dealroom.co
