Summary: On April 8, 2026, Forum Markets announced short-dated bridge loans backed by NVIDIA GPUs for neocloud operators, with a first US facility of at least $25 million expandable to $50 million and a mid-teens annualized return target. Maturities run 60 to 120 days, with USD.AI lined up for takeout financing. The article flags residual-value, concentration, tokenization, and cash-flow risks.
The real-world-asset tokenization firm Forum Markets stepped into AI chip infrastructure financing on April 8, 2026, unveiling short-dated bridge loans that cover the purchase and deployment of NVIDIA GPUs. The first committed facility goes to a neocloud operator based in the United States: a minimum of $25 million with room to expand to $50 million, and an annualized return target in the mid-teens. The announcement was likewise filed with the SEC as an exhibit to an 8-K.
The dollar amount is not the story — set against data center investments counted in the billions, $50 million barely moves the needle. The story is the financial-engineering signal: GPUs have become a debt instrument that can be packaged, priced, and tokenized. Once compute hardware itself counts as collateral that generates cash flow, a market for GPU debt starts to exist.
How the Deal Is Built: Short Bridges, Takeout Already Waiting
The design targets one specific headache. A neocloud that has just ordered its GPUs has to wait for those machines to be powered up and earning before revenue appears, and the bridge loan exists to carry the operator across that income-free stretch:
- Maturities run 60 to 120 days, with proceeds going to hardware acquisition and deployment
- Sourcing and underwriting each transaction falls to a third-party AI infrastructure bridge credit originator, and a loan origination partner shepherds every loan from closing through final repayment
- Every loan already has a takeout waiting from USD.AI, an institutional term lender, so permanent financing is in place when the bridge ends
- Through its PALM platform, Forum has the option to put some or all of the loans on Ethereum as tokens
Put plainly, Forum operates as the structuring layer standing between capital and collateral — not the seat banks traditionally occupy.
Where a Mid-Teens Yield Actually Comes From
The release carries its own comparison point: loans against AI chips with longer maturities generally yield mid-single digits. Bridge capital is different money — it arrives during the stretch when funding is hardest to find, and borrowers accept noticeably pricier terms for it. Those richer coupons flow straight into the blended yield across Forum’s tokenized structures. Interest is not the only line item: pre-tokenization yield, origination economics, asset management fees, and secondary-market activity all count.
The same document sketches the backdrop: the data center GPU market is forecast to climb from about $120 billion in 2025 to more than $228 billion by 2030, and NVIDIA’s latest quarterly data center revenue was up 75% year over year. What these lenders are really underwriting is the slope of the supply-demand curve, not any single customer.
The Market for GPU-Backed Debt Is Assembling Itself
Forum has company in this trade. USD.AI — the same firm supplying the takeouts — has said publicly it aims to originate $1 billion in loans by the end of Q4 2026, and it talks bluntly about wanting to “fully disintermediate the neocloud business.” Per CoinDesk’s reporting, an AI infrastructure firm skipped banks entirely in January 2026 and drew as much as $500 million through an onchain loan, with verified GPU deployments converted into tokenized collateral and lenders able to follow how the hardware actually performs. Making “where the machines are, how they run, and what they earn” readable as verifiable onchain data is the technical precondition this credit category rests on.
What Could Go Wrong at Mid-Teens Rates
A yield like that is never handed out for free. These are the assumptions worth pressing on:
- Residual value: accelerator generations replace each other quickly, and the collateral marking on a chip only two years old can slide a long way
- Concentration: nearly every loan leans on a single vendor’s silicon, so one supply or demand shock lands on all deals at once
- Structural risk: moving the loans onchain adds smart-contract and platform exposure that conventional credit never has to think about
- Cash flow risk: servicing the debt rests on contract coverage and utilization; the moment a cloud price war begins, falling unit rents feed straight into repayment capacity
If Your Team Buys Compute, Three Things Follow
One: rental prices for GPUs will sooner or later embed the cost of capital — with borrowing costs in the mid-teens, no neocloud can quote below that line forever, so knowing the funding side strengthens your hand in negotiations. Two: a working debt market accelerates supply growth, but when the cycle turns, forced selling of collateral can deepen the drop in GPU prices, and that scenario deserves a stress test before any long-term contract is signed. Three: for teams whose balance sheets are thin yet still need dedicated capacity, financing built on “hardware as collateral” could become a real alternative to renting from the hyperscalers.