On April 8, 2026, Forum Markets — a firm that tokenizes real-world assets — announced it is entering AI chip infrastructure financing: short-term bridge loans that fund the purchase and deployment of NVIDIA GPUs. The first committed deal starts at a minimum of $25 million with an option to go up to $50 million, lent to a U.S.-based neocloud operator, targeting annualized returns in the mid-teens. The announcement was also filed with the SEC as an 8-K exhibit.
The story here is not the dollar amount — next to multibillion-dollar data center investments, $50 million is rounding error. It is the financial engineering signal: GPUs are now a debt instrument you can package, price, and tokenize. Once compute hardware itself qualifies as cash-flow-producing collateral, a GPU debt market starts to exist.
The Structure: Bridge Loans with a Pre-Committed Takeout
The design solves one specific pain point. Between the moment a neocloud orders GPUs and the moment those machines are energized and earning revenue, there is a window with no cash flow. Forum’s bridge loans fill exactly that gap:
- Terms run 60 to 120 days, funding hardware acquisition and deployment
- Deals are identified and underwritten by a third-party AI infrastructure bridge credit originator, with a loan origination partner managing each loan from origination through repayment
- Every loan carries a pre-committed takeout from USD.AI, an institutional term lender, so permanent financing picks the loan up when the bridge ends
- Forum may tokenize some or all of the loans on Ethereum through its PALM platform
In other words, Forum sits as a structuring layer between capital and collateral — not in the traditional-bank seat.
Why the Yield Reaches the Mid-Teens
The press release supplies its own benchmark: longer-term AI chip loans typically yield mid-single digits. Bridge lending earns more because it covers the phase when capital is scarcest — hardware ordered, revenue not yet started — and borrowers pay for that. For Forum, the higher coupons raise the blended yield across its tokenized structures. Revenue comes from more than interest, too: pre-tokenization yield, origination economics, asset management fees, and secondary-market activity all factor in.
The market context is in the release as well: the data center GPU market is projected to grow from roughly $120 billion in 2025 to over $228 billion by 2030, and NVIDIA’s most recent data center revenue grew 75% year over year. The lenders are underwriting the slope of the supply-demand curve, not any single customer.
A GPU Debt Market Is Taking Shape
Forum is not alone in this trade. USD.AI, the firm providing the takeouts, has publicly stated a goal of $1 billion in loans originated by the end of Q4 2026 and talks openly about wanting to “fully disintermediate the neocloud business.” CoinDesk reported in January 2026 that an AI infrastructure firm bypassed banks entirely and secured up to $500 million in onchain financing, with verified GPU deployments converted into tokenized collateral — lenders can track how the hardware actually performs. Turning “where the machines are, how they run, and what they earn” into verifiable onchain data is the technical precondition that makes this credit possible at all.
The Risks Behind the Yield
Mid-teens returns are not free. Several assumptions deserve scrutiny:
- Residual value: chip generations turn over fast, and the collateral marking of a two-year-old accelerator can fall sharply
- Concentration: loans are backed almost entirely by one vendor’s silicon, so a supply or demand shock hits every deal at once
- Structural risk: onchain tokenization adds smart-contract and platform exposure that conventional credit does not carry
- Cash flow risk: a neocloud’s ability to repay depends on contract coverage and utilization; a cloud price war compresses unit rents and, with them, debt service
What It Means for Teams Buying Compute
Three practical effects. First, GPU rental pricing will eventually reflect the cost of capital — when a neocloud’s borrowing rate sits in the mid-teens, its quotes cannot stay below that line forever, and understanding the funding side gives you negotiating leverage. Second, a functioning debt market is fuel for supply expansion, but in a downturn, collateral fire-sales could amplify downward swings in GPU pricing; anyone signing long-term contracts should stress-test that scenario. Third, for teams with limited balance sheets that still need dedicated capacity, “hardware as collateral” financing may become a genuine alternative to renting from the hyperscalers.
Sources
- Forum to Participate in AI Chip Infrastructure Financing to Purchase NVIDIA GPUs — Forum Markets
- Forum Markets to deploy capital in AI chip financing deals — Investing.com
- AI infrastructure firm secures up to $500 million onchain loan after bypassing banks — CoinDesk
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.