GMI Cloud is seeking a $635 million loan secured against customer contracts for GPU capacity, according to a report by Data Center Dynamics.

The figure is the story. Six hundred thirty-five million dollars, raised not against property or servers, but against signed commitments from customers who want GPU compute. That structure marks a shift in how AI infrastructure gets funded, and it deserves a close look at the mechanism.

How the financing works

A conventional data center loan leans on the building. Lenders take the land, shell, and power infrastructure as collateral. GMI Cloud's reported deal works differently. The collateral is contractual: agreements in which customers commit to paying for GPU capacity over time.

For the lender, this changes the risk profile. The loan's value depends on whether customers keep paying, not on real estate valuations. For the borrower, it unlocks capital without mortgaging physical assets directly.

Three elements drive the structure:

• Long-term customer commitments for GPU capacity
• Predictable revenue streams those contracts create
• A lender willing to price risk against contract quality rather than bricks

This model already exists in adjacent sectors. Energy projects, telecom networks, and satellite operators have all raised debt against offtake agreements. GPUs are the newest asset class to get the treatment.

What does this change for AI infrastructure finance?

The practical question: what does this change for cloud providers and their customers?

For providers, it opens a funding route that does not dilute equity. If GMI Cloud closes the reported $635 million, it signals that lenders now treat AI compute contracts as bankable. That could pull more debt capital into a sector that has relied heavily on equity and vendor financing.

For customers, there is a second-order effect. Providers that borrow against capacity contracts have a strong incentive to sign longer, larger deals. Expect more multi-year commitment requirements attached to GPU allocations, because those contracts are now the raw material for financing.

For lenders, the model concentrates risk in one question: will AI demand hold? A contract is only as good as the counterparty behind it.

Why GPU capacity is becoming collateral

GPUs are scarce, expensive, and in demand — three traits that make them plausible security for a loan. Scarce hardware plus committed buyers equals revenue visibility, and revenue visibility is what project finance needs.

The report does not specify the lenders involved, the interest terms, or the contract durations underpinning the deal. Those details will determine whether the structure becomes a template or stays a one-off.

The bigger picture

Data center construction has consumed enormous capital in the AI buildout, and traditional financing tools have strained to keep pace. Debt secured on customer contracts is one way the market is adapting — the $635 million request shows a provider betting that lenders will follow the revenue rather than the real estate, a shift that could reshape funding across the sector if it succeeds.

Whether this deal closes on the reported terms will be the signal to watch. If it does, expect comparable structures from other GPU cloud operators within the year.