Demand Heats Up For Spend Now, Lease Later AI Financing
In the financial model that governs the AI data center build-out, one of the most important questions is who the tenant will be. The goal is to ensure that demand from big cloud firms and AI labs for computing capacity—and their ability to pay—will line up with supply as trillions of dollars pour into data centers and related infrastructure.
Bankers and lawyers like to point out that all of these details are worked out before most financing is arranged, cutting the risks of overbuilding while making the debt funding the build-outs attractive to investors. In practice, though, data center developers are racing to bring projects online so quickly that financing can’t get arranged entirely. That’s created a need for a financing stopgap.
“There are a lot of asks for what we call equipment financing or other types of bridge financings,” John Greenwood, Goldman Sachs’ global head of infrastructure and real asset finance, said at a video event held by The Information last week. That kind of financing helps pay for early work before contracts and leases are signed and permanent financing is in place, he said.
For the bulk of the funding needed to move ahead with data center construction, financiers want to know the tenant’s creditworthiness. Shareholders in data center companies want to know who’s paying the lease underpinning the revenue streams they are modeling. That’s why we’ve begun to see cloud firms and chip companies offering what effectively amounts to insurance that the ultimate tenant, in many cases an AI lab, will pay the lease: If the lessee can’t pay, some of the largest tech companies in the world have agreed to backstop those commitments.
But even before a lease is locked in, data center developers may still want some money to level the land, pay for the power studies, or get in line for one of the increasingly scarce gas turbines they might need to generate their own power.
One solution is for a developer to convince a bank or other lender to give them a loan before a tenant even signs a lease. Demand for that kind of capital has risen sharply. “This is probably one of the largest areas of increased inquiries from our clients,” Greenwood said.
There too, new forms of financing are emerging that tap cloud firms to cut down on lending risks. Greenwood noted that there’s a “ton of innovation happening in that space.”
For instance, a number of tenants, including AI labs and cloud firms, have begun using what he calls a cost reimbursement agreement. Under a CRA, a prospective tenant says to the developer: “‘Hey, we haven’t signed the full lease yet, but go ahead and start making some of these early purchases,’” Greenwood explained.
The developers can then use that CRA to get capital, he said, with lenders looking to it as protection for the loan. If the lease falls through, hyperscalers or the AI lab will reimburse the developer, and by extension the lenders, for any costs incurred. The lenders or hyperscalers may also be able to take control of assets purchased with those funds, such as gas turbines, and redeploy or sell them.
That financing can help advance a project enough to make it attractive for a major tenant. While data centers themselves aren’t all that difficult to construct, developers first need to complete a complex series of approvals, agreements and equipment purchases around land, power and construction materials to build efficiently and on time. Building that project pipeline requires capital.
A similar idea can apply on the power side. Applied Digital, a former crypto miner turned AI data center developer, used a CRA with Cleco Power, its prospective energy provider. The agreement differed from the one Greenwood mentioned because a developer, rather than an AI lab or an investment-grade cloud firm, agreed to reimburse the early development costs.
Cleco had billed Applied Digital $25 million under the agreement as of the end of March, filings show. The next month, Cleco signed an electric service agreement with Applied Digital. Fitch Ratings, which rated some bonds issued by Cleco, noted in late May that the contractual protections, including the cost reimbursement agreement for early development costs, helped reduce risks for bondholders.
There are other variations on the theme. Last year, Macquarie Group agreed to a loan facility with Applied Digital that the developer said it intended to use to fund the pre-lease development costs for new data center projects, according to a filing. It would use an initial $100 million it had borrowed to develop the site in the midst of what it described as advanced negotiations with an investment-grade hyperscaler client.
Still, these kinds of arrangements don’t always hold. Fermi America, a data center and power company building a complex in West Texas, offers one public example of that. The company, whose board includes former Texas Gov. Rick Perry, disclosed last September ahead of its initial public offering that it was negotiating a potential cost reimbursement agreement with a prospective anchor tenant.
Fermi explained that the agreement would provide money it could use to build out its site. If the tenant ultimately signed the lease, Fermi would pay the money back over several years in the form of rent credits. If the lease did not get finalized, Fermi would repay the money to the prospective tenant.
Last November, Fermi fleshed out the details of its negotiations with the prospective tenant, saying it had entered into a $150 million Advance in Aid of Construction Agreement, money Fermi could have used to begin building.
The next month, however, after a period of exclusivity had ended without the parties reaching agreement on a lease, Fermi said the prospective tenant had withdrawn the $150 million advance. Fermi’s stock price plunged on the news. (When I was at Business Insider, I broke the news that the prospective tenant was Amazon.)
Overall, developers and lenders are betting that speed to market means more than waiting for a confirmed lease. Cost reimbursement agreements and pre-lease bridge loans give them the ability to do that, but they also reflect a growing tension in the industry. With hyperscalers and AI labs in a land grab for data centers, they may be looking at many sites and only moving forward on the most promising ones. Bridge financing can give developers a head start, but it’s no guarantee their projects will cross the finish line.
New From Our Reporters
