Data center financings in the United States were $30 billion in 2024 and are expected to reach $60 billion this year. The financing structures have some similarities with traditional project finance structures used in power and LNG deals, but there are also differences.

Two data center developers and two bankers talked about this and other subjects on a live podcast in April. The following is an edited transcript.

The panelists are Melih Ileri, senior vice president of capital markets and strategy for DC BLOX, a data center company based in the southeastern United States, Tim McGuire, head of capital markets for Rowan Digital Infrastructure, a hyperscale data center developer in Denver, Claus Hertel, a managing director with Rabobank, and Mike Johnson, [ ] with Nomura. The moderators are Christy Rivera and Christine Brozynski with Norton Rose Fulbright in New York.

Spiraling Demand

MS. RIVERA: Last year in the US, we saw $34 billion in data center financings, and that number will be higher this year. The US power market is likely to need another 150 to 250 terawatt hours of electricity by 2030. That is the equivalent of adding another New York City's worth of power just for data centers. What is driving that demand?

MR. ILERI: The internet was created to communicate information. Think about the amount of data it takes to send an e-mail versus the amount of data it takes to communicate video content.

Cloud applications have been growing, especially since the start of the COVID pandemic when people were locked in their homes trying to work over the internet. There was a massive acceleration in the amount of data being consumed at home and, all of a sudden, ChatGPT comes out and you have to build infrastructure for AI learning models. The computing power need has increased exponentially.

Back in the good old days of cloud deployments, a data center of had maybe 5 to 10 megawatts of electricity load. Today, we are talking about campuses with hundreds of megawatts of electricity load. All of the video and image data that is being collected for large language models is being put through machine learning algorithms.

MR. MCGUIRE: Projects that would have seemed like a big undertaking even as recently as a few years ago are now barely worth the time.

Rowan has focused so far on multipurpose sites. That means we have multiple use cases for our tenants, including cloud AI training and AI inference. If we locate within the cloud zones near population centers, we can unlock all of those use cases, which makes our source of demand a little bit more robust than simply relying on the AI models, which have been disrupted by models like DeepSeek.

The cloud business is a tried-and-true model for a number of hyperscalers, and we see their unfulfilled cloud services revenue continuing to pile up. There is a very deep well of demand. There is a significant premium for locations that can facilitate those cloud uses. We see AI training opportunities, too. Those are a little bit less sensitive to latency and therefore are a bit more location-agnostic, but we think that as we move from the AI training phase into more of an AI end-user phase, those applications will have a similar premium on location as the cloud services do today.

MS. BROZYNSKI: Tim McGuire, given the spiraling electricity demand, are data centers looking for electricity from any source at any cost?

MR. MCGUIRE: I would not go quite that far. We still see our customers wanting sustainable energy solutions. We are housed in a private equity fund that is an article 9 fund, meaning we focus on sustainability.

That said, with power lead times in most geographies extending out multiple years, we are looking at bridging solutions to grid interconnection. Some of those are onsite natural gas generation, with longer-term solutions potentially using either co-located or offsite solar and battery systems. We have a sister company called Primergy that specializes in such projects. There are potential synergies with renewable energy developers.

We have seen customers willing to pay more for accelerated power delivery, but their willingness to absorb higher energy rates or construction costs has a limit.

The electricity cost is the number one cost for data centers and can run into hundreds of millions of dollars a year for large sites.

MR. JOHNSON: Green loans being considered, which can benefit the borrowers from a pricing perspective. We do not see “power at any cost” as a trend. However, power as the constraining factor is a big focus for data center developers.

MR. HERTEL: For each deal, we look at the “PUE” or power usage effectiveness. The ratio is fairly high, meaning the power usage is efficient. A majority of the financings qualify as green loans.

Gating Items

MS. RIVERA: That covers how you are viewing your power sources. How are you selecting new data center sites?

MR. ILERI: I can speak for DC Blox specifically. We are hyper-focused on the southeastern United States. We have good relationships with state and local governments.

We want to make sure the communities are aligned on data centers coming to their communities. Frankly, local governments tend to love internet infrastructure. For example, we put a cable landing station in Myrtle Beach, bringing international internet traffic into South Carolina. We did that with great local engagement and coordination.

From a financing perspective, lenders pay attention to whether we are in an established data center market. The situation tends to be more favorable in a brand new market. We also try to follow where our customers want to go. It is important to find large population centers with talented workforces that can construct and operate the data centers.

MR. MCGUIRE: I can speak for Rowan. We are not a spec developer. We go to sites where we know our customers want to be. We are looking at sites nationwide. Our commercial business development teams are in frequent communication with customers about what markets, timelines and budgets work for them.

The common key drivers have been timeline and process to secure power, the cost of land, water availability and sales and use taxes.

Sales and use taxes are often a gating issue for us. Our customers spend a couple billion dollars every few years for large data centers to update their IT systems. Getting an exemption from paying sales tax on that is pretty important.

We are seeing constraints pop up in some target markets. In northern Virginia, timelines to get power are growing significantly longer, and land costs are already very high at $3,000,000 an acre.

Best Markets

MS. BROZYNSKI: Tim McGuire, where are the primary data center markets today?

MR. MCGUIRE: Greater Ashburn, Virginia, the Texas triangle, the Pacific Northwest, New York City, the tri-state area and Columbus, Ohio. All those markets offer multiple use cases to end users who place a pretty substantial premium on low latency for cloud services.

When it comes to other use cases like AI training, you can put data centers anywhere as long as you have power and water access. Latency is not as big of a deal. Thus, for example, AI training use cases are less focused on the tier-one markets. However, when we move from the training phase of AI to the inference phase, meaning people are actually using AI tools in their everyday lives or we have AI agents that are virtual co-workers, low latency performance and proximity of end users will be more of a factor, and the tier-one markets will remain the most attractive.

MR. ILERI: Siting tends to follow where the customers want to go, and that follows GDP growth not only in the United States, but also globally. There is a correlation between where the GDP is highest and the greatest data center demand, other than for AI training.

Because of the large sizes of these projects, you need a good workforce. You cannot set up a hotel to house hundreds of workers and then find talent in the middle of nowhere. The primary data center markets tend to follow where the GDP is, which is where the workforce and talent are.

Financing Structures

MS. RIVERA: Let's talk about financing. We are building a lot of data centers. The data centers need money to be built. Bankers, what is the market for data center loans?

MR. HERTEL: Current data center investment on a global basis is roughly $443 billion annually, according to Moody’s. We have teams in Europe and Asia, and I cover North America. We think $60 billion is probably the right number for actual debt going into these facilities on an annual basis. That is to pay for internal servers, computing equipment, the data center facility, the shell mechanical infrastructure and new generating capacity.

These financings have access to capital markets, commercial banks, permanent capital and the private credit markets.

Some are massive elephant deals similar to LNG. We closed a deal earlier in the year that was $3.4 billion, and that encompassed quite a few campuses. They are large transactions that come in various forms. The traditional project finance form is construction plus term debt, but there is also warehousing where you aggregate projects into one larger permanent capital vehicle.

There are holdco facilities where the developers can use funds to develop their projects and borrowing base facilities where cash flows are used to extend credit for a portfolio. There are a lot of financial institutions in the market. We see more than 50 commercial banks active in the space.

MR. JOHNSON: The data center financing market in North America has doubled over the past year and doubled the year before that. Prior to 2020, the greenfield construction market volumes were in the range of $200 million per year, and now we are talking about $30+ billion just in the US, and we are on pace to more than double that again for this year.

At Nomura, we have seen a broad appetite and range of debt investors. In December 2024, we were the coordinating lead arranger for a $600 million financing supporting construction of a data center in Virginia where the tenant was a large AI computing provider. In that transaction, we syndicated pretty broadly, and we saw interest from everybody from your traditional project finance lenders to private credit funds as well as some regional banks. There is not only a wide lender group, but also a deep volume.

MR. ILERI: You are dealing with the best customers on the planet in terms of creditworthiness. They are looking to do long-term transactions: in most cases 15 years and in some cases 20 years all the way down to 10 years, but still very long-term contracting, and all kinds of barriers to entry from the perspective of it taking a couple of years to put together one of these giant facilities.

MR. MCGUIRE: Lenders see an underlying stream of cash flows coming from hyperscalers who are very high investment grade-rated counterparties. That is pretty much bulletproof.

If you look at where the longer-term ABS or CMBS data center transactions are pricing relative to where hyperscale corporate bonds are pricing, there are probably 100 to 150 basis points of premium in the data center space.

That makes it an attractive opportunity for a lot of lenders who are not even focused specifically on data centers, but are looking at the relative value.

MR. HERTEL: Hyperscalers are maybe rated AA or AA+. If you are already lending to them on a corporate basis, here is another way to lend that carries a higher premium. It is an extra 150 basis points for taking pretty much the same risk, triple-net-lease risk with hell-or-high-water payments. It is similar to a bond.

Differences

MS. BROZYNSKI: A lot of our listeners are used to project financings of power projects. What are the key differences between financing for a data center and a power project?

MR. JOHNSON: I think the term project financing can be somewhat limiting. We see a need to structure innovative solutions and to have the right structure and appropriate level of recourse for a project throughout its lifecycle, both through development and operations.

Similar to renewables, execution certainty and speed to market are critical components. The clients have to deliver a service on time to their tenants, who are, again, some of the most creditworthy and largest institutions financially on the globe.

Similar to renewables, we are looking at cash flows generated by long-term contracts, in this case from leases on the assets.

Data centers have provided investors with a similar risk-return profile to some renewables projects in the sense that high upfront capital requirements are leveraged by a long-term stream of cash flows.

In some cases with renewables you have investment-grade power purchase agreements. In this case, we have long-term leases with tenants who use the data center.

The underlying characteristics of the assets is what drives some of the differences in the financings.

For hyperscale tenant data centers, for example, these assets provide a level of cash flow stability that is not necessarily seen in the renewables space given that those projects rely wind or sunlight to drive revenue. This cash flow stability opens up the assets to a range of financing solutions, during both construction and operations, that are not necessarily available in the renewables space. I think someone earlier mentioned the ABS market, for example.

Roughly 50% of the market is concentrated among the top 20 developers and operators of these types of assets, according to one of the analyst reports.

We are dealing with a unique supply and demand imbalance. The key difference between renewables and data centers is that data centers are solving for a supply gap in an electricity demand curve that is rising steeply for reasons that we discussed earlier.

Unlike renewables, where the supply gap can be replaced by conventional power, data centers are the only thing that can solve the gap in supply.

The power supply is the main constraint on growth in the sector. There is a limited supply of data centers in the key markets, and that drives up the residual value of the lenders’ collateral as well as the investors’ returns.

MR. HERTEL: There are nuances, but from a credit risk perspective, in terms of structural risk factors, you are looking at a lot of the same things as with a traditional project financing. You are looking at developer and contractor experience and track records and the financial stability of the developer. What is the money behind the developer? You are looking at the construction budget and timeline. You are looking at the exits through a permanent financing or refinancing.

The asset class is unique. It straddles both project finance and commercial real estate. It depends on the specific transaction, but you are looking at real-estate-specific risks as well: location, lease structures and property valuation.

You can put the financing in different books. You can put it in a real estate book, you can put in the project finance book or some people have it in the infrastructure book. It is a combination of real estate and project finance.

The loans use mini-perm structures. We are financing construction plus three or four years, and then the loans get refinanced.

There is more certainty around refinancing data centers because you have multiple potential markets in which to refinance. You have the CMBS market, you have the ABS market, you have a private placement market. You potentially have insurance companies that are looking for safe long-dated instruments that they can book.

Contrast that to a traditional solar deal with developers that have large utility-scale portfolios. You can do a private placement, but there is less certainty on the refinancing aspect.

Another difference is you do not have any tax equity in these transactions. There is a lot of complexity around tax equity in every solar and wind deal.

You are not in a back-levered position. You are in a senior lien position. That makes it much easier from a documentation perspective.

Other Terms

MS. RIVERA: Why do we see more guarantees in data center financings compared to solar deals? 

MR. HERTEL: We see construction guarantees sometimes from the developer itself and sometimes from the financial sponsor that is backing the developer. Those are staggered and, in all cases, they go away at the end of construction. The market is moving away from that as lenders get more comfortable around construction risk.

MS. RIVERA: We talked about the growing need for power for data centers. In cases where the data center has a co-located power plant, are the two assets financed together in a single financing?

MR. JOHNSON: Lenders are contemplating it. Co-location can take two forms. There can be physical co-location or coordination with PPAs to purchase renewable power for the data center. The choice could be specified in the leases with tenants.

We think we will see more physical co-location with the growing constraints on the grid. It will add complexity with respect to permitting timelines.

Once you have your permits in hand, the powered shell can be constructed quickly. A combined-cycle gas facility may take longer to build. For solar, you have to deal with the intermittency of generation, which can be can be mitigated by batteries.

Then we open up a third layer of space constraints. The industry is focused currently on putting assets in tier-one or tier-two markets that are generally very close in proximity to population centers for latency purposes. Obviously it is harder to build the solar park in the middle of the downtown region.

MR. HERTEL: Let me add two more things.

We see a lot of our renewables clients signing PPAs with data centers. There is a convergence between the hyperscalers that are financing based on lease revenue and hyperscalers that are offtakers under PPAs.

You can almost put them into one client bucket. On the one hand, you have the cash flow stream from the data center. Then you have a PPA and a renewables portfolio that is signed by the same hyperscaler. It is a strong convergence.

Equipment

MS. BROZYNSKI: What is actually being financed? Is it just the buildings or is it also the computer infrastructure inside the buildings?

MR. JOHNSON: That depends on the situation. For example, it depends on whether the developer is catering to hyperscale tenants or using a co-location retail model.

For the most part, when you are financing construction, you are not necessarily financing a material amount of equipment. We are seeing graphics processing units or GPUs financed separately. There are ongoing discussions about whether and when that will be molded into one type of facility structure.

Equipment tends to be brought in by the tenants. Through the lifecycle of the asset, we have been focusing on supporting financing of different things.

MR. HERTEL: We are financing mostly the shell and the building housing the data center, including the security access around it, the lines, the cooling systems, the interconnection and things of that nature. That is all the responsibility of the developer.

There have been instances where we have also financed some of the underlying equipment and backup equipment, such as generators. The data center has to be online 100% of the time. It needs 2X redundancy.

We have a structured leasing team that finances the backup generators for some data centers. That is another component beyond project finance and traditional financing structures. There is an equipment financing involved as well.

MR. ILERI: The majority of the financings taking place here are contracted cash flow financings. In some cases, there is a booked-but-not-billing component of contracted cash flows, but you have to deliver the capacity to receive the cash flows.

The contracted cash flows can also be in the form of purely booked-but-not-billing, and that could be a ring-fenced, completely new project where lenders are financing against future expected cash flows.

What becomes really critical is the real estate must be provided as collateral. It is also really important for the borrower to have the right kind of management team with experience in the industry.

Not everybody can find the right location and have the ability to pre-lease a piece of dirt to an exceptional customer base and be able to put up a billion dollar project in a matter of a couple of years.

It takes industry veterans. The lender is underwriting the risk profile of the team and assessing whether it can actually deliver the expected future cash flows and construct the data center with no hiccups and no delivery delays. That is what is being underwritten. We lend against contracted cash flows.

MR. MCGUIRE: The value that lenders are underwriting is long-dated streams of very high quality cash flows with very strong credit supporting the cash flows from hyperscalers.

There is still a real estate element. The leases that are the source of cash flow are long-dated, but they all have expiration dates.

When we talk about a powered shell, it is a building. It is horizontal infrastructure. It is a substation. Those are long-lived assets. They do not depreciate quickly.

Contrast that to a business model where the borrower is in the business of providing GPUs, servers, racks and the low-voltage system. That stuff depreciates much more quickly. More importantly, the product lifecycle for the technology elements is much shorter than it is for something like a generator. A generator does not become functionally obsolete in a period of three to four years.

There are different gradations of what is being leased to the tenant. We focus on the powered shells, but there are also turnkey products where we provide the mechanical, electrical and plumbing systems, but not the high tech elements.

When to Finance

MS. BROZYNSKI: When is a data center project ripe for financing?

MR. MCGUIRE: We like to bring a deal to market as soon as we have land control and when the gating issues to development and energization as well as the lease contingencies are resolved. That usually means we have our procurement timelines nailed down. We have a signed construction contract and any high-level regulatory approvals have been received.

Speed to financing is super important, especially when we have high land costs or substantial other pre-development costs. Paying those costs with equity leads to be a very high cash drag on our equity returns.

That said, it is a balancing act. If we bring a deal to market too soon when there are still moving pieces, it can lead to a lot of wasted time. In the best case, we end up having to answer a lot of hypothetical questions about hypothetical scenarios and how those moving pieces could play out.

We try to box those as best we can before we start talking to the banks.

MR. ILERI: A project is ready for financing when you have purchased the land and you are moving on to customer discussions. If I am getting close to signing a contract and I would like to figure out whether the project is financeable, then it is a good time to talk to lenders.

You want the financing to be buttoned up and aligned with the timeline to get to an agreement with the customer that is pre-leasing the project that you are going to build. After that, it is off to the races.

The more creditworthy and clean your lease is, the better the financing terms you will be able to get.

As you move through the lifecycle evolution of a data center, if you are simply building one campus entirely in a built-to-suit hyperscaler financing, you are tapping into the project finance market. We talked about the completion guarantees earlier. Those completion guarantees are dropped and now you have a data center with long-term high investment-grade tenant contracts in place and stable cash flows.

The next step is to move from project finance to a take-out financing. It can come in multiple flavors. The most common way to do a take-out financing recently has been the asset backed securitization market.

Historically, there have also been solutions where you would tap into the real estate financing bucket for stabilized cash flows. It is less common now because of how attractive the ABS market is, but that generally tends to be the lifecycle of how you finance this built-to-suit bespoke product that you are putting in the ground.

MR. JOHNSON: Sone of the financings can be bespoke as well. There is not a one size fits all.

MR. HERTEL: One more comment. The lifecycle follows the same path as renewables. There are bespoke pre-NTP solutions. The data center world is moving down the same path. These could be borrowing base facilities or something similar, a holdco or mezzanine facility that supports the build out of a portfolio or development pipeline.

Small Data Centers

MS. RIVERA: We have time for one more question.

We have been talking about big financings. Will the data centers remain huge or is there the possibility of doing smaller data centers, with just a couple megawatts of energy load and rooftop solar as an energy source?

MR. JOHNSON: Due to power constraints and space constraints, we have not seen smaller data centers being developed in metropolitan or really dense areas as much as the hyperscale volume in the past couple years.

There is a need for smaller data centers. The market will eventually find a way to plug that gap.

MR. MCGUIRE: We actually do still build 1- and 2-MW data centers with a line of sight to 5- to 10-MW opportunities inside of those data centers.

We do those in two flavors. We built a number of them in the early stages of our company, where we went to the secondary and tertiary markets where there was a significant under-supply of data center capacity.

We are continuing to build those. We started with enterprise buy-and-sell customers as well as government agencies, research institutions and schools as customers.

We have also expanded our product offerings to some hyperscalers who are building capacity at the edge to recast the internet to new uses. Those are being financed. 

I do not think a 1- or 2-MW project can stand on its own with economics to support a project financing.

What we do is try a hybrid approach where we finance cash flows from existing sites mixed in with cash flows from smaller-sized new projects. You can cross-collateralize multiple sites and tap into a combination of existing cash flows in a form of hybrid financing that allows financings of smaller sites.

In September, we closed a transaction where we did exactly that: financed multiple build-outs of single-digit-megawatt data centers across nine locations in the southeastern United States.