PFI Data Centre Roundtables 2025 - Project Finance Transcript
Nic Stone: Welcome to the project finance panel. We’ll jump straight into it. Omer, could you paint the picture of where we are right now? What kind of structures are being used to finance data centres?
Omer Farooq: When you zoom out, these projects are significantly larger than anything we’ve seen on the project finance, real estate or the public-private partnership side. Our first deal was around US$10m whereas some recent projects, such as Meta Platforms’ deal, are US$27.3bn. To make these work, investors from multiple markets are required.
We are seeing structures that combine tools from real estate, project finance and leveraged finance to appeal to a broad set of investors. Private credit and investment-grade investors are entering at an earlier stage than historically seen. For context, a US$2bn renewable project would typically involve only banks, but a US$13bn–$15bn data centre project requires banks, private credit and often a rating from agencies during construction. You now need to worry about how to hedge exposure and negative carry so structures are evolving.
Leases with hyperscalers may look similar on the surface but when you dig deeper, the structures differ significantly. The tenant contract looks very similar but location, how insurance is managed and the power story become very important.
While project finance markets are used to underwriting these risks, the scale with data centres is unprecedented.
Nic Stone: Joe, do you see any issues in sourcing capital for these buildouts?
Joseph Luca: Capital availability is not necessarily an issue. In our eyes it’s coming from the private side. The challenge is identifying the right opportunities due to inherent project-specific risks.
On the surface, a lot of the risks may appear similar but factors such as the power structure, potential takeout and the management team are critical. When we look to underwrite these deals, capital isn’t necessarily the issue. Deployment depends on whether these idiosyncratic risks are acceptable to us or not.
Nic Stone: Christine, you deal with these contracts and risks. What are you seeing in terms of the paperwork and structuring of these deals?
Christine Brozynski: The perceived risk depends on your background. For those from real estate, I understand that these deals are complex. From an energy perspective, they are simpler than many energy projects, which often carry higher cashflow, counterparty and political risks. As real estate practitioners got involved, they carried over things like completion guarantees – something unique to real estate projects – into these deals. These risk mitigating elements have remained, so overall, data centre deals tend to be less risky in general than energy projects but involve different risk considerations. It’s just a different financing.
Nic Stone: Omer, from a banking perspective, how are these risks being managed? Are there emerging products specific to data centres?
Omer Farooq: The risks are different – not necessarily less risky. It just depends on the project.
Concentration risk is very important. You might have a data centre with US$10bn of capex but that’s just the shell – you also have graphics processing units and backup power. So, maybe there is US$20bn–$30bn at risk in a concentrated location and you need to get property and casualty insurance for it. The real estate world is used to full replacement value coverage but getting it at this scale is challenging.
Project finance experience with probable maximum loss helps, but concentration remains a focus.
In terms of emerging products, we are seeing active construction-to-term bank loans, combining commercial real estate and project finance expertise. Teams from both areas work together on transactions. Features from real estate financing are included and we are beginning to see private securities structures, like the Meta bond, in a sinking fund-style product but we haven’t seen long-term “permanent” financing at this scale yet. It does exist in the data centre space but not for this scale. That will be the next phase when funding exits the bank balance sheet for a more permanent home.
Nic Stone: Dhaval, you were lead analyst on the Meta/Blue Owl Capital deal. Can you walk us through that transaction?
Dhaval Shah: The deal used a JV structure, which is increasingly considered or used. Key considerations included how cashflow flows from the opco or JV to the debt issuer, and potential leakages affecting debt repayment.
The Meta lease is four years with a renewal option, but the concern was what would happen if the lease is not renewed. Guaranteed minimum future value and residual value guarantee mechanisms were included to guarantee minimum payments if leases are not renewed or terminated.
We were focused on whether the GMVs were sufficient to cover the debt and if any leakages would reduce the GMV. That could affect the debtholder’s repayment.
Another concern was construction risk because this transaction involved significant power capacity buildout by Entergy. The power availability was delinked from lease commencement so even if power is not ready, then lease payments begin on a fixed date, which is June 1 2029. That mitigates construction risk.
Nic Stone: The US$27.3bn bond is rated A+ by S&P compared with Meta’s AA– and priced around 6.50% (around 225bp over Treasuries) versus Meta’s 4.40% yield (44bp over Treasuries) on its 4.75% 2034s, giving an asset-specific spread.
Joe, how do you look at companies like Meta in terms of renewing leases in the long term – will Meta even exist in 20 years when the bond is due?
Joseph Luca: Renewal risk has two components. Hyperscalers are generally considered stable – I don’t think anybody says they’re not going to be around in 20 years – but alternative tenants – the AI tenants or non-investment-grade tenants – carry more uncertainty. That’s where the analysis comes into play of developing a house view on whether they will be around in five or 10 years and if they are creditworthy in themselves.
When it comes to renewals, most people are comfortable with hyperscale risk, but not everyone is comfortable with AI and non-investment-grade counterparties.
Market and replacement analysis is critical to understanding how long it would take to replace a tenant if they depart.
The analysis is broken into two parts. You have your hyperscalers and non-hyperscalers and that’s how we quantify the risk associated with renewals and different types of tenant.
Nic Stone: Dhaval, what are the key elements for achieving investment-grade ratings in this market?
Dhaval Shah: From a credit perspective, data centres have low operational and construction risk complexity so the rating is primarily driven by the contractual structure and the counterparty exposure.
Investment-grade ratings depend on creditworthy counterparties, strong contractual structures and adequate coverage ratios with resiliency. For large-scale projects, the key concern during construction is labour supply and the equipment procurement within the time to complete construction.
AI facilities built in a remote location then releasing risk, particularly in an emerging market, require substantially higher coverage ratios.
Omer Farooq: Releasing risk depends on the type of data centre. Investors will take a view on renewal on large cloud-focused campuses in strong availability zones.
The other item for large campuses is whether what’s inside the data centre has a life of four or five years. There are a lot of new technologies emerging in terms of cooling and chips, so the question is whether the design is going to work for the latest technology in 15 years, and will it need such an expensive retrofit that the tenant will move to a new location.
Nic Stone: Christine, looking at SPVs like in the Blue Owl deal, what are the challenges across jurisdictions?
Christine Brozynski: These are generally non-recourse to the sponsor, other than the completion guarantee during construction. There’s a lot of focus on making sure lenders are protected at all stages of the deal, irrespective of what happens with the project, so insurance is very important.
Covenants are going to be very tight and the wraps expensive but, overall, lenders can get comfortable with that. The completion guarantee from a developer certainly helps.
The issues we see tend to be from the regulatory side, particularly with respect to the power source. There are lot of rules around that and they change all the time. They also differ by state, so there’s multi-jurisdictional risk with how you’re connecting your power.
You have to make sure that part is airtight because the rules were not designed to accommodate data centres. Deals must account for these legal complexities to ensure compliance.
Nic Stone: Omer, how important is power in structuring these deals?
Omer Farooq: The power story, if not the first, is one of the first three things we consider. Sites may be grid connected but does the grid need an upgrade? It may be behind-the-meter gas or behind-the-meter renewables, but what is required to build it, how much will it cost and will it be delivered on time? Leases typically keep power risk with the tenant but lenders focus heavily on ensuring availability to avoid operational delays. That’s what we focus on.
Locations are selected for excess power and favourable regulatory environments.
Christine Brozynski: Power markets are challenging; there’s not enough power coming onto the grid in the next couple of years. The US administration is favouring some forms of power over others that will take much longer to build and there is going to be a gap.
I’ll illustrate with a project we did in ERCOT [Electric Reliability Council of Texas]. The project was co-located with a wind farm taking the power through a four-party PPA [power purchase agreement] to determine when power went to the grid versus the data centre. The challenge was that the data centre was being constructed in phases, which meant that there were multiple project companies developing and financing each phase and one entity signing the PPA.
There are rules in ERCOT forbidding parties to resell power if they’re not a utility. So, the entity couldn’t resell the power to the project companies owning the different phases. We had to have the entity pass it through in a sub-lease, a rule that was probably designed for an apartment building where you’re passing power costs to your tenants. We made it work for data centres.
There is a growing acknowledgement that we need a new regime that will work but, until then, you’ve just got to be creative sometimes.
Nic Stone: Dhaval, when you look at data centre deals, are you looking at the power element?
Dhaval Shah: Reliance on power availability by the time the lease starts is one of the considerations that plays into the ratings of a data centre. It is an element that has delayed some projects so it is a focus for us. If the lease start date depends on power construction, then that is a risky feature for us and may not sit well in achieving an investment-grade rating, depending on risk of its completion by the lease start date.
Nic Stone: Omer, resource constraints – equipment, turbines, land, labour. Are they slowing down dealmaking?
Omer Farooq: There are supply chain and labour concerns, but firms are being creative. Hyperscalers are placing orders two years out for procurement of long lead items, for instance, and rather than building a combined-cycle gas plant next to the site, they’re building a simple-cycle plant, they’re building wind-plus-solar. Getting power online is so important. It determines which data centre gets built first.
Nic Stone: Beyond the physical aspects, what about financial products? Insurance was mentioned earlier, along with hedging, which has faced challenges recently in data centre deals, and capital itself. Do you see any shortages in any of those markets and how is that being addressed?
Omer Farooq: The Meta deal demonstrates how to access deeper pockets of capital. Banks are very active and that market is deep. The next phase is how to get this exposure into publicly rated securities, where it can be tranched and broken into smaller pieces. That will be welcome.
Currently, there is no constraint on capital. Large deals have been successfully placed across banks and demand for this paper remains strong. Once you understand the deal, it can be replicated faster than other project finance deals, with only a few key differences to focus on.
Joseph Luca: Bringing capital to projects from the bank or private credit side of the market is currently is not an issue, but we also look three to five years into the future to anticipate potential capital constraints.
A lot of these construction facilities are structured as three years, plus one, plus one, with extension rights at the borrower’s discretion. The three years are to bridge the construction period. The extensions provide flexibility for the sponsor to refinance once the asset reaches stabilisation, allowing better pricing and leverage relative to the construction facility.
Once assets reach commercial operation date or stabilisation, we expect them to be taken out in a different market, whether ABS, CMBS or in a long-term structure. The question is whether markets can absorb the approximately US$50bn of paper expected from upcoming projects. Right now, capital is available, but we are strategically considering future takeout markets.
Christine Brozynski: Refinancing risk is not a major concern in the documentation. Banks are comfortable that there will be a takeout. In renewables, refinancing risk is higher due to political uncertainties but in data centres, construction financing is supported by strong takeout expectations.
Nic Stone: Regarding potential bubbles, there are stories of data centre loans guaranteed by AI companies, chips purchased in exchange for data centres being built, etc. Dhaval, how are you evaluating these risks?
Dhaval Shah: It is too early to determine whether there is a bubble. For AI giga campuses, key considerations are the lease term and protections for the issuer, similar to Meta’s GMV. Lenders are not typically taking renewal risk on AI campuses; they want to ensure the lease term or protections should the tenant leave cover debt repayment. Beyond AI, a key risk could be sovereign data policies and regulations that could restrict data storage outside of a sovereign.
Nic Stone: Omer, is the potential for a bubble leading to any reticence in your approach to data centres?
Omer Farooq: As an institution, we are strategic and tactical. We look at the external risks: location, construction quality, management and offtaker strength. And then we look at our internal exposure: concentration risk by state and hyperscaler.
As a lender, we feel protected. These deals are well structured, and product demand remains strong. We’re not taking technology obsolescence risk in these deals.
Nic Stone: Are we going to see portfolio deals or will financing remain project specific?
Christine Brozynski: These gigantic deals consume enough capital that combining them into megadeals is unlikely. Nonetheless, we are starting to see early-stage financings/development capital/land acquisition financings applied to multiple sites and smaller data centres. Early-stage financings tend to be revolvers and projects will come in and out of the facilities, for instance. But I don’t see the large single asset deals being combined into portfolio financing any time soon.
Omer Farooq: Even within one large deal, power and data centre financings are often separate to access different investors. A US$100bn combined deal is unlikely.
Nic Stone: We’ll wrap up our project final panel here and I’ll now call upon my colleague Richard Leong who will host the structured finance panel. Thank you to the panel.