Powering Progress Episode 4 – The Power Supply Challenge: Delivering Energy for AI Growth

Aubrey Bishai [00:00:09]

Welcome back to Powering Progress, a podcast series exploring the infrastructure, energy, and market dynamics shaping the next generation of data centers and AI development. I’m your host, Aubrey Bishai, Chief Innovation Officer at Vinson & Elkins.

Artificial intelligence is driving an extraordinary surge in data center demand, but behind every server and processor is a far more fundamental question: where will the power come from, and how will it be delivered reliably and at scale?

In this episode, we’re joined by Mark Brasher, who advises clients on power development, power purchase agreements, EPC contracts, and the complex interconnection challenges that increasingly define large scale digital infrastructure.

Together, we’ll explore the questions developers, investors, and operators are asking as they plan the power strategies behind the AI economy.

Mark, welcome and thanks for making time.

Mark Brasher [00:01:28]

Thanks, Aubrey. Great to be here.

Aubrey Bishai [00:01:30]

So how secure is power availability relative to data center growth timelines?

Mark Brasher [00:01:38]

Well, I think the short answer is that supply is not keeping up, and there’s a real structural gap or deficiency in the market in terms of power availability. We saw a report this week that gas plant construction costs have increased by almost 70% over the last two years. And I think that reflects not sort of opportunism on the part of the main suppliers and builders of these projects, but a real deficit in capacity. And that’s across the board. There are only a certain number of EPC contractors who can do this work well and support it with robust balance sheets. There are labor shortages. The supply chain is very stressed. The original equipment manufacturers of equipment are sort of a little bit reluctant to ramp up their manufacturing capacity to meet demand, because I think they still have some skepticism about sort of long term where demand’s going to settle. And then there are raw material shortages as well.

So all of that means that there’s a real insecurity in terms of the power that’s available to meet the data center growth. And we’re starting to see reporting on the AI companies themselves saying they’re unable to access enough power to provide all of the services that they want to provide. So this is structural, and I think it’s going to last for quite a few years to come.

And what that’s resulting in at the moment is a kind of all-of-the-above strategy in terms of how to get power to data center projects. So we’re seeing a lot of the really well capitalized investors and developers are pursuing the best grid connected opportunities and powered land opportunities. Those opportunities are becoming more and more cost prohibitive to pursue. So they’re starting to concentrate in a smaller number of hands. That’s pushing many others into pursuing behind the meter projects where there’s a slightly lower barrier to entry, although even their tenants and lenders are becoming a lot more selective about who they deal with on the power generation side.

And then also, we’re seeing a lot of hybrid opportunities being pursued as well, where developers are looking to supplement grid capacity with behind the meter generation, or use behind the meter as a bridge to fill the gap between when they need to start data center projects and when grid power is available. So there’s a lot of kind of making this up on the fly, and the name of the game is power capacity, however you can get it.

Aubrey Bishai [00:04:49]

With some of these challenges around power, how flexible are PPA structures as power demand scales?

Mark Brasher [00:04:58]

Depends what you mean by flexible. I think we’re seeing a lot of variability or variation in the market. There are a lot of different structures showing up, and one of the main things driving that is the regulatory constraints that override every project. So focusing on behind the meter, where I think this is most interesting, the developers are desperate to avoid being classified as a regulated utility. That brings rate regulation, service obligations, oversight, and various other constraints that kill the economics of behind the meter projects. So you have to work on a project-by-project basis, looking at both federal and state regulatory constraints that are really going to dictate the structure.

What that means is we are seeing conventional PPA structures, often using a retail sleeve to overcome the utility regulation issue. But we’re also seeing tolling structures, so with a gas plant arrangement where the end user sources the gas and then tolls that through the gas plant and offtakes the power. I mean, that solves a regulatory issue in a lot of places. It also solves a financing issue, which is the gas supply itself is something that’s a big risk. It’s a big infrastructure project in itself to get gas to these sites on a reliable basis.

And shifting that to, say, on the hyperscale projects, the hyperscalers themselves, means the gas companies are facing a really creditworthy counterparty. The developer of the power plant has de-risked their exposure somewhat by laying off the risk of sourcing the gas and having gas available over the term of the contract. So all of that kind of helps to make some of these projects more bankable. That’s an example of one of the key trends that you see across these projects — that everyone is trying to find a way to access the credit of the handful of big name data center companies who are really underwriting this entire build out.

Aubrey Bishai [00:07:27]

How about on the construction side? Where are the principal construction and EPC risks allocated?

Mark Brasher [00:07:34]

This is kind of interesting, because intuitively people think that lump sum turnkey EPC contracting is the way to go and the way to de-risk major capital projects like this. But when you’re thinking about behind the meter microgrids, and particularly in light of the constraints in the EPC contractor market, that’s not necessarily the case. So first of all, there are only a very small number of EPC contractors who have the balance sheet and the reputations to really underwrite all the risk on these projects. And they have so many opportunities, they really don’t have to agree to onerous risk allocation.

So what that’s resulting in is not just in power, but across the board for major capital projects in the U.S., it’s becoming very difficult to negotiate lump sum turnkey contracts and to get conventional bankable terms and conditions in those agreements. In the microgrid context, that’s not necessarily fatal — you look at the market and the secondary, second-tier EPC contractors, you pretty quickly come to the conclusion that you’re better off facing the OEMs themselves. So the Siemens, GE, Vestas, those types of OEMs have a much better credit profile, and so negotiating the key agreements for the generating equipment with them and using their performance guarantees and credit to support the projects is a good solution.

A big challenge there, though, is they know that it’s a seller’s market. They know how strong their leverage is. The terms you can get, particularly on, let’s say, gas turbines, compared to what you could get three or four years ago are starkly different. And so what that means is developers and lenders having to bear more risk than they would have a few years ago, and then also having to bear more risk just via the nature of that multi-prime contracting structure itself. So we have multiple OEMs supplying equipment, potentially multiple contractors working on a site. There’s a lot of interface risk that needs to be managed, and a lot of the work we’re doing on the construction side of things these days is trying to manage those risks and use best practices in the contracts that we’re negotiating to make the gaps — which are inevitably going to exist and fall back onto the owner — as narrow as possible.

Aubrey Bishai [00:10:04]

What interconnection constraints or delays could impact delivery?

Mark Brasher [00:10:11]

So I mean, there’s two things here. Obviously, on the grid side, there’s a lot of uncertainty at the moment, as in various jurisdictions rules are evolving — the requirements for large load studies and what the transmission companies themselves are requiring start to evolve. So the picture remains somewhat unclear there, and again is why people are looking at behind the meter as a bridge.

The other one that’s interesting, given how prolific the proposed use of gas turbines is, is the gas interconnection itself, which I mentioned before is an important risk item, and there structures are obviously preferred because you can, as a power developer, shift that risk away from yourself and onto the offtaker. And so what that means, both in a power context and in data center leases — you know, we’re seeing the same thing with the gas or the power. If you can shift some of the risk of supplying the gas or just the power supply onto the end user, then you can try to negotiate terms that provide for the commencement of revenue when you’ve done everything that you are required to, even if the gas is not there or the power is not there.

Now, how far you can shift that risk is going to depend. And oftentimes, you know, even if you can shift the gas supply and sourcing risk onto your offtaker, you may still remain responsible for constructing the gas interconnect and maintaining the transportation agreements that are required to get the gas to the plant, so that typically a developer still has to manage some of that risk.

Aubrey Bishai [00:11:56]

This is really interesting, Mark. So with all this in mind, what assumptions around load growth could create future exposure?

Mark Brasher [00:12:06]

I think clearly the biggest risk is regulatory. I think at this point, everyone tends to agree that the technology’s real. All of us have used AI and seen how powerful it is. And I think, to me at least, the demand picture is very clear — much clearer than it was a year ago, when I and others were probably asking themselves, where’s the market for this? Now I think it’s clear.

That creates its own problems or risks on the regulatory side, I think. We’ve focused today on the power constraints. And I think what that means, if some of this power — or a large amount of it — is going to be sourced from the grid, that’s going to really start to stress retail electricity prices. I was reading recently that retail electricity prices are expected to increase by another 40% by 2030. That’s going to have political implications — one of the reasons why this structural deficit in power, I think, is going to continue.

Then there’s also, in terms of risk for developers who are investing in that very favorable structural environment, I think that the two big concerns that could create risks for them are, obviously, the societal concerns about AI and impacts on jobs, concentration of power, and data privacy — all those kind of things. So if we had a less permissive regulatory environment down the track and people were trying to constrain AI growth, that may hit the demand side of this power picture. And then obviously environmental — the emissions issues are real, the water usage is real — and all of that could create conditions where we end up in a tighter regulatory environment.

So I think the current investment activity kind of assumes that AI is going to continue to be developed in a relatively permissive regulatory environment. If that changes, that may create an issue for projects and the investment theses that people have relied on to start developing those projects.

I think the last thing I would say is, you know, maybe a moderating factor for some of those risks is, even if you did have a tighter regulation, the market’s so far out of balance at the moment in terms of the power supply being able to keep up. If there were regulation that hit the demand side in terms of AI growth, I think all that would probably do is bring the market into equilibrium.

Aubrey Bishai [00:14:50]

How are reliability and redundancy being addressed contractually?

Mark Brasher [00:14:55]

Yeah, so there’s a really interesting tension here, and kind of a meeting of two different worlds that’s happening. So on the data center customer side, they have near zero tolerance for downtime of the baseload power solution that they’re contracting for. And then on the power generation side, people are used to developing power plants where you do have downtime, you have scheduled maintenance every year, you have all sorts of excused outages.

And so in the contracts, both in data center leases and in the power offtake — and I should mention, you know, there’s a bit of a trend at the moment towards the data center customers wanting to bundle the power inside the lease, even if it’s being generated on site. That’s a relatively new development, and so we’ll have to work through that. But the central challenge is kind of marrying up the expectations with the reality of what it takes to operate a power plant and the things that result in power plants being offline from time to time.

Those challenges are really — the trend is the contractual requirements of the data center tenants are incredibly stringent and heavily negotiated. So they do expect, you know, you hear of five nines availability. That’s their expectation. The solution is redundancy. So across generation, the equipment on a project in a microgrid, developers are building in a lot of redundancy so that things like scheduled maintenance can be performed on a unit-by-unit basis without taking down the capacity from below the guaranteed levels. Batteries also form part of the solution.

And so that’s really an engineering problem, and that’s how the power developers are resolving it and get you most of the way. But then the remaining part, and the real hard work, is negotiating availability guarantees and reliability guarantees in a way that gives the operators of the power plant, or if it’s a landlord who’s taken responsibility for the power supply, relief when they ought to receive it, because there are still going to be circumstances that are beyond the control of the landlord or the power generator that need to be addressed in these contracts. And to make them bankable and ensure there’s not an uncontrolled exposure on the power supply obligations, you really need to get that piece right.

Aubrey Bishai [00:17:32]

Thank you, Mark, and thanks for joining us for Powering Progress. If you’d like to explore more insights on AI infrastructure and the legal frameworks behind it, visit velaw.com.