Funding the Power Gap: Who Finances AI-Era Energy Infrastructure?
Written by Bruno P. Allaire
Trillions in generation and transmission need is colliding with a project finance model built for nine-figure tickets. The gap won't be filled by banks alone, and that opening is where new capital structures earn their place.
Share this perspective
Sizing the Gap
Start with the shape of the number, because its shape matters more than its size. Forecasts of AI-era power infrastructure spending over the coming decade now run well into the trillions of dollars, McKinsey puts the capital need for data center infrastructure alone at $6.7 trillion through 20301, and the instinct is to treat that as an abstraction. It isn't. It is a population of physical projects: gas turbines and reciprocating units, pipeline laterals and storage, transmission upgrades, grid equipment, and the balance of plant nobody writes headlines about. Each of those projects has to be financed individually, on its own merits, by someone who will hold the risk.
The scale is the first insight. An obligation this large cannot be absorbed by any single channel of capital. The banks cannot do it, the utilities cannot rate-base it fast enough, and the hyperscalers will not put all of it on their own balance sheets, whatever their announcements imply. The money will have to come from everywhere at once: institutional balance sheets, private credit, infrastructure funds, family offices, and individual accredited investors. When capital has to come from everywhere, the interesting question is which structures can reach it.
That question is usually discussed as a technology story. It is actually a finance story, and it is the story I have spent my career on: what happens when real assets need capital at a scale and in a granularity the incumbent system wasn't built to serve. The AI buildout is simply the largest recent instance of an old problem.
Where Traditional Finance Falls Short
The incumbent system has three channels, and each fails the AI-era pipeline for a different reason. Bank project finance is engineered for scale. Lenders commit to deals large enough to justify their own diligence and monitoring overhead, which in practice means nine-figure tickets to investment-grade counterparties. The utility channel is constrained by its balance sheet and its politics: rate-base growth is a negotiation with regulators and ratepayers, and it moves on a timeline that has nothing to do with a model release cycle. And the public markets, as I can attest from personal experience, simply will not carry a mid-market energy issuer; the research and listing economics collapse below a certain capitalization, and the gap between what the assets warrant and what the tape will fund becomes a permanent feature of the CEO's life.
I lived that gap from the inside. I took an oil and gas company public through a reverse merger, ran it as a small-cap issuer, and spent years watching our projects clear every technical bar while the available capital structures passed us by. Below the bankable threshold, an asset either waits for a balance sheet to adopt it or it doesn't get built. That lesson is why I have approached the AI power buildout less as an technology opportunity than as a capital formation one: the constraint I recognize is not electrons, it's who can write the check and on what terms.
What makes this cycle different is that the stranded middle has arrived at power-industry scale. The assets the buildout needs most, in the aggregate, are exactly the ones the incumbent channels skip: too small for project finance, too merchant for rate base, too infrastructure-like for venture capital. A financing system that only clears mega-deals will clear a fraction of the pipeline the demand forecasts assume.
The Mid-Market Generation Opportunity
Look at what the queue-busting strategies from my last article actually require, and the mid-market inventory writes itself. Peaking capacity and fast-start reciprocating units for campuses that need firming before the grid matures. Behind-the-meter gas generation for co-located load. Pipeline laterals and fuel storage that turn an interruptible fuel arrangement into a contractable one. Storage hybrids and the interconnection upgrades that make them useful. These are five-to-fifty-million-dollar assets. They have contracted offtake, measurable heat rates, and useful lives measured in decades. They cash flow like infrastructure, and they fund like orphans.
This is precisely the market need that led us to form Defiant Power Solutions. Sitting where I sit, watching the demand shock collide with an interconnection queue that peaked at nearly 2,600 gigawatts of active requests, with a median of more than five years from request to commercial operation for the projects that do get built2, it was clear that the bridges and co-located builds the AI era needs would not be financed by the channels that serve utilities and hyperscalers, and would not be built by developers who couldn't underwrite fuel supply, equipment, and permits as one integrated discipline. The company was built for that seam: deliver firm power to compute on the compute's timeline, as a sponsor rather than a consultant. The venture isn't a bet on AI. It is a bet that the stranded middle of power infrastructure is a permanent, profitable market, whoever the anchor tenant turns out to be.
The opportunity is broader than one sponsor can serve, which is the point. There is room for a population of disciplined mid-market operators building the same inventory, and the binding constraint on all of them is the same: none of them match the ticket sizes the incumbent capital channels were built for.
New Capital Stacks
So how does the gap actually get filled? In layers, and the layering is already visible. The hyperscaler balance sheets take the top: anchor campuses and gigawatt-scale commitments funded on investment-grade credit, which solves nothing below the top of the market but does absorb the headline projects. Private credit has moved in beneath it, financing the merchant generation and the fuel infrastructure at rates that reflect real risk and real diligence, and it will take more of the stack as the asset class demonstrates delivery. Equipment vendors are financing their own sales, a classic late-cycle signal that deserves an article of its own.
Beneath those layers sits the segment I find most interesting: assets that are too small for private credit's diligence economics and too numerous for direct relationships to cover. That is where fractional structures earn their place, and it is the argument I have made in my tokenization articles applied to a new asset class. A partial tokenization of a cash-flowing mid-market generation asset, sold to accredited investors under the Reg D framework with board-grade reporting, is an administration-cost solution for precisely the population the banks skip. And the framework is not a niche channel: issuers raised roughly $1.8 trillion through Regulation D offerings in 2022 alone3. The digital rails don't create the investability. The offtake contract and the heat rate create the investability. The rails make the check sizes workable and the reporting affordable at a scale private placement overhead never could.
None of these layers is speculative on its own. Each is a familiar instrument: project equity, private credit, a Reg D private placement. What is new is that they are stacking, because the pipeline is too large and too granular for any one of them to clear alone. The sponsors who understand how the stack fits together will have access to capital their competitors are still waiting for.
The Discipline Premium
I want to be honest about what hype does to a capital market, because I have watched it happen twice now, in digital assets and in energy. AI-power enthusiasm will produce its share of mispriced projects. There will be generation built on spec against announced load that never materializes, fuel infrastructure against campuses that don't clear permitting, and sponsorships whose real product is the story of the buildout rather than a kilowatt. The demand forecasts are directional, not contractual, and finance that forgets the difference gets corrected eventually.
The underwriting standards that separate durable energy finance from narrative finance are old and unforgiving. Contracted offtake, or load so proximate it is functionally contracted. Firm fuel at a basis you can survive. Equipment procurement with delivery dates backed by something stronger than a salesperson's optimism. A permit posture that reflects the state you are actually in, not the state you wish you were in. And a capital structure that can carry the asset through a dip in the hype cycle without a forced sale. That last one is where sponsors fail, and it is why the trough-era lesson applies here too: the operators who can hold through the correction are the ones who bought their discipline before the market demanded it.
The discipline premium cuts in the sponsor's favor, though, and that is worth stating as plainly as the risk. In a market awash in narrative, cash-flowing assets with boring documentation stand out. A mid-market generation project with a real offtake contract and reconcilable financials is, right now, one of the easiest capital raises in energy precisely because so much of the competition for capital is uninvestable on its face.
Getting Deal-Ready
Which brings the argument to the practical question: what should a sponsor have in place before approaching any of these capital sources? The list is short and mostly unglamorous. Contracted offtake, or a letter of intent from a load that has committed capital behind it. Firm fuel supply, priced and term-matched to the offtake. Permit posture documented well enough that a lender's counsel can verify it in a week. A clean entity holding the asset, with financials a CFO can hand over without apology. A reporting cadence the sponsor can actually sustain, because the obligation outlives the closing.
This is the same preparation standard we advise property and asset owners through Defiant Consulting, and the reason we formed the venture is a corollary of everything above: the owners sitting on the land, the fuel, and the proximity to load are frequently energy-adjacent rather than energy-native, and they don't know what their position is worth to the AI buildout or how to make it fundable. Advising them is not a sideline to the power business, it is the upstream of it. Every well-prepared site that reaches the capital market with its offtake, fuel, and permits in order makes the whole stack work better, and every unprepared one is an argument for hiring someone who has sat on the sponsor's side of the table.
The gap between what the AI era needs and what the incumbent system funds is not going to close on its own. It closes deal by deal, sponsor by sponsor, as operators prepare assets the new stack can actually price. The frenzy will eventually cool, and when it does, the correction will fund the prepared and bury the announced. That has been true in every energy cycle I have worked, and I have never seen a reason, at any scale, for it to change.
Notes & Sources
- 1.McKinsey projects that data centers will require $6.7 trillion in cumulative capital investment worldwide by 2030 to keep pace with demand for compute power. McKinsey & Company, "The cost of compute: A $7 trillion race to scale data centers," April 2025
- 2.The US interconnection queue peaked at nearly 2,600 GW of active generation and storage requests at the end of 2023, and the median duration from interconnection request to commercial operation exceeded five years for projects built in 2025. Lawrence Berkeley National Laboratory, "Queued Up" (2025 and 2026 editions)
- 3.Issuers raised approximately $1.85 trillion through Regulation D offerings in 2022, according to SEC staff statistics on exempt offerings. SEC Division of Economic and Risk Analysis, "Market Statistics of Exempt Offerings," 2025