The capital problem is solved. The deployment problem isn't.

Moody's now projects that the six largest US hyperscalers will spend approximately $700 billion on data-centre capital expenditure in 2026 — roughly six times their combined 2022 outlay — and forecasts that the US data-centre market will surpass $1 trillion by 2030. Those are the headline numbers. The structural change underneath them is more interesting, and considerably more consequential for any institution allocating capital into infrastructure this vintage.

For most of the last fifteen years, "data-centre infrastructure" sat awkwardly between technology equity and traditional real estate. It was too operationally intensive to be core property, too capital-intensive to be venture, too concentrated geographically to be conventional infrastructure. That awkward position is over. The category has institutionalised. The question for allocators is no longer whether the sector is investable. It is what wins inside it now that everyone agrees it is.

The answer is unintuitive.

The pool arrived in every structure

The clearest evidence that AI-infrastructure has crossed the line into a mature institutional asset class is the diversity of vehicles bringing capital to it.

Digital Realty closed $3.25 billion of equity earlier this year for its first commingled data-centre fund, taking 80 per cent stakes in five properties — a transaction structured and priced as core real estate, not technology speculation. Blackstone is launching the Digital Infrastructure Trust under the BXDC ticker with a target of approximately $2 billion, a pure-play AI-data-centre REIT bringing the category into the retail-accessible part of the market at scale for the first time. Crusoe, Blue Owl and Primary Digital are now in the second phase of a $15 billion joint venture funding 1.2 gigawatts across six buildings at Abilene, Texas, energising mid-2026 — a template for off-balance-sheet financing of single-tenant AI campuses that other consortia will copy. Equinix and CPP Investments are acquiring the Nordic operator atNorth for around $4 billion, layering pension capital onto a renewable-powered consolidation play.

The capstone is the sale of Aligned Data Centers to a consortium of BlackRock GIP, the AI Infrastructure Partnership and MGX for approximately $40 billion, expected to close in the first half of 2026 — the largest digital-infrastructure transaction on record, and one that resets the valuation benchmarks across the asset class.

Outside the United States, the same dynamic plays out at sovereign and corporate scale. SoftBank has indicated up to €75 billion for roughly five gigawatts of capacity across France. Thailand's Board of Investment has approved a TikTok-linked $29 billion hub. Meta and Reliance are advancing inside an Indian AI ecosystem now estimated at approximately $400 billion of committed capital. Capital has arrived from every direction.

Abundance moves the bottleneck

When a pool of capital this large arrives this quickly into an asset class with a finite supply of investable projects, the binding constraint moves. It does not disappear. It moves.

For a decade, the constraint on AI-infrastructure deployment was raising the money. Funds needed to convince LP committees that data centres belonged in their infrastructure allocations, that hyperscaler revenue was creditable, that long-tenor capex against intangible offtakers was bankable. That argument has been won. LP committees no longer require persuasion; they require deployment.

What they will discover, if they have not already, is that every vehicle in the market is now bidding for the same finite pool of closeable, qualified deals. The pool is bounded — by powered land that can credibly host or import the megawatts required, by counterparties that can underwrite and operate at hyperscale, by structures that can survive lender diligence and reach financial close inside a window that matters. Sovereigns are bidding into it. Pensions are bidding into it. Retail REIT vehicles are bidding into it. Strategic balance sheets are bidding into it.

At jumbo scale, the cost of getting the matching wrong compounds non-linearly. A failed $400 million term sheet costs everyone involved a few million in fees, advisory time and lost opportunity. A failed or restructured $15 billion joint venture costs an order of magnitude more — eight or nine figures in committed fees, opportunity cost, and the calendar months of deployment a fund cannot recover. The constraint that used to be "do we have the money" is now "do we have the deal." And the cost of being wrong about a specific deal is now structural, not incremental.

Defining the scarce asset

The phrase "high-quality deal flow" is loose enough to mean nothing, so it is worth being specific about what makes a deal high-quality in the current market.

A high-quality deal is one in which the technology counterparty has bankable performance data and a credible balance sheet; the site has a verifiable power path inside the offtaker's window, whether grid interconnection, behind-the-meter generation, a firm-power purchase agreement, or a hybrid; the regulatory exposure has been mapped and is clean against current rules; and the deal is already shaped to a form a financier can underwrite without reconstructing it. The fund evaluating it is choosing between options that have been pre-qualified, not sourcing and de-risking from raw inbound.

That is a different procurement problem from the one most infrastructure funds were built to solve. Traditional infrastructure deal-flow processes were built for slower velocity, lower deal complexity, and more linear counterparty alignment. The current market does not give a fund eighteen months to convert a memo into a closing. The funds that will deploy this vintage at scale will be the ones that figure out, by build or by partnership, how to be in front of pre-qualified deals before peer funds arrive at the same table. The capability is sourcing infrastructure, not capital infrastructure.

The double opportunity

For a fund manager, the consequence of this shift is a double opportunity, and the two halves of it matter equally.

The first half is deployment quality. The vintage will be won by the funds that reach qualified deals first and deploy against them with the right structure. Sub-deployment risk is concentrated, in 2026 and 2027, in the mid-market — funds with $300 million to $2 billion in committed capital, large enough to need a steady cadence of qualified inbound and small enough to be outbid at the top of the market by the consortia closing transactions at $40 billion. The mid-market fund that solves its sourcing problem will return capital and re-fundraise. The mid-market fund that does not will spend two years writing apologetic LP letters.

The second half is harder to put a number on but matters more strategically. The same matching capability that delivers deployment also routes capital into the energy infrastructure that actually gets built. Nuclear and small modular reactor PPAs are becoming the default AI baseload thesis — illustrated by Meta's roughly 6-gigawatt arrangements with Oklo, TerraPower and Vistra, and the 1.92-gigawatt AWS–Talen transaction that has now been widely studied. Chevron is pursuing additional behind-the-meter data-centre power partnerships. Prometheus Hyperscale has been approved for a 506-acre Wyoming campus targeting 1.25 gigawatts initially and 5 gigawatts at full build. QTS is advancing a $10 billion mega-site in Van Wert, Ohio.

These are not data-centre deals separable from energy deals. They are energy-infrastructure deals dressed in data-centre clothes. The funds that win the sourcing race for these transactions are, in aggregate, financing the rebuild of the US energy industry — firm power, grid upgrades, behind-the-meter generation, storage, and the connective tissue between them. That is a category of impact that institutional capital has spent more than a decade searching for a credible vehicle to express. The vehicle, finally, is here.

What we built to solve this

This is the work we built SitePower to do. The matching layer that sits between technologies, sites and capital — qualifying each side before introduction, pre-shaping the structure to a form that can close, and routing pre-qualified deal flow to the financiers whose mandate, geography and tenor actually fit. We do not place capital, we do not raise funds, and we do not take fees from allocators; SitePower's economics come from sponsor-side carry, co-investment, and operating roles inside the deals we help shape. What we deliver to financiers is access to the resource that has become the binding variable in this market: closeable, qualified deal flow, faster than the peer set is reaching it. If your fund has a mandate to deploy into AI-infrastructure or the energy backbone underneath it, the qualification gate is at sitepower.ai/apply/financiers.