$710 Billion and Nowhere to Plug It In

Microsoft. Google. Amazon. Meta. Four companies. $710 billion in committed capital expenditure for 2025 and 2026. That number is not a projection — it's a disclosed, investor-facing commitment. It represents the single largest coordinated infrastructure buildout in the history of private enterprise. And it has a problem that no amount of capital can immediately solve.

The bottleneck isn't compute. The chips exist. The cooling technology exists. The construction crews exist. The capital definitely exists.

The bottleneck is power — specifically, the physical infrastructure required to deliver it at the scale these buildouts demand. And that infrastructure operates on a timeline that doesn't care about quarterly earnings calls or hyperscaler urgency.

What $710 Billion Actually Buys

To understand the constraint, you have to understand what this capital is actually being deployed to build. AI training and inference workloads are extraordinarily power-dense. A single AI data center cluster can require 50 to 100 megawatts of continuous power draw. A large campus-scale buildout — the kind Microsoft, Google, and Amazon are planning across multiple US markets — can demand 500 megawatts to a gigawatt or more.

For reference: one gigawatt of continuous power is roughly equivalent to the output of a large nuclear power plant. It can power approximately 750,000 average American homes.

The hyperscalers aren't building one of these. They're building dozens, simultaneously, across the United States and globally. The capital is there. The demand signal is clear. The execution bottleneck is at the grid.

The Queue Nobody Talks About

Here's the number that reframes everything: grid interconnection queues in major US markets currently run three to five years. That means if you submit an application today to connect a new large-scale power consumer — say, a 200-megawatt data center — to the transmission grid in a constrained market, you're looking at 2029 to 2031 before you're operational. Maybe later. This isn't a bureaucratic inefficiency that can be engineered around. It's a physical reality. Transmission infrastructure takes years to permit, finance, and build. Transformer lead times — a component most people have never thought about — have extended to 18 to 24 months in some markets due to demand surges. Substation upgrades require coordinated shutdowns, regulatory approvals, and construction timelines that don't compress under pressure. The queue isn't a line you can cut. It's a calendar.

Virginia Is Full

The clearest illustration of the constraint is Northern Virginia — specifically, the corridor around Ashburn that handles roughly 70 percent of US internet traffic and hosts the highest concentration of data center capacity on the planet. Virginia is effectively closed to new large-scale utility connections.

Dominion Energy, the primary utility serving the region, has publicly communicated that its transmission system cannot accommodate significant new load additions without major upgrades that are years away from completion. Data center developers who showed up in 2023 expecting to replicate the previous decade's growth trajectory found a very different conversation than they expected.

The land is available. The capital is available. The power is not.

This isn't a Virginia-specific problem. It's Virginia as the leading indicator. Phoenix is experiencing similar pressure. Dallas is tightening. Northern New Jersey — the next largest US data center market — is watching its own interconnection queue extend. The constraint is propagating across every major market where digital infrastructure has historically concentrated.

Permitted, Energized Land Is a New Asset Class

When a physical constraint becomes binding at scale, the assets that sit on the right side of that constraint don't just become more valuable — they become a different kind of asset entirely. Permitted, energized land — sites that already have grid interconnection approved, substations in place, and power delivery infrastructure operational — has crossed that threshold.

These sites existed before the AI infrastructure buildout accelerated. They were developed for earlier generations of data center demand, industrial use, or utility-scale operations. Many of them are underutilized. Some are sitting idle. A few are actively being repositioned.

But here's what's changed: the hyperscalers and their contractors are actively hunting for them. When Microsoft or Google needs to deploy 200 megawatts in a given metro area and the standard interconnection queue runs four years, a site that already has that power infrastructure in place isn't just convenient — it collapses a four-year timeline to months. The value of that compression, at the scale these operators are working, is measured in billions. That's not a real estate transaction. That's a strategic infrastructure acquisition.

What Happens Next

The capital doesn't stop flowing because the grid is constrained. It redirects. Operators who planned ahead — who secured power infrastructure, private generation capacity, or grid interconnection agreements before the queue extended — are now holding assets the market desperately needs. Private equity firms that acquired data center land and power rights at 2020 and 2021 prices are sitting on positions that look prescient in retrospect.

The next phase of the buildout will be characterized by three things:

Private power development. Solar arrays, gas peakers, fuel cells, and eventually small modular reactors positioned directly on or adjacent to data center campuses. The regulatory environment — particularly FERC Order 1920 and state-level renewable mandates — is pushing operators toward private generation at exactly the moment hyperscaler demand makes it economically viable. Asset repositioning. Industrial sites, legacy power plants, and underutilized commercial facilities with existing grid infrastructure will be acquired and converted. The due diligence criteria for these deals will look nothing like traditional real estate — power capacity, interconnection status, and transmission headroom will be the primary value drivers.

Geographic redistribution. Markets that have historically been secondary data center locations — parts of the Midwest, the Southeast, rural markets with excess transmission capacity — will receive investment that would have been unthinkable five years ago. The constraint is forcing diversification.

The Bottom Line

$710 billion committed. Grid infrastructure that can't absorb it fast enough. A three-to-five year interconnection queue standing between capital and deployment. This is not a temporary bottleneck that gets resolved with the next infrastructure bill or utility upgrade cycle. It's a structural gap between the pace of AI demand and the pace at which physical power infrastructure can be built. The operators who understand that gap — and who hold assets positioned on the right side of it — are not just well-positioned for the current cycle. They're holding the infrastructure that the next decade of AI runs on.

Rich Washburn is a technology strategist and AI infrastructure advisor. He works at the intersection of AI deployment and physical infrastructure through his work with Data Power Supply and Eliakim Capital.