How much of the AI data center boom is actually real?

Start with the thing that makes this measurable at all. To an electricity grid, a data center is simply a very large customer that turns up asking for as much power as a mid-sized city. It cannot just go and buy that power. It has to join a queue, and the grid operator has to work out whether the wires and the generating plants exist to serve it.

That queue is where the exaggeration gets caught, because each step costs the developer something more than the last. Announcing a project costs a press release. Entering a study costs paperwork. Signing an interconnection agreement costs a commitment. Posting collateral costs actual money. Pouring concrete costs a great deal of it. At every step a large share of the projects quietly walk away.

So the gap between what gets announced and what clears those steps is a rough measure of how much of the AI buildout is real. Nobody has to take our word for any of it: the grid operators, utilities and regulators publish these numbers themselves, in filings and testimony, and they have every reason to be conservative rather than promotional.

For a sense of scale, Texas alone is fielding about 474 gigawatts of connection requests, roughly 90 percent of it data centers. The all-time record demand across the entire Texas grid is 85.5 gigawatts. Governor Greg Abbott described the pile of requests as more than five times the state record when he ordered an audit of it in August 2026.

Pennsylvania counted its own data center projects at every step in August 2026. This is what the drop-off looks like.

The ones already in the ground. A rule that slows every new entrant does not hurt a project that is already sited, permitted and locally supported. If anything it widens the moat. These are the names with real Pennsylvania exposure that mostly clear the new bar.

The ones that will move anyway. These names have no Pennsylvania exposure we can find. They tend to trade on data center policy headlines regardless of geography, which is a sentiment reaction rather than a change in their business. Worth separating the two before treating a move as information.

There is a live precedent for that split. When New York paused state data center permits on July 14, 2026, TeraWulf finished as the biggest decliner among the AI infrastructure names we track and Nebius was close behind, despite Nebius having nothing to do with New York.

These are the counterweight. Real capacity is being built at very large scale, and these two figures are the cleanest evidence of it anywhere in the data.

Of all generation and storage capacity that entered a US interconnection queue between 2000 and 2020, 13 percent reached commercial operation. Attrition did not stop at the paperwork either: 41 percent of the capacity that signed an interconnection agreement between 2000 and 2022 had withdrawn by the end of 2025. The median project reaching operation in 2025 took 61 months from its request, up from 36 months in 2015.

This measures generation and storage, not data center load. Lawrence Berkeley National Laboratory states plainly that large loads sit in separate queues its report does not cover. It belongs here as the long-run base rate for how queue capacity turns into working plant, not as a data center figure. Lawrence Berkeley National Laboratory, Queued Up: 2026 Edition (June 2026)

This is deliberately not a count of blocked or cancelled projects. That number exists, it is quoted everywhere, and it is close to useless. It comes almost entirely from one organisation whose clients are the AI industry, its recent project-level data sits behind a paywall, its own pages disagree with each other, and its delayed column contains projects that were later approved. Announcement froth is also perfectly normal: companies have always announced more than they build, and counting the froth on its own tells you nothing.

What matters instead is the conversion rate, and whether it is getting worse. A project that clears these screens has secured land, defined how it will be powered and, in the strictest cases, posted collateral. Those projects are largely on schedule. The useful question is how much of the announced pipeline ever becomes one of them, because permits and interconnection agreements run 18 to 36 months ahead of concrete. That makes this a reading on capacity arriving in 2028 through 2030, which is where the valuations actually live, rather than on anything being delivered this year.

Two things cut against a bearish reading, and they belong here rather than in a footnote. Local defeats often do not stick: Saline Township, Michigan rejected a $16 billion campus, was sued two days later, settled for about $14 million in community benefits and saw ground broken within two months. And friction runs both ways, because West Virginia has moved in the opposite direction and preempted local ordinances that limit large data center projects.

Every figure here comes from the screening body itself: an ISO presentation, a public utility commission filing, an earnings call, a federal reliability assessment or a state executive order, each linked in the row it supports. Nothing is estimated, modelled or interpolated. The screens are deliberately not added together, because they cover different regions, vintages and definitions, so the range and median describe the spread across them rather than a total.

• Pennsylvania will not permit a data center until the town says yes
• Texas froze new data center grid connections pending an audit
• New York paused state permits for new data centers
• ECB researchers say an AI stock market correction is likely
• AI infrastructure backlog leaderboard

Last updated August 19, 2026. This is a record of what grid operators, utilities and regulators have published about their own screening, not a forecast and not investment advice. Figures move only when one of those bodies files something new, so this page updates as filings land rather than live.