I’ve been watching the numbers for a while now, and something still doesn’t add up. Developers keep announcing enormous data center projects that would need enough electricity to light entire cities, yet the grids keep quietly signaling they cannot deliver. When the latest projections showed that more than two-thirds of the power sought for these facilities will probably never materialize, I wasn’t surprised. I was just relieved someone finally said it out loud.
The Scale Of The Mismatch Is Hard To Ignore
Picture this. Grid operators across the United States have been handed requests totaling more than a thousand gigawatts for data center projects alone. That figure sits close to the entire utility-scale generating capacity the country currently has. One traditional nuclear reactor equals roughly one gigawatt. So the industry is essentially asking for the output of more than a thousand new reactors. Nobody seriously expects every single request to turn into concrete and servers. Still, the sheer volume creates real headaches for the people trying to keep the lights on.
In one major state the interconnection queue ballooned to nearly five times the record peak demand the system has ever seen. Most of those requests came from data centers. Regulators eventually hit the pause button. That single decision put billions in potential revenue on hold and slowed a sizable share of the national pipeline. I’ve spoken with people who work these queues, and the consensus is clear: a large portion of the applications are speculative, duplicated, or simply unrealistic.
Phantom Projects And The Shotgun Approach
Developers have learned to cast a wide net. They file the same project with multiple utilities, then wait to see which one offers the best terms and the fastest path. The applications that don’t win get abandoned. Those abandoned filings still sit in the queue for months or years, clogging the system and distorting forecasts. Industry analysts now estimate that only a fraction of the requested power will ever support an actual facility.
Some estimates put the credible portion somewhere between twenty and thirty percent. Others land a bit higher once you account for projects already under construction or backed by well-known tenants. Either way, the majority of the announced capacity is unlikely to appear. That gap between the headline numbers and the probable outcome is what some quietly call the bragawatt problem. It sounds clever until you realize how much planning and capital get wasted chasing ghosts.
Grid operators don’t know which ones are real and which ones aren’t.
That uncertainty forces utilities to spend time and money studying projects that may never break ground. The cost of those studies and the eventual infrastructure upgrades still gets socialized through rates. Ordinary customers end up paying for capacity that was never going to be needed. I’ve found that part particularly frustrating. The AI story is supposed to create value, not quietly transfer risk onto households and small businesses.
Why First-Time Developers Struggle
Not every applicant has a track record. Many first-time developers propose facilities that would rival mid-sized cities in power draw. Their projects tend to be larger on paper and less likely to finish. Established technology companies with strong balance sheets and signed tenants clear the bar more easily. Utilities have responded by raising the bar themselves. Upfront collateral, near-perfect credit scores, and hefty application fees now filter out a lot of the noise.
The shift from “first-come, first-served” to “first-ready, first-served” makes sense on paper. In practice it still leaves smaller players scrambling. Some of those smaller players used to build the facilities and then sell them to larger operators. That model is under pressure. When the queue is closed for business, the entire chain slows down.
In my experience the most credible projects share a few traits. They already have physical progress on the site. They have investment-grade tenants lined up. The sponsor has delivered similar facilities before. Press releases and PowerPoint decks do not count. When analysts apply those filters, the pipeline shrinks dramatically.
Regional Differences Tell Different Stories
Not every grid is drowning in the same way. One large interconnection region that serves tens of millions of people still sees roughly half of its applications as reasonably credible. Another fast-growing state sees closer to fourteen percent. The difference often comes down to how aggressively developers have targeted the area and how transparent the screening process has become.
Some utilities have already cut their own internal pipelines by nearly forty percent after applying stricter criteria. They are choosing the projects most likely to reach commercial operation and setting the rest aside. That kind of pruning is healthy, even if it disappoints people who prefer the more optimistic forecasts.
The broader national picture remains sobering. The total volume of requests still represents a large share of existing generation capacity. Building the transmission, substations, and generation needed to support even the credible portion will take years. In the meantime the queues stay long and the uncertainty stays high.
What The Numbers Mean For Actual Build Rates
Analysts who dig into the data rather than the announcements tend to land on similar conclusions. Out of several hundred gigawatts currently tracked in detailed databases, roughly one-third look credible once you apply realistic filters. Construction progress carries the most weight. Tenant quality comes next. Sponsor experience matters but less than the first two. Legislative and community opposition exist, yet they have not been the primary reason projects disappear so far.
The timing question is almost as important as the volume question. Even the more optimistic pipelines show target delivery dates that look aggressive. A large share of the capacity labeled for the next couple of years is highly unlikely to arrive on schedule. A more realistic horizon stretches five to ten years for the credible portion. That slower ramp changes the entire investment thesis for companies counting on rapid power availability.
I’ve watched similar cycles before. Semiconductor companies once double- and triple-ordered equipment during tight markets. When the excess capacity finally arrived, the correction was painful. Energy markets can behave the same way. The difference here is that the “equipment” is power plants and transmission lines that take far longer to build and cost far more to idle.
The Cost Of Getting The Forecast Wrong
Utilities and grid operators have to plan years ahead. They cannot simply ignore the queue. Yet treating every application as real would lead to massive overbuilding. Treating too many as phantom would leave the system short when genuine demand arrives. The middle path requires better screening, higher barriers to entry, and more transparent data sharing. Progress is happening, but it remains uneven.
Ratepayers sit in the middle of this tension. Capital spent on studies, upgrades, and reserves eventually shows up in monthly bills. When a large share of that capital turns out to have supported projects that never materialized, the public has every right to ask hard questions. The industry’s answer so far has been that the alternative—slowing the entire AI build-out—would be worse. That argument only holds if the credible projects actually get the power they need. Right now the system is still sorting that out.
Perhaps the most interesting aspect is how little public discussion focuses on the filtering process itself. Headlines celebrate the next giant campus announcement. Few dig into the probability that the campus will ever draw the power it claims. The gap between announcement and reality is where the real story lives.
Practical Steps Utilities Are Already Taking
Several practical changes are already visible. Higher application fees and collateral requirements push speculative filings out early. Credit reviews favor established players. Some operators now require evidence of site control, financing progress, or signed offtake before advancing a project deeper into the queue. These measures slow the process for everyone, yet they improve the signal-to-noise ratio.
Another approach involves staged screening. Projects move through multiple gates. At each gate the developer must demonstrate additional readiness. Those that stall get removed. The method is imperfect, but it beats treating every filing as equally serious.
- Require meaningful financial deposits early
- Demand proof of site control and zoning progress
- Prioritize projects with committed tenants
- Apply credit and experience thresholds
- Publish clearer queue statistics so markets can adjust
None of these steps solves the underlying generation shortfall. They simply make the shortfall clearer. Once the phantom projects are filtered out, the remaining gap still looks substantial. Closing that gap will require new generation, new transmission, and new operating practices. All of those take time and money.
The Investor Angle Few Want To Discuss
Capital has already flooded into data center development, related equipment, and the supporting energy infrastructure. Some of that capital will support real projects that generate real returns. A meaningful share will not. When markets finally price the difference between announced capacity and delivered capacity, the adjustment could be sharp.
I’ve seen this pattern in other infrastructure cycles. The early announcements create a narrative of unstoppable demand. Later the physical constraints reassert themselves. The companies that planned for realistic power availability tend to fare better than those that assumed every press release would turn into megawatts. Right now the market still seems to lean toward the more optimistic story. That may change once more utilities publish their filtered pipelines and more states follow the example of pausing new approvals until reviews are complete.
The question is not whether AI needs power. Of course it does. The question is how much of the currently claimed power will actually be available on the timelines the industry is using in its models. The latest independent assessments suggest the answer is far less than the headlines imply.
Looking Ahead Without The Hype
None of this means the data center boom is finished. Credible projects with real tenants and real construction progress will continue. Power will be found for many of them, sometimes through behind-the-meter generation, sometimes through creative contracting, sometimes through accelerated transmission upgrades. What will not continue is the assumption that every announced gigawatt is equally likely to appear.
The healthier path is the one already emerging: stricter screening, clearer data, and more honest timelines. That path will disappoint people who preferred the larger numbers. It will also reduce the risk that ratepayers and investors end up funding capacity that never gets used.
In the end the math is straightforward. The grid has finite capacity. Building more takes years. Requests that exceed both current and near-term capacity by multiples cannot all be real. Once the industry and the public accept that reality, the conversation can shift from chasing phantom load to delivering the power the genuine projects actually need. That shift feels overdue.
The next few years will show how many of today’s announcements survive the filters. My own expectation is that the final build rate will land closer to the conservative estimates than the optimistic ones. If that proves true, a lot of capital and a lot of planning will need to be reassessed. Better to start the reassessment now than wait until the gap becomes impossible to ignore.
Power remains the quiet constraint on the entire AI infrastructure story. Until the queues are cleaned and the credible projects are properly prioritized, the gap between ambition and reality will keep widening. The numbers already point in that direction. The only remaining question is how long it takes for the rest of the market to catch up.
