DX Today | No-Hype Podcast & News About AI & DX
The DX Today Podcast: Real Insights About AI and Digital Transformation
Tired of AI hype and transformation snake oil? This isn't another sales pitch disguised as expertise. Join a 30+ year tech veteran and Chief AI Officer who's built $1.2 billion in real solutions—and has the battle scars to prove it. No vendor agenda. No sponsored content. Just unfiltered insights about what actually works in AI and digital transformation, what spectacularly fails, and why most "expert" advice misses the mark. If you're looking for honest perspectives from someone who's been in the trenches since before "digital transformation" was a buzzword, you've found your show. Real problems, real solutions, real talk. For executives, practitioners, and anyone who wants the truth about technology without the sales pitch.
DX Today | No-Hype Podcast & News About AI & DX
The Seventy Two Percent That Isn't There: Phantom Data Center Load and the AI Power Queue - August 20, 2026
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Welcome to the DX Today Podcast, your daily deep dive into the AI ecosystem. I'm Chris, and joining me as always is Laura. Today we are digging into the gap between the electricity that artificial intelligence data centers are asking for and the electricity they will actually take.
SPEAKER_01I have been waiting to talk about this one all week, because it might be the single most misunderstood number in the entire artificial intelligence build out. And almost every energy forecast you have seen this year quietly depends on it being right.
SPEAKER_00Set the table for me first. When we say a data center is asking for power, what does that request physically look like from the perspective of the utility sitting on the other side of the desk?
SPEAKER_01A developer files an interconnection request with a utility or a regional grid operator, stating how many megawatts they want, roughly where, and roughly when. It is a piece of paperwork, and critically, in most places, it has historically cost almost nothing to file.
SPEAKER_00So it is closer to a restaurant reservation than a signed lease. You put your name down, you hold a slot on a Saturday night, and nobody charges your card if you never show up for dinner.
SPEAKER_01That is exactly the right analogy, and it is precisely why these cues have become so absurd. Wood Mackenzie published an analysis on August 13, 2026, and the headline number in it is genuinely staggering.
SPEAKER_00Give it to me straight. I have seen the summaries floating around this week, but I would much rather hear the actual figures than somebody's aggregated hot take on them.
SPEAKER_01Across the United States, developers have requested 1,066 gigawatts of grid capacity for data centers. Wood Mackenzie expects roughly 298 gigawatts of that to reach actual grid commitment. That leaves 768 gigawatts of what the industry now calls phantom load.
SPEAKER_0072%. So nearly three out of every four gigawatts sitting on that national list is, in some meaningful sense, never going to be built by anyone, anywhere, ever.
SPEAKER_0172%, correct. And to put 1,066 gigawatts in perspective, the entire installed data center capacity in the United States right now is about 24 gigawatts. The request pile is more than 40 times what actually exists today.
SPEAKER_00Okay, that lands hard. Now the obvious question, and I suspect it is the exact question forming in every listener's mind right now, why would anyone request power they have no intention of ever using?
SPEAKER_01Because it is perfectly rational for each individual developer, even though it is destructive in aggregate. Power has become the binding constraint on building anything at all. So developers file with multiple utilities in multiple states for the very same project.
SPEAKER_00It is hedging behavior. You do not know which utility will say yes first, so you ask five of them, and then take whichever answer happens to arrive soonest.
SPEAKER_01Precisely, and the industry has a name for it now. They call it Q shopping. Peter Freed, who used to run energy strategy at Meta, described the mess bluntly. Tech companies are getting the same project bid into them multiple times, and their energy teams are so busy trying to isolate good quality opportunities from junk.
SPEAKER_00I love that quote because it reveals the second order problem underneath the first one. Even the hyperscalers, the people supposedly driving all of this demand, cannot reliably tell which proposals in front of them are real.
SPEAKER_01And it gets worse because a single building can appear three or four times in the national tally through completely different developers pitching the same site to the same customer. Astrid Atkinson, who runs Camue Energy, estimates that requests exceed actual data centers by five to ten times.
SPEAKER_00Five to ten times over. So the 72% phantom estimate might actually be the conservative reading, depending on how carefully you de-duplicate across regions and across separate developer entities.
SPEAKER_01It might well be, and the Wood Mackenzie analyst Ben Hertz Sargell points straight at the reason. He notes that much of the planned capacity belongs to new developers with a small number of massive, speculative projects. These are not the established operators.
SPEAKER_00Let us talk about that distortion because I think this is where the story stops being an interesting industry curiosity and starts becoming something that shows up on ordinary people's electricity bills.
SPEAKER_01This is the part that makes me genuinely angry. Utilities plan generation and transmission against forecast demand. If your forecast contains 768 gigawatts that never arrives, you either overbuild enormously or you paralyze yourself trying to work out what is real.
SPEAKER_00And overbuilding is very much not free. Somebody pays for those transmission lines and those gas turbines, whether or not the data center that justified them ever actually pours a foundation.
SPEAKER_01Rate payers pay. Look at what has already happened inside the PJM territory, which covers 13 eastern and Midwestern states. Its capacity auction went from $2.2 billion for the 2024 to 25 year to $16.4 billion for 2027 to 28.
SPEAKER_00That is more than a seven-fold increase in what an entire region pays purely to guarantee that generation capacity will be available when needed. And that cost flows straight through to residential bills.
SPEAKER_01It already has flowed through. National residential electricity prices are up around 8.25% year over year. And in Virginia, which hosts the densest data center corridor anywhere on Earth, prices are up more than 19%.
SPEAKER_00I want to push back gently here because I do not think all of that increase is fairly attributable to phantom projects. Some of it is real demand, real fuel costs, and real transmission that genuinely needed replacing anyway.
SPEAKER_01That is a completely fair correction, and honestly, I should have made it myself. Real load growth is real. The United States went from about 0.2% annual demand growth in the 2010s to a projected 5.7% through 2030.
SPEAKER_00Right, so the grid genuinely is waking up after decades of essentially flat demand. The question is not whether we need to build. The question is whether we are building against a signal or against pure noise.
SPEAKER_01And right now it is mostly noise, which is exactly why the regional numbers look so wild. Take the Texas grid operator. Its large load queue went from 63 gigawatts at the end of 2024 to 226 gigawatts by November of 2025.
SPEAKER_00Nearly quadrupled in 12 months. And I am assuming the overwhelming majority of that is data centers, rather than factories or hydrogen plants or anything else industrial in nature.
SPEAKER_01About 77% of that 226 gigawatts is large data centers targeting a grid connection by 2030. And Texas regulators are now scrutinizing something on the order of 474 gigawatts of requests statewide.
SPEAKER_00And on the other side of this, you now have utilities publicly saying out loud that they do not believe their own pipeline numbers, which is a remarkable thing for a regulated monopoly to admit in public.
SPEAKER_01Exelon is the clearest example of that. It has a 65 gigawatt pipeline and has said publicly that only about 22% of it is likely to materialize. Internally, it plans against roughly 11 gigawatts of high probability load over 10 years.
SPEAKER_00So Exelon is essentially applying a 78% haircut to its own order book. That is an extraordinary statement about how little the raw Q number actually means to the people who own the wires.
SPEAKER_01And that skepticism is spreading fast. Brian FitzSimons, who runs Grid Unity, says one major utility saw nearly 30% of its 2024 load applications simply canceled. He also noted that transmission owners were handling these processes using spreadsheets and project binders.
SPEAKER_00Spreadsheets. So what is actually being done about it? Because a problem this visible usually produces a regulatory response. And I know the federal regulator has been unusually aggressive on this front all year.
SPEAKER_01The Federal Energy Regulatory Commission moved on June 18, 2026. It issued show cause orders to all six regional grid operators, telling them to justify their existing large load rules or file changes within 60 days.
SPEAKER_00What are the actual reform categories in there? Because 60 days to rewrite the rules for connecting every large load in the country strikes me as a very tight clock indeed.
SPEAKER_01Five buckets in total. Faster application processes, cost transparency, rules for collocation and behind-the-meter generation, flexible transmission service, and study processes for generations sited near the load. The commission chair, SWET, framed the whole package as safeguarding consumers.
SPEAKER_00Flexible service is the interesting one to me by far. That is essentially telling data centers they can connect years faster if they agree to curtail during the worst hours, which is a genuinely elegant trade.
SPEAKER_01It is the single best idea in the entire package. A data center willing to turn down during the 60 worst hours of the year can connect far earlier because you are no longer sizing the grid around its absolute peak.
SPEAKER_00And states are clearly not waiting around for the federal process to finish either, from what I've read. There is a whole patchwork of screening mechanisms appearing at the state and even city level.
SPEAKER_01A real patchwork. Chicago now charges a $1 million fee for any request above 50 megawatts. Ohio requires large loads to pay for 85% of their projected energy use, whether they end up consuming it or not.
SPEAKER_00That Ohio rule is the sharpest tool on the entire list because it converts a free option into an expensive commitment. Phantom projects die instantly the moment somebody attaches a real price to the paperwork.
SPEAKER_01Virginia is requiring 14-year contracts from large load customers, and Texas passed a law directing its utility commission to require disclosure when a customer has filed multiple similar requests, with that rulemaking due by December of 2026.
SPEAKER_00If you make interconnection expensive enough to filter out the speculators, you also filter out legitimate startups and smaller operators who do not have hyperscaler balance sheets behind them. You entrench the incumbents entirely by accident.
SPEAKER_01That is a real cost, and I do not want to wave it away. High fees are regressive across company size. The counter is that phased fees, escalating as a project proves itself, preserve access while still punishing pure speculation.
SPEAKER_00Which is basically what any lender does when underwriting a construction loan. It is genuinely strange that the electricity system did not already work this way, given the amount of capital involved.
SPEAKER_01It is strange until you remember the history behind it. From about 1950 until 2020, this was a solved problem. Demand grew slowly and predictably, so there was no reason to build a sophisticated screening apparatus.
SPEAKER_00The system was optimized for a world that stopped existing roughly three years ago. That is the real story here, and artificial intelligence is simply the thing that finally exposed it.
SPEAKER_01Tom Bailey at Flexential made exactly that argument. Phantom data centers did not break the power grid. They proved that it was already broken. The interconnection process was built for 0.2% growth, not 5.7%.
SPEAKER_00And the transmission numbers back that up completely, do they not? I remember a fairly brutal comparison between what actually gets built annually and what the energy department says the country needs.
SPEAKER_01Brutal is the right word. The country built about 888 miles of high capacity transmission in 2024. The Department of Energy estimates we need roughly 5,000 miles every single year to keep pace.
SPEAKER_00So even if every phantom request vanished tomorrow morning and we were left with only the 298 real gigawatts, we would still be short on wires by a factor of five.
SPEAKER_01Correct. And that is exactly why I think the Phantom Load story is ultimately optimistic rather than pessimistic. The real number is roughly 300 gigawatts, not 1,000. That is hard, but hard is not impossible.
SPEAKER_00And it also suggests that some of the more apocalyptic forecasts about artificial intelligence energy consumption are being built on top of a Q number that the utilities themselves openly do not believe.
SPEAKER_01That is the part I would most want listeners to take away today. When you see a chart projecting artificial intelligence electricity demand, ask whether it is built on requests or on commitments, because those two differ by nearly a factor of four.
SPEAKER_00What should we be watching over the next few months as the concrete signal that this is actually getting cleaned up rather than just being talked about at conferences?
SPEAKER_01Three things specifically: the tariff filings coming out of that 60-day federal clock, the Texas rulemaking due in December, and whether reported pipeline numbers start falling as deposits and disclosure requirements begin to bite.
SPEAKER_00A falling cue would be the healthiest possible headline, and I suspect it will get reported as a slowdown in artificial intelligence investment when it is really just the noise finally being filtered out.
SPEAKER_01That is my prediction as well, and I want to be on record before it happens. The first big pipeline decline will get written up as the bubble popping. It will actually be a measurement correction.
SPEAKER_00That's all for today's episode of the DX Today podcast. Thanks for listening, and we'll see you next time.