The Utility CEO Just Said AI Demand Isn't Slowing — and He's the One Who Has to Build the Power

The Utility CEO Just Said AI Demand Isn't Slowing — and He's the One Who Has to Build the Power

Let me tell you something about reading tea leaves in this AI buildout: the most important interviews are never the ones with the chip CEOs. The chip guys are selling you the future. The utility CEO is the one who has to physically build the thing that makes the future work — the power plant, the transmission line, the substation — and he's the one who gets blamed when the lights flicker. So when Chris Womack, the CEO of Southern Company, sat down with Bloomberg today and said AI data center demand isn't slowing, I paid attention. Not because I trust him. Because I know exactly what he's selling and why.

Southern Company is the utility behind Georgia Power — my home state's electric company. It's one of the biggest vertically integrated utilities in America, and it's sitting in the middle of the biggest demand surge since the dot-com days. Womack told Bloomberg the company's pipeline of data center projects keeps growing, pointed at the 25-year power deal it signed with OpenAI, and made the case that the AI buildout is not a bubble about to pop. That's a strong statement. It's also a statement that needs to be read with one hand on your wallet. Let me walk you through what he said, what the numbers actually show, and why the incentives matter more than the headlines.

The Setup — What the Utility CEO Just Said

Here's the thing about utilities: they don't do hype. Regulated monopolies don't need to pump a stock. They need approval from state regulators to build things, and they need demand to justify the build. So when Womack goes on Bloomberg and says demand isn't slowing, he's not doing a victory lap for shareholders — he's laying the groundwork for the next round of rate cases and the next Integrated Resource Plan. The message is simple: we need to build more, and here's why.

And to be fair, the man has receipts. Southern just reported a second quarter where data center electricity usage was up 55 percent year over year — not a projection, an actual meter reading. System-wide data center load now exceeds 1.2 gigawatts, up more than 500 megawatts in a single year. Retail electricity sales grew 2.3 percent in the first half of 2026, which management called the highest growth through June in nearly two decades. This is not a company that's inventing demand. The meters are moving.

The Numbers — This Isn't a Story About Vibes

Here's the scoreboard from Southern's own earnings call, and it's worth sitting with these numbers for a second:

Contracted large load — the signed, collateralized deals — is now over 17 gigawatts scheduled through the mid-2030s. They added 6 gigawatts of new contracts in a single quarter. The prospective pipeline is over 75 gigawatts in various stages of development, with 8 gigawatts in late stages and 3 gigawatts nearing finalization. Data center usage up 55 percent. Commercial sales up 7.4 percent in the quarter. Adjusted earnings per share of $1.13, up 21 cents, and management guiding to the top of the $4.50-to-$4.60 full-year range.

And here's the part that should make every hosting provider in the Southeast sit up: the company says it costs roughly $2 billion per gigawatt of new generating capacity. They got approval for 10 gigawatts of new company-owned generation — thermal, battery, and solar — and the CEO says the OpenAI contract alone pushes them about 1 gigawatt beyond what was recently approved. That's the scale we're talking about. Not megawatts. Gigawatts, plural, at two billion dollars a pop.

The OpenAI Deal — 3.2 Gigawatts and a Gigawatt That Can Turn Off

The centerpiece of Womack's Bloomberg appearance was the deal with OpenAI — a 25-year electric service agreement for a planned data center in Effingham County, Georgia, that's expected to need roughly 3,200 megawatts. For context, that's more than three times the 1.2 gigawatts of data center load Southern has on its system today, serving all of its customers. One facility. Three times the current AI load of the entire Southeast utility.

But read the fine print, because this is where the story gets interesting. OpenAI has agreed to provide up to 1,000 megawatts of demand response to Georgia Power's system in times of high demand — one of the largest single-site demand response commitments in the country. Womack described it as the ability to "shave the peak," meaning the AI facility itself can dial down by a gigawatt when the grid is tight. OpenAI pays the full infrastructure and electric service costs. There's an $80 million community investment fund, $71 million in Codex credits for Georgia students, thousands of jobs promised, and hundreds of millions in projected tax revenue.

Now here's what I find genuinely remarkable about that deal, and it's not the size. It's that OpenAI — the most important AI company on the planet, the one Nvidia just reported a $96 billion quarter against — had to build a gigawatt of shutdown capacity into its own facility just to get connected. The grid can't serve 3.2 gigawatts of continuous load, so the load itself has to learn how to bend. The data center isn't just a customer anymore. It's a battery.

The Two Readings — Real Demand, or a Rate Base in Search of a Story

So which is it? Is the demand real, or is the utility CEO selling you a story? The honest answer is both — and that's exactly why you need to be careful.

Reading one: the demand is real. The 17 gigawatts of contracted load is backed by $21 billion in collateral, with a credit target of A- or better. These aren't handshake deals; they're contracts with credit protections. Data center usage is up 55 percent on actual meters. The company joined something called the National Ratepayer Protection Pledge and is holding retail base rates stable in Georgia and Alabama until 2029. By every available metric, the AI buildout in the Southeast is happening.

Reading two: the forecast is not neutral. Here's the structural fact about regulated utilities that almost nobody in the AI conversation talks about: a utility makes money on its rate base. More approved infrastructure — more plants, more wires, more substations — means more capital invested, and under regulation, the allowed return is applied to that capital. Demand forecasts are not academic exercises at a utility. They're the justification for the next decade of construction. Every gigawatt of "prospective" demand in that 75-gigawatt pipeline is a reason to build, and building is how the business model works.

That's why the skeptics matter. Staff at the Georgia Public Service Commission have already said they doubt the utility's demand predictions. The nonprofit Science for Georgia looked at dozens of proposed data centers and found projected demand could reach 29 to 58 gigawatts — potentially outpacing what the state can actually supply. Georgia Power's own Integrated Resource Plan calls for 25 percent more demand by 2030 than its previous plan, with roughly 80 percent of that crunch driven by data centers lured by tax incentives. And we've all seen what happens to announced gigawatts in this cycle: Wood Mackenzie found 28 percent of the gigawatts in interconnection queues are real; the rest are phantoms. Utilities are not immune to that math.

The Secondary Bottleneck Nobody's Talking About — the Rate Base Incentive

Here's the part I want you to take home, because it's the hidden amplifier in this whole story: the rate base incentive is a structural bias toward optimism baked into the most important demand forecast in America.

Think about it. Every player in the AI infrastructure chain has a reason to talk the buildout up — Nvidia wants to sell chips, hyperscalers want to justify capex, data center developers want to raise money. But the utility is different. It doesn't need to talk anything up to sell anything. It needs regulators to approve construction, and the single best argument for approving construction is a demand forecast that says the load is coming. The incentive isn't to lie — it's to believe. And when the load doesn't materialize — when a 75-gigawatt pipeline turns out to be 10 gigawatts of real projects — the wires still got built. The rate base is still there. And the cost of that infrastructure gets recovered through rates, one way or another, because that's how the regulated model works.

This is the same pattern we've seen all over the AI buildout, from interconnection queues to GPU leases: announced capacity runs ahead of realized capacity, and the gap gets filled by someone else's balance sheet. With utilities, the gap gets filled by ratepayers. The Ratepayer Protection Pledge holds until 2029. Ask yourself what happens after 2029, when the first wave of 10-gigawatt generation approvals hits the rate case cycle and the load forecasts start getting haircut.

What This Means for Independent Hosting Providers

First — read the utility's Integrated Resource Plan like it's your power bill, because it is. In a state like Georgia, every IRP cycle is a preview of your future colo power rates. When the utility adds gigawatts of generation and transmission, you're not just reading about infrastructure — you're reading about the next rate case.

Second — flexibility just became a feature, and it's about to be a price differentiator. OpenAI's 1-gigawatt demand response commitment is the tell: hyperscalers are now trading curtailment for interconnection. Facilities that can shed load on a utility signal will get grid access faster and cheaper than facilities that can't. If you run colocation, start documenting your own demand-response capability today, because it's going to be in every interconnection negotiation for the next decade.

Third — lock your power contracts before the rate cases land. Southern's $2-billion-per-gigawatt buildout is real, and that capital gets recovered through rates. Multi-year fixed-power colo agreements signed before the next IRP cycle will look very different from the ones signed after it.

Fourth — don't price your business against announced gigawatts. The 75-gigawatt pipeline is not 75 gigawatts of power. The phantom-gigawatt pattern applies to utilities just as much as it applies to data center developers. Plan for the demand that's contracted and collateralized, not the demand that's in a slide deck.

Fifth — and this one's personal for me — watch the politics. I run servers in Atlanta. Georgia Power's rate cases are my power bill and yours. The "ratepayer protection" promises are real until 2029; what happens after that depends on whether the load forecasts hold. The best hedge for any hosting business in a utility's footprint is a diversified power strategy: multiple facilities, multiple states where possible, and a contract that doesn't put your business on the wrong side of a rate case.

The Bottom Line

So is AI demand slowing? The utility CEO says no, and he's closer to the physical reality than almost anyone — the meters are moving, the contracts are collateralized, and the buildout is real. But here's the thing about asking the guy who profits from building whether we need to build: he's going to say yes, and he's going to mean it, and he's still going to be wrong sometimes.

The real tell in this whole story isn't Womack's confidence. It's the gigawatt of shutdown capacity OpenAI had to build into its own facility. The most powerful AI company on earth couldn't get 3.2 gigawatts of continuous power from the biggest utility in the Southeast — so it agreed to be the battery. That's not a demand story. That's a constraint story wearing a demand story's clothes. The AI buildout is real, the grid is the bottleneck, and the load is going to have to learn how to bend. Ent? Plan for that.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Sources: Bloomberg Television (Aug 26, 2026), Southern Company Q2 2026 earnings call transcript (The Motley Fool), Economic Times, Georgia Public Broadcasting, The Current, Macon Newsroom, Atlanta Journal-Constitution.