At Yotta, not a word on slowing down the AI infrastructure build-out

You know what I didn’t hear at Yotta 2026 last week? “We’re in a bubble.”

The 6,500 or so in attendance instead talked of shortages of electrical gear so severe that a grey market for turbine blades and power station spare parts have sprung up, accessible only via Signal chat. Across three days, we talked about the wildly complex contracting arrangements now required to build an AI data center; the evolution of flexibility protocols; the tradeoffs inherent in building off-grid; and the elegant little solid state transformers that de-clutter increasingly messy power system architectures and lay the groundwork for a rapid move to 800 VDC.

Sure, the top-level numbers of this market have started to rationalize as projects get canceled or downsized. But the hundreds of gigawatts in utility load queues were always a fiction. All these data centers were never going to be built in three or four years, so the culling underway today is considered the mark of a maturing, healthy market, not one in retreat.

Nevertheless, Goldman Sachs, while acknowledging that community opposition has made it harder to build data centers, actually raised its forecast for 2026 U.S. data center capacity by five GW, to 64 GW, and trimmed its 2027 forecast to 90 GW. McKinsey has revised its own projections upwards as well, forecasting an average annual pace of 17 GW of data center additions through year-end 2030.

And more of this load is real than it was just a few years ago. Among the utilities we cover at Latitude Intelligence, 62% of signed load sits under approved data center tariffs;19 of the 24 approved tariff classes make customers pay at least 80% of contracted demand whether they use it or not. That’s firm commitment, not speculation.

This new demand should be encouraging to renewables and storage developers, and the ones I met with at Yotta all said they are deploying as fast as they can, in every region of the U.S. But things could still be better.

One of the largest developers told me they had over a hundred projects fully ready to go, awaiting only final permitting signatures from the federal government; all are being withheld for unclear reasons. The delays are particularly long for wind. The news of a permitting reform bill, released last week, came as welcome news for everyone (though its benefits will also accrue to gas pipeline developers).

But will renewables and storage and “unlocked” clean capacity actually dominate the AI data center power buildout? After three days of talking, and a look at the data that the Latitude Intelligence team has collected, I’m beginning to doubt it.

The idea that gas is winning out was unavoidable throughout the week. Every question posed to developers on a panel or during a cocktail hour led to the same response: a shrugging admission that we’re back in the realm of “all of the above” — and that a lot of the above will be gas.

And no one said they expect any real softening in the market. Bankers said all the gas turbine OEMs are supply constrained well past 2030, and Jefferies, in a note this week, said “hyperscalers are increasingly prioritizing capacity availability and time-to-power over [levelized cost of electricity], supporting investment across the power equipment value chain.” Read: gas will get priority over any other generation source if it’s the fastest.

To me, these conversations (depressing as they were) were a cry for batteries. Storage is the answer to firming renewable power, a predictable source of flexibility, and the solution for mitigating congestion on transmission grids and unlocking any stranded capacity. Batteries on the distribution grid reshape load while adding capacity that can either be accredited directly to a data center or at least help buffer their impacts across the larger system.

Meanwhile, in the data center itself, fast, short duration batteries can obviate the need for legacy UPS and replace diesel backup with clean capacity. And if long-duration storage scales as it should, those batteries could replace even gas peakers.

Learn how utilities can adopt modern energy forecasting tools to help bridge planning, flexibility programs, and real-time operations.

Looking outside the data center market, utilities are already deploying renewables and storage at pace. Latitude Intelligence found electric utility plans still need about 104 GW of new supply through 2035, and current IRPs and procurement plans meet 70% of that demand with solar (41 GW), storage (26 GW), and wind (7 GW). New gas generation stands at roughly 20 GW.

In other words: The energy transition is still happening, it’s just stalled at the data center.

There remains a real gold rush atmosphere in the market — I saw at least a dozen cowboy hats last week, maybe more — which I suspect has led many in the clean energy sector to think the AI boom’s early emphasis on gas was just the folly of speculators. But I didn’t see that on the ground.

As the real scale of AI labs and hyperscaler ambitions sets in, I saw the most serious characters in the industry say with confidence they will be building with gas, building gigawatt-scale off-grid power campuses, and including renewables where they can. Flexibility will come when it’s ready, but no utility really wants to offer a service that will consistently be curtailed, nor would any data center operator choose one if it didn’t have to. They want growth and speed, everywhere they can buy it.