Power Reality Check: Why Bridge Solutions Are Becoming Permanent

Utilities promise 2-7 year delays for data center power. Why temporary solutions are becoming permanent and hyperscalers are building their own grids.

The AI infrastructure boom promised to transform the grid. Instead, it's revealing how far behind utilities have fallen—and why temporary power is becoming the new normal.

When OpenAI announced plans for 100 gigawatts of computing capacity, the assumption was simple: utilities would scale up to meet demand. Data center operators would secure power commitments, break ground, and connect to the grid within a reasonable timeframe.

That's not what's happening.

Instead, utilities are telling data center operators to expect two to seven-year delays. Mobile power generation companies are keeping gigawatts of equipment in stock for "temporary" deployments that increasingly look permanent. And hyperscale operators are investing in modular nuclear reactors because waiting for the grid isn't an option.

At the infra/STRUCTURE Summit in Las Vegas this week, hosted by Structure Research to bring together power providers, data center operators, and investors, the message was clear: the power infrastructure challenge is worse than anyone anticipated, and bridge solutions aren't bridging to anything—they're becoming the foundation.

The Utility Time Bomb

David Dorman, Director of Commercial Operations at APR Energy, has spent his career in power generation. He knows what it takes to bring a 1.5 gigawatt combined cycle facility online. The answer: years of development and years of construction.

When asked if the industry is being realistic about timelines for 100 gigawatts of AI infrastructure, his response was measured but clear.

"I think the timelines are extremely difficult," Dorman said. "That's going to be very difficult to achieve within a decade. You have supply chain constraints and labor constraints. There are a lot of things that need to change."

The problem isn't just building new generation capacity. It's everything around it: transmission lines, substations, transformers, distribution infrastructure. And utilities, which experienced flat electricity demand for a decade before AI arrived, aren't equipped to handle this surge.

"On average, I've been hearing anywhere from two to seven years' delays in getting infrastructure to these facilities," Dorman explained during a briefing at the Structure Research conference.

Those aren't outlier cases. That's the new normal.

What "Rapid Deployment" Actually Means

APR Energy specializes in rapid power deployment. They keep close to a gigawatt of power generation equipment in stock—mobile turbines, transformers, switchgear, fuel systems—ready to deploy on short notice.

Their typical installation timeline is 30 to 90 days.

But that's only for the equipment itself. The infrastructure required before APR Energy can even begin installation—roads, foundations, gas pipelines, site preparation—takes considerably longer.

"A gigawatt of power would probably take tranches of 250 megawatts within maybe 60 days," Dorman said. "But again, it's just that outside infrastructure."

And here's the reality check: even with all of APR Energy's fleet deployed, they can provide about a gigawatt of power. OpenAI alone is talking about 100 gigawatts. Meta is spending $72 billion this year on AI infrastructure. The scale of demand dwarfs what mobile solutions can provide.

Yet companies like APR Energy are busier than ever, because data center operators can't wait years for utility connections.

The Bridge That Never Ends

APR Energy's business model traditionally centered on temporary deployments—providing power for countries experiencing demand surges, supporting disaster relief efforts, or filling gaps while permanent infrastructure came online.

Dorman estimates that temporary work represents about 90% of their projects. But increasingly, those temporary installations are becoming something else.

"I would say probably on the latter half, that 10% has been power generation facilities built in a permanent fashion rather than with our modular equipment," Dorman explained. "With a permanent solution, we would install a different solution than we would use for a bridge."

In other words, when it becomes clear the grid isn't coming, temporary equipment gets replaced with permanent installations.

This shift reveals something fundamental: the industry initially viewed mobile power as a stopgap measure. Now it's becoming Plan A.

Bill Thomson, VP of Product Management at DC BLOX, sees this from the operator side. His company works across the Southeast with multiple utilities, and the infrastructure gap is widening.

"I do foresee that a lot of future data centers will probably have to start up with some kind of bridge solution," Thomson said. "Whether it be mobile generators or other technology, potentially land-mounted turbines. Until that infrastructure and utilities are able to bring power to them, a behind-the-meter solution is going to happen."

Behind-the-meter means power generation on-site, independent of the grid. What was supposed to be temporary is becoming standard operating procedure.

The Utility Culture Clash

Thomson describes the situation as a fundamental mismatch between two industries operating on completely different timelines and risk profiles.

"This is the clash of cultures," he said. "Most of us in the data center industry are very aggressive risk takers, flexible, and adapting to market opportunities. Utilities, not so much. Very conservative, different business operating models."

Georgia Power exemplifies the challenge. For at least a decade before 2022, they experienced flat electricity demand. Their business model is optimized for efficiency and avoids capital expenditures. Then AI arrived.

"All of a sudden, data centers and AI came and literally changed everything—20 to 30 times the demand," Thomson said.

Utilities are adapting, but not fast enough. Some utilities now ask data center operators to pay for substations and transmission lines to their sites. Data center companies are willing to do that. But the utility policies create other problems.

For projects over 20 megawatts, Georgia Power requires commitments to use 80% of that power by a certain date. For colocation providers who need to sign customers before committing to power usage, achieving this is impossible.

"You're going to box out the majority of the market—80% of the market—by these kinds of policies," Thomson explained. "Colocation third-party development of data centers is five times what self-builds are."

Duke Energy in Florida recently asked regulators for an entirely new rate framework because the company doesn't currently serve large-scale data centers. The request acknowledges what everyone in the industry already knows: utilities aren't prepared for this.

Supply Chain Reality

Even if utilities could move faster, supply constraints create another bottleneck.

Thomson noted that turbine manufacturers are "blowing their doors out" with orders, and delivery timelines stretch years into the future.

"I can't get a turbine in the next couple of years," Thomson said. "My whole value proposition—that's tough."

The same applies to transformers, switchgear, and other critical electrical infrastructure. Global demand for these components has exploded, and manufacturing capacity can't scale overnight.

Dorman emphasizes that money can help solve supply chain constraints—companies can pay premiums to jump the queue or incentivize expanded production. But there are physical limits to how fast manufacturing can scale.

"More money can solve the supply chain constraint," Dorman said. "More money can solve labor constraints as well. But plans need to be in place sooner than later, because it is going to be very difficult to get the supply chain and labor forces ready for a surge of 100 gigawatts."

The Nuclear Bet

Hyperscale operators aren't waiting to find out if utilities and supply chains can catch up. They're making their own plans.

Thomson mentioned that hyperscalers have invested heavily in modular nuclear technology that can be deployed on-site in standardized containers.

"These modular reactors, 75 megawatts—they're just in these containers," Thomson said. He noted that recent executive orders reducing regulatory approval timelines could accelerate deployment from eight to ten years to a more manageable timeframe.

This represents a fundamental shift in how data center operators think about power. Instead of connecting to the grid, they're planning to generate power independently.

Amazon's Project Rainier in Indiana, Microsoft's investments in nuclear energy, and Meta's exploration of alternative power sources all point in the same direction: hyperscalers are building their own energy infrastructure because they can't rely on utilities to provide it fast enough.

The Cost of Delay

The infrastructure gap isn't just slowing down AI deployment. It's creating real economic consequences.

Communities that reject data center projects because of grid concerns are turning away billions in investment. States with proactive energy policies are capturing that capital instead.

Dorman pointed out that Nevada's competitive energy rates make locations like Las Vegas attractive for data center development. Georgia's willingness to allow data centers to pay for their own transmission infrastructure keeps projects moving forward there.

But regulatory uncertainty and utility delays create risk that some companies aren't willing to accept.

"The demand is too high, the economic value is too high to wait for them," Thomson said of utilities. "The more aggressive and agile ones will play a role, but it's going to bypass many of them."

What Happens in Three Years

Structure Research's infra/STRUCTURE Summit emphasized that the next few years will be particularly challenging as the industry transitions to new power solutions.

"The next few years are going to be probably more challenging because this transition continues until some of those more nuclear or other alternative energy source things come into play," Thomson said. "We're probably going to stumble a little bit in meeting the demand."

That stumble has consequences. Data center operators with power solutions move forward. Those waiting for utility connections fall behind. Customers seeking AI infrastructure capacity will go where power is available, regardless of whether it comes from the grid or from on-site generation.

Dorman expects that permanent behind-the-meter solutions will become increasingly common.

"In some cases, the cost to develop that infrastructure may be too great, and utilities may only be able to provide partial power," he said. "A behind-the-meter solution, whether it be mobile or permanent, may still need to be on those specific sites."

The Long-Term Picture

If temporary becomes permanent and hyperscalers build their own power infrastructure, what happens to utilities?

Thomson sees two paths. Aggressive utilities that embrace the opportunity and adapt their business models will remain relevant. Conservative utilities that resist change will get bypassed.

"Maybe we need to consider private investment that's willing to take on that risk, which is what helps generate that energy infrastructure," Thomson said. "If utilities can't figure out how to get aligned with that, then it's just going to bypass them."

The data center industry isn't going to wait. The economic stakes are too high, and competitive pressure is too intense.

Dorman frames it practically: "Everything can always be done, especially if you throw more money at it. More money can solve the supply chain constraint. More money can solve labor constraints."

But there's one thing money can't solve: time. Building a gigawatt of generation capacity takes years. Getting utility approvals takes years. Expanding transmission infrastructure takes years.

The industry needs that capacity now. So bridge solutions stop being temporary and start being permanent. Mobile turbines get replaced with land-mounted installations. On-site generation becomes standard. And the vision of data centers seamlessly connecting to an ever-expanding grid fades into something more practical: data centers that generate their own power because waiting isn't an option.

That's the power reality check. The bridge isn't leading anywhere. It's becoming the road itself.