“Speed to power” is a defining phrase of the data center era. The conversation used to center on one question: How do we build power fast enough to meet AI demand?
But that’s changing as the industry increasingly seeks ways to mine the system for spare capacity, flexibility and efficiency.
This idea came out during a panel discussion I moderated on July 10 at The Great Transformation conference in Bend, Oregon.
The grid’s hidden capacity
Utilities have traditionally planned for peak electric demand. That conservative approach has produced a grid that’s reliable, but expensive and underused.
“We have an eight-lane highway, but we only use two of the lanes. We have to use the other lanes when it gets to be rush hour,” said Bill Messner, director of customer solutions for Portland General Electric (PGE).
At PGE, the rush hour is short. Messner said the utility worries about roughly 200 hours a year. During the other hours, that capacity sits stranded.
So rather than requiring every customer wait for new substations and transmission lines, PGE is using creative contracting to tap the stranded capacity, allowing customers to connect sooner in exchange for operational flexibility during those relatively few peak periods.
“If you want to be my customer, what flexibility can you bring me?” is how Messner described the conversation.
The approach is already producing deals. For example, a data center is building a battery that PGE will operate upstream of the customer’s meter, giving the utility another source of system flexibility.
The strategy represents a different way of thinking about speed, power, and customers. Flexibility becomes an infrastructure resource. And instead of being viewed solely as enormous new loads, customers begin contributing assets that help utilities make better use of existing infrastructure.
The technology driving the demand surge could also help it
Jacob Sandry, co-founder and CEO of Euclid Power, offered a more humble way to improve speed to power: reduce the paperwork.
A single renewable energy project can involve 500 documents and contracts, by Sandry’s count. Much of that information exists as unstructured text that must be reviewed repeatedly by developers, investors, attorneys and lenders. The gaps between what the documents say, what the financial model assumes and what gets communicated to the field are big enough to delay projects — or kill them.
AI can organize and analyze that information in a fraction of the time. “We are seeing our ability at Euclid to do things that used to take months down to days,” Sandry said.
That’s significant because many project delays aren’t only about interconnection queues. They stem from the complexity of developing and financing billion-dollar infrastructure projects.
Sandry also offered a window into why data center developers behave so differently from the renewable developers his company serves. A solar project is underwritten on a 35-year return, on technology that doesn’t improve once it’s in the ground. Data centers are underwritten on roughly a three-year horizon because Nvidia ships a better GPU every six months.
That math helps explain what Sandry described as the first era of the buildout. He recalled one hyperscaler cutting an enormous check with minimal due diligence for land with access to power. The company was driven by the urgency to get GPUs running. Sandry sees that giving way to the more rigorous project scrutiny that the renewable energy industry has already learned.
From backup power to “backup power plus”
Perhaps the most pointed exchange of the panel involved microgrids.
For years, the primary value proposition for microgrids has been resilience. Hospitals, military bases and critical facilities invested because they needed electricity when the grid failed.
Now, data center demand creates a different economic rationale. Onsite generation and storage as a way to reach commercial operation before utility infrastructure can be expanded. Microgrids become bridge power — temporary infrastructure that allows facilities to begin generating revenue sooner.
Not everyone at the conference was impressed. A speaker the previous day had called microgrids “an inefficient solution to a solvable problem” — a design failure, essentially.
I asked Cameron Brooks, executive director of Think Microgrid, an education and advocacy non-profit, to respond. “You might as well come over and kick my dog,” he joked.
But he took the criticism seriously — and turned it around. The real design constraint, he argued, belongs to the grid itself.
“There’s certain things that a microgrid can do that grid-delivered power is simply never going to be able to do.” No amount of undergrounding or hardening eliminates the need for onsite power, which is why backup generation already sits all over the system.
That’s what makes treating these assets as temporary such a waste.
“If we’re going to go through all the headache of putting some kind of resource onsite,” Brooks argued, “why would you want that to be there only to turn on and off” during outages?
Instead, Brooks suggested thinking about “backup power plus.” A battery installed to accelerate interconnection can later provide grid flexibility. A microgrid built for resilience can participate in utility programs, reduce peak demand and strengthen local reliability. Today’s bridge asset becomes tomorrow’s permanent grid asset.
An audience member pushed the logic further: if bridge power works this well, does it eventually disintermediate the utility?
“That’s not my nightmare currently, but thank you,” Messner, who works for a utility, deadpanned. But then he answered more seriously that a grid of disconnected microgrid islands would be a failure, not a triumph. Brooks agreed. Distributed resources are “the capillaries of the system,” he said. “We’re not going to get rid of the trunk line or the aorta.”
Could AI erase the peak?
Brooks summed up the panel’s two threads — AI and flexibility — with a challenge.
The industry spends its AI efforts on incremental applications, he noted: better customer segmentation and better program targeting. He suggested opening up the imagination. “It could well be that we use AI combined with storage embedded throughout the system to completely erase peak power.”
His analogy: long-distance phone calls once cost real money. Then the underlying technology changed, and an entire pricing model — an entire way of thinking about scarcity — simply vanished. Nobody thinks twice now about calling France and talking for an hour.
If AI and distributed storage do the same to the peak, the eight-lane highway problem doesn’t just get managed. It disappears.
Speed to power definition changes
So is “speed to power” becoming an outdated description of what’s actually happening? Not the speed part. Urgency still permeates industry discussion in light of demand forecasts. But with interconnection lines long for new power plants and equipment on backorder, it’s become a race to mine the system for more flexibility, efficiency and capability.
