.png)
GRID CAPACITY · LOAD GROWTH & COORDINATION
IS AI REALLY GOING TO BREAK THE GRID?
By David Swank, CEO, i3 Power & Energy
The story that AI is going to break the American grid has become a fixture in industry commentary. It shows up in earnings calls, congressional testimony, keynote panels, and cover features. There is a real load growth story underneath it. The numbers are large, the timelines are compressed, and the pressure on utilities and system operators is genuine. What the doom narrative misses is not the scale of the challenge. It is where the actual failure risk lives.
Grids do not break the way bridges break. They fail through coordination lags between the parties responsible for developing, connecting, and operating them. When those parties work in isolation, small mismatches compound into large problems. When they work together, even substantial load growth gets absorbed with less friction than the current narrative suggests is possible. The AI grid conversation is one of the clearest examples of this dynamic in the industry today.
.png)
WHAT THE LOAD GROWTH NUMBER ACTUALLY TELL US
The load growth projections that dominate current commentary are directionally correct. Data center demand is real, its concentration is unprecedented, and the pace of announcements has outstripped the pace at which most regional grids were planning to evolve. Taken on its own, that would be an alarming picture. But the numbers themselves are not the failure point. The failure point is the sequence that produces them: a developer decides, then a utility responds, then an ISO reacts, and each step introduces a lag that the parties then blame on the next party in the chain.
A grid capable of absorbing this load growth is not fundamentally a different grid. It is the same grid, operated by the same parties, using the same infrastructure categories, but with a coordination discipline that the current process does not require anyone to bring.
THE VIEW FROM THE UTILITY SEAT
RB Sloan and I sat in that seat when new load arrived faster than the system was built to absorb. What I can tell you from that experience is that the problem was rarely that we could not build. Building takes time, but it is manageable time when the parties are working from a shared understanding of what is coming. What was not manageable was discovering, halfway through interconnection studies, that the load profile had changed, the site had shifted, or the developer's timeline had accelerated. Each of those surprises reset the clock on our work, and the surprises came because we were participating too late in the developer's process to shape it, and the developer was participating too late in ours to be shaped by it.
This is not a criticism of developers or of utilities. Both parties are behaving rationally inside a process that was not designed for the pace or scale of what is now being asked of it.
.png)
“The AI grid problem is not primarily a capacity problem. It is a coordination problem wearing a capacity costume.”
.png)
WHAT COORDINATED DEVELOPMENT ACTUALLY LOOKS LIKE
A coordinated approach brings developer, utility, and system operator into shared conversation from the earliest planning phases. Load forecasts are surfaced when they are still assumptions, not after they have become commitments. Generation options are considered alongside load, not after siting has been finalized. Transmission implications are worked through in advance rather than discovered in queue studies. Interconnection design is a joint product of the parties who will build, operate, and connect the asset.
None of this requires new regulatory authority, new financing structures, or new technology. It requires a different theory of who should be in the room, when they should be in the room, and what decisions they should have a hand in shaping. The parties are all willing. The industry structure has just not made this the default.
WHY THIS MATTER OVER THE NEXT FIVE YEARS
The wave of load driven by AI, industrial electrification, and continued data center growth is not slowing. Firms and utilities that can convene the parties earlier will move faster and produce more bankable projects. Firms that do not will remain stuck in queues that, in many cases, did not need to exist. The grid is not going to break. The parties that fail to coordinate around it may find that their projects do.
The AI-will-break-the-grid narrative is a useful signal. It tells us the industry knows something is straining. What it gets wrong is the diagnosis. What is straining is not the physical infrastructure. It is the model of how parties work with each other to build it. Change that, and the load growth story becomes an opportunity rather than a crisis.