Before the AI buildout, a large utility might see a handful of requests a year from customers looking to connect unusually large loads. By 2024, AEP Ohio had told regulators it had more than 50 data centers in its queue, representing over 30,000 megawatts of additional demand. American Electric Power said in May that it had 63 GW of contracted new load through 2030, backed by customer agreements, with another 190 GW of active projects in its interconnection queue.
But those two numbers tell very different stories. One represents customers who have made commitments. The other represents possibilities. That distinction matters because the AI infrastructure boom has created an enormous market for optionality. A developer considering several locations has every reason to investigate the power at each location before making a final commitment. Multiply that behavior across hundreds of developers, and suddenly a utility is trying to plan generation and transmission around a queue containing many more proposed megawatts than will ever be built.
The grid has a capacity problem. It increasingly has a decision problem as well. And from what I have seen across large capital programs, capital owners need to pay much more attention to the part of that equation they can control.
Indecision Now Has an Infrastructure Cost
A site for an AI data center is not simply a parcel of land. It is a decision about power, transmission, water, permitting, construction capacity, financing, customer demand, and increasingly community acceptance. Each variable changes while the project is being evaluated.
Yet many organizations still make these decisions through planning processes built for a slower era.
Site assumptions live in spreadsheets. Financial models are rebuilt for successive committees. Utility information is with one team, while construction assumptions are with another. When an input changes, leaders often need another round of analysis to understand what has changed in the business case. I have seen large capital organizations struggle to answer what ought to be a simple question: Of all the projects we are considering, which ones should we fund now?
In AI infrastructure, taking months to answer that question is no longer an administrative inconvenience. It can change the answer. A viable power arrangement disappears. A site gets taken. Construction pricing changes. A tariff moves. A competing project moves ahead in the queue. By the time the organization reaches a decision, the conditions that supported the original business case may already have changed.
The Market Is Starting to Charge for Uncertainty
Grid operators and regulators are beginning to respond to the same problem. In June, FERC ordered all six regional transmission organizations and independent system operators under its jurisdiction to justify or reform the way their tariffs handle large loads, citing the speed and scale of demand from facilities such as data centers and manufacturing plants.
AEP has already run the experiment. In July 2025, the Public Utilities Commission of Ohio approved a data center tariff requiring new loads above 25 MW to pay for at least 85% of contracted capacity regardless of what they actually use, on twelve-year terms with exit fees. Amazon, Google, Meta, Microsoft, and the Ohio Manufacturers’ Association all challenged it. The commission affirmed it. Wood Mackenzie now expects data center-specific tariffs in at least a dozen states by the end of this year. Optionality that used to be free is being repriced, and the customers being charged are the ones filing the requests.
PJM is moving toward the same principle on the supply side. Its new Expedited Interconnection Track for qualifying generation projects requires meaningful evidence that a project is ready to move: site control, support from the relevant siting authority, a $500,000 nonrefundable study deposit, and a $15,000-per-MW readiness deposit. PJM expects qualifying projects to reach a generation interconnection agreement in roughly ten months.
That tells us something important about where infrastructure markets are heading. A place in line is becoming less valuable than the ability to demonstrate readiness. The scarce resource is no longer simply power. It is a credible commitment.
Capital Owners Should Start Measuring Decision Latency
This is where I think leaders need to change the conversation inside their organizations. We carefully measure construction duration. We measure permitting schedules. We measure procurement lead times and energization dates. But very few organizations measure the time between identifying a viable capital opportunity and actually approving the money to pursue it. Some of our customers have been able to move viable capital opportunities from identification to funded approval in just a few weeks, significantly faster than the roughly ten-month timeline PJM is targeting for expedited interconnection.
In a market moving this quickly, that interval is no longer just an internal efficiency measure. Leaders need to understand where time is being lost and whether their approval process can keep pace with changing market conditions. If the external opportunity is changing in ten months and your organization requires twelve months to decide whether to pursue it, improving the construction schedule later will not recover the lost time. The first schedule that needs managing is the schedule of the decision itself.
Three Things Owners Can Change Now
The solution is not to weaken governance. Large capital commitments deserve scrutiny. It is to remove the unnecessary time between having the information and acting on it. Capital owners should do three things.
First, put the whole program on one shared record. Every site under consideration, every business case, and every commitment already made should be visible in one place rather than scattered across spreadsheets, inboxes, and individual teams. You cannot make a high-stakes decision quickly if the answer still lives in six different people’s heads.
Second, treat re-pricing a business case as a live capability, not a quarterly exercise. If a tariff, power price, construction cost, or other critical assumption changes, the organization should be able to reflect it in the business case within days, not wait for the next scheduled review.
Third, compare your internal approval timeline with the pace at which the market around you is changing. If power, site availability, pricing, or other critical conditions are shifting faster than your organization can reach a funding decision, the business case can weaken before the project ever moves forward.
This is the part of the AI infrastructure problem owners actually control. The United States will need more generation, transmission, and grid capacity. Those projects will take time. But adding power to the grid while allowing capital decisions to move through months of fragmented analysis solves only half the problem.
I have spent more than twenty years watching large organizations plan and deliver infrastructure. The organizations that perform best are rarely the ones that make reckless decisions fastest. They are the ones who have built enough institutional clarity to make good decisions without repeatedly rediscovering what they already know.
AI infrastructure is about to make that capability much more valuable. The companies that can reach a decision while an opportunity is still available will build. The others may discover that by the time they decide, someone else already has.
