Core Strategic Assessment
The AI buildout is running into a harder constraint than model hype: power that can be delivered to the right site, at the right time, under rules that decide who pays for the grid upgrades.
A data center is not just another commercial customer when it asks for hundreds of megawatts of firm power. In grid language, it can become a "large load" - a very large electricity user, generally above 20 megawatts in FERC's large-load docket. The strategic issue is not only total electricity demand. It is the queue to connect, the transmission upgrades needed, the reliability rules for unusual computing loads, and the political fight over whether ordinary ratepayers or data-center customers pay.
Key Actor Objectives
The main actors are not debating whether AI needs power. They are deciding how that power is approved, priced, and protected.
- US - Federal regulators: FERC is weighing reforms for large-load interconnection so big new loads can connect in a timely, orderly, reliable, and non-discriminatory way.
- US - Reliability authorities: NERC is pushing utilities and grid operators to treat computational loads as a technical reliability issue, not only a demand forecast.
- US - Grid operators: PJM is trying to manage a supply-demand imbalance largely driven by data-center growth through capacity-market and interconnection proposals.
- US - State consumer advocates: Maryland's Office of People's Counsel is challenging whether data-center-driven transmission costs are being pushed onto customers who did not cause the load growth.
- Data-center operators: Hyperscalers and developers need power certainty fast enough to support AI deployment schedules.
Strategic Dynamics
The bottleneck is shifting from "can the site be built?" to "can the site be energized?" That changes the map of AI infrastructure.
If a region can offer land and fiber but cannot offer interconnection certainty, it becomes less attractive. If another region can provide firm power, transmission capacity, clearer tariffs, and faster approvals, it gains leverage over where compute gets placed. That does not mean AI growth stops. It means growth gets rerouted, delayed, repriced, or tied to new power commitments.
Cost allocation is the most politically sensitive part. Grid upgrades can benefit a private data-center customer, but the costs can be spread across a wider customer base. That creates a direct fight between speed-to-power and ratepayer protection.
The federal docket is only one layer. Developers, utilities, and states are also using bilateral utility arrangements, co-located generation, behind-the-meter structures, curtailment agreements, and state public-utility commission proceedings to move around slow grid queues. That does not weaken the interconnection thesis; it shows how severe the speed-to-power problem has become.
Evidence and Indicators
The evidence is strongest because it comes from regulators, reliability authorities, grid operators, state advocates, government data, and company filings.
- FERC action clock: FERC said it will act by June 2026 on the RM26-4 large-load docket, which includes data-center demand and asks whether large loads should pay the full cost of grid upgrades needed for interconnection. That action may be a regulatory milestone rather than a final rule.
- Different interconnection problem: FERC Order 2023 restructured generator interconnection queues. RM26-4 is different: it concerns how major electricity users such as AI data centers connect to, draw from, and pay for impacts on the transmission system.
- Demand pressure: EIA says data centers are driving U.S. electricity-demand growth and identifies PJM and ERCOT as the fastest data-center demand-growth regions through 2027.
- Reliability risk: NERC issued a Level 3 alert for computational loads, including AI training and data-center uses, after finding that existing processes were not sufficient for these load risks. The alert requires acknowledgment and reporting, but it is not itself a mandatory Reliability Standard.
- Regional implementation: PJM's 2025/2026 capacity auction cleared at a record-high $269.92/MW-day RTO-wide, reflecting tighter reserves as load grew and retirements outpaced new-resource interconnection. PJM has also filed large-load reforms tied to data-center-driven imbalance, including an expedited generation-interconnection track and a capacity-market price collar.
- Ratepayer dispute: Maryland's consumer advocate filed a FERC complaint alleging that PJM rules assign about $2 billion of data-center-driven transmission capital costs to Maryland customers. That allegation is not a final ruling, but it is a concrete sign that the cost fight is already live.
Market and Sector Implications
This is not an investment recommendation. The market signal is that AI infrastructure exposure is no longer limited to chips, cloud platforms, or data-center landlords.
- Verified direct exposure: Quanta Services (PWR) is exposed to the grid-build side: transmission, substations, distribution, grid modernization, data-center infrastructure, and interconnection services. Its 2026 filing directly links strong demand to data centers, reshoring, electrification, and interconnection work.
- Verified direct exposure: Eaton (ETN) is exposed to electrical equipment and power-management demand. Its Q1 2026 release reported Electrical Americas orders up 42%, driven by data-center momentum, and Electrical-sector backlog up 48%.
- Sector pressure: Developers, utilities, and equipment suppliers will not all benefit equally. The strongest exposure should come from companies that can turn grid access, electrical equipment, permitting, and reliability compliance into delivered capacity. Projects that look attractive on compute economics can still stall if they cannot secure power on acceptable terms.
Summary: The Strategic Chessboard
| Issue | Actor Objective | Leverage Used | Likely Dynamic |
|---|---|---|---|
| Large-load interconnection | Connect major power users without destabilizing the grid | FERC docket, tariffs, upgrade rules | Speed-to-power becomes a strategic AI variable |
| Reliability | Model and manage computational-load behavior | NERC alerts, standards work, reporting | AI loads get treated as a technical grid-risk category |
| Regional capacity | Serve fast-growing data-center demand | PJM reforms, capacity auctions, interconnection rules | Power-rich regions gain siting leverage |
| Cost allocation | Decide who pays for transmission upgrades | FERC complaints, state advocacy, tariffs | Ratepayer protection becomes a constraint on AI buildout |
| Public-company exposure | Identify direct infrastructure channels | Filings, orders, backlog, project demand | Grid and electrical suppliers matter alongside chips |
Bottom Line
AI infrastructure is becoming a power-delivery race. Chips still matter, but the next constraint is increasingly whether a data center can get connected, supplied, studied, protected, and paid for under rules that regulators and grid operators can defend. The chokepoint is not just electricity generation. It is interconnection.