Operating Under Load Growth: Why the Real Constraint Is Institutional, Not Technical

The U.S. grid has managed load growth before. What it has not managed is load growth that is clustered, speculative in timing, and large enough to change planning assumptions within a single planning cycle.

At Operating Under Load Growth: Forecasting, Interconnection, and Grid Capacity, the panel converged on a structural diagnosis: the engineering toolkit for operating power systems is robust, but the institutional toolkit—forecasting credibility, interconnection process design, cost allocation, and incentive alignment—is strained by acceleration.

The defining shift is not simply “more demand.” It is demand arriving in ways the system was not designed to absorb: hyperscale data centers, advanced manufacturing, and other concentrated loads that can change location, ramp quickly, and create upgrade requirements that outlive the commercial projects that triggered them.

---

1) Forecasting Is No Longer About Load Size — It’s About Load Credibility

Sean Kelly (CEO, Amperon) described how forecasting is evolving from macro planning inputs into an operational capability. Amperon forecasts electricity demand at the meter level and in aggregate—at horizons from next-hour to multi-year—across what he described as roughly 40 million meters under management.

The technical point is straightforward: we can forecast weather-driven supply and demand with growing sophistication. The new operational problem is distinguishing credible load growth from aspirational load growth.

Kelly gave a concrete example from early work forecasting a 25 MW casino load on the Las Vegas Strip—pairing that demand profile with renewable offtake, and using coincident peak programs to reduce load when the system is most constrained. Coincident peak—an incentive structure used by utilities/ISOs to reduce demand during the highest system peak intervals—has become a central lever for large customers to behave like flexible grid participants.

The stakes of forecast error are no longer marginal. NERC’s latest long-term assessment emphasizes that demand growth uncertainty and lags in resource additions are contributing to heightened resource adequacy concerns across multiple regions (NERC, Jan 2026). Separately, the DOE/LBNL data center assessment highlights how wide the forecast band can be: data centers consumed about 4.4% of U.S. electricity in 2023 and could rise to a materially larger share by 2028, depending on growth trajectories (DOE/LBNL, Dec 2024). The message is not one number—it is that the planning distribution is widening.

Implication: Planning must shift from deterministic “load forecasts” to probabilistic load credibility frameworks—including customer vetting, phased energization assumptions, and explicit confidence bands that influence reserve and transmission decisions.

---

2) Demand Flexibility Works — But It Is Not Symmetric Across Large Loads

Kelly emphasized that “large loads are surprisingly flexible,” but the degree of flexibility varies dramatically by business model. Some loads can only shave marginally; others can drop near zero. He cited Bitcoin mining as highly flexible, and discussed how demand response participation has broadened from the early 2000s to become common for many megawatt-scale customers today.

He also cited a stress-event proof point: during Winter Storm Fern, PJM “toggled down” large-load demand by roughly 10 GW more than expected, while ERCOT was closer to 7–8 GW. Whatever the exact composition, the point is operationally meaningful: flexible load can materially change the net load that operators must serve during tight conditions.

But the panel also flagged the core policy tension: who flexes, when, and under what rules.

Joshua Macey (Professor of Law, Yale Law School) raised a crucial issue that goes beyond voluntary programs: at some point, curtailment becomes mandatory. When large loads ramp to hundreds or thousands of megawatts, curtailment design becomes a question of priority rights, technical limits, and equipment impacts (e.g., transformer cycling concerns that data centers have raised).

Implication: Demand flexibility cannot be treated as a generic resource. Operators need tiered curtailment frameworks that reflect differing economic value, technical ramp constraints, and grid impacts—rather than assuming all large loads will respond similarly to price or requests.

---

3) Curtailment Is Becoming a Market Power Question, Not Just a Reliability Tool

Brenden Millstein (Cofounder and President, Centinel) offered a sharp economic frame that changes how curtailment politics should be understood. For a typical commercial building, utility costs might be ~$3 per square foot, rent ~$30, and salaries ~$300. For a data center, he argued, the revenue density could be orders of magnitude higher—potentially ~$3,000 to $30,000 per square foot.

His conclusion was blunt: hyperscalers will not shut down mission-critical compute “because the grid asks nicely.” They will either:

• install large batteries (his example: “$30 million of batteries”), or

• lobby for rules that protect their continuity.

He predicted that curtailment burden may shift toward lower-margin users who cannot win that political fight.

This is not merely cynicism; it is a warning about incentive compatibility. If reliability policy depends on voluntary curtailment but the highest-value loads opt out through technical self-supply or regulatory leverage, the remaining system bears the adjustment.

Macey’s later discussion of secondary markets for curtailment rights makes the underlying point explicit: if curtailment has value, property rights and markets will emerge—either informally through influence or formally through market design.

Implication: Grid operators and regulators will need to decide whether to formalize curtailment rights and secondary markets (so allocation is transparent), or allow allocation to be determined implicitly through political and economic power.

---

4) Interconnection Reform Is Necessary — But Queue Discipline Alone Won’t Solve Scarcity

Macey argued that interconnection backlogs are not just a procedural inconvenience; they are now a binding constraint on reliability planning and investment timing.

He explained that under current frameworks, generator interconnection can proceed under different service models—NRIS (Network Resource Interconnection Service) and ERIS (Energy Resource Interconnection Service). NRIS generally confers stronger deliverability and, in capacity market regions, eligibility for capacity payments; ERIS typically implies greater curtailment risk and fewer deliverability rights.

His high-level thesis: we should let resources connect faster if they are willing to accept risk, similar to “connect and manage” approaches—an idea popularized in the interconnection reform debate.

He then went further: if scarcity is real, queue position should not be treated as an administrative artifact. He described the concept of auctioning interconnection positions so developers “put their money where their mouth is,” and suggested that either:

• auctions, or

• a subscription model where upgrades are built prospectively and developers pay standardized fees could better align infrastructure buildout with where value is highest.

This aligns with the broader direction of interconnection reform in the U.S., which has already moved toward cluster studies and stricter readiness requirements (FERC Order 2023 and related compliance actions). (FERC, 2023–2024)

Implication: Under structural load growth, interconnection needs mechanisms that allocate scarce upgrade capacity based on value and readiness, not solely on timestamp—and must be coordinated with capacity market timing so markets can actually signal new entry.

---

5) Reliability Risk Is Becoming Dynamic, Not Just Resource Adequacy

Antonio J. Conejo (Professor, The Ohio State University) emphasized a technical point with direct regulatory consequences: the biggest new risk is not simply more load—it is the speed and geography of load change, interacting with high renewable penetration.

He described a dynamic stability problem: large loads can shift dramatically between locations in response to small price differences, creating a “combinative effect” where load rises sharply in one area while falling elsewhere. That is not the classical planning problem of steady growth; it is a dynamic system stressor that can contribute to instability and—if unmanaged—blackouts.

Conejo connected this to real-world events, referencing instability episodes such as the Iberian Peninsula system event and noting the growing importance of reactive power and nonlinear behavior in modern grids with power electronics.

Conejo’s core recommendation was institutional: regulation must require “good behavior” from large loads—not because markets will naturally produce it, but because stability is a public constraint. He expressed confidence that engineering solutions exist, but emphasized they must be implemented through enforceable rules and safeguards.

Implication: As load becomes more mobile and renewables more dominant, reliability policy must incorporate dynamic stability requirements—including technical standards for both large loads and inverter-based resources, not just static reserve margin math.

---

The New Scoreboard for Operating Under Load Growth

The panel’s most important takeaway is that the next phase of grid planning is institutional:

• Forecasting must measure credibility, not just magnitude

• Load flexibility must be operationalized with clear rights and obligations

• Interconnection must allocate scarce upgrade capacity based on value, readiness, and risk acceptance

• Resource adequacy must pay for performance, not theoretical availability

• Utility incentives must not default to capital expansion when lower-cost flexibility exists

The grid is not running out of ideas.

It is running out of time.

---

Sources

• North American Electric Reliability Corporation (NERC) (Jan 2026). 2025 Long-Term Reliability Assessment.

• U.S. Department of Energy / Lawrence Berkeley National Laboratory (Dec 2024). 2024 Report on U.S. Data Center Energy Use.

Federal Energy Regulatory Commission (FERC) (2023–2024). Order No. 2023 and related interconnection queue reform guidance.