Orientation

Key Takeaways#

A summary of the report's principal findings, organized as answers to three questions, with a confidence tag on each point so settled facts stand apart from informed expectations. Confidence key:

 SETTLED  adopted rule or occurred event

 LIKELY  proposal on a clear path / strong evidence

 EMERGING  proposal or position taken but not yet resolved

 SPECULATIVE  genuine forecast — direction supported, date or magnitude not

1. The three risks that dominate the record#

• Load behaviour that the system was not built to absorb. Gigawatt-scale computational load disconnects itself in under a cycle during a normally cleared transmission fault, converting a disturbance the grid should absorb into a large, uncommanded loss of demand. The record is continuous rather than anecdotal: 26 ERCOT events above 100 MW between January 2023 and September 2025, and about 1,500 MW across 60 delivery points in northern Virginia in July 2024 during faults cleared correctly in 42 to 66 milliseconds — the archetype, because nothing malfunctioned. A larger west Texas event in December 2022 removed about 1,600 MW and took twelve and a half minutes to recover, though its clearing was abnormal and its load mixed (Section 4). ERCOT studies indicate that frequency excursions become significant near about 2,600 MW of instantaneous loss, and has already reduced operating limits on some interfaces because the possibility has itself become an operating contingency. Ramping is ungoverned everywhere: training clusters swing hundreds of megawatts in seconds, and no North American jurisdiction presently imposes a generally applicable ramp-rate limit on an individual large computational load, though individual service agreements may contain one.  SETTLED 
• Cost allocation that shifts large-load costs onto existing ratepayers. PJM's independent market monitor estimates roughly 40–45% of recent capacity-auction cost as attributable to data-center load, much of it for facilities not yet built. Capacity-driven increases already visible on household bills run approximately $10–20 a month in affected PJM zones. Because capacity prices clear against forecast load, the system socializes those costs today on the basis of projects that may never exist — which makes queue integrity a precondition for solving cost allocation rather than a parallel effort, and one of the principal areas of policy debate in this system expansion.  SETTLED 
• A timing inversion between load arrival and infrastructure delivery. Load reaches commercial operation in 18 months to three years; the generation, transmission, and rules meant to serve it take roughly four years, seven to ten years, and several years respectively. Planning cycles therefore commit capital and set tariffs against demand that arrives long before the system can serve it. NERC's own assessment places MISO and PJM in elevated shortfall risk from roughly 2028 and 2029 — before the remedies take effect. Two factors constrain infrastructure deployment: an equipment supply chain booked through 2029–30, and unresolved politics over who pays.  LIKELY 

2. The reforms that matter most#

• Flexible, curtailable load service. Among the largest levers available. Giving up 0.25 to 0.5% of annual consumption — partial reductions falling in 85 to 177 hours a year, averaging about two hours each — could accommodate a modelled 76 to 98 GW of new demand on the existing grid, on the same timescale the load itself arrives. The underlying study states its participation and curtailment assumptions, and does not model network constraints. Current evidence indicates the principal barriers are contractual and market-design related rather than technical.  LIKELY 
• NERC registration of computational loads. Among the few levers that reach these loads even when they island off-grid, and the precondition for the rest: a Reliability Standard binds only the entities NERC has registered. NERC targets a first mandatory standard for year-end 2026; the registration criteria determine which entities that standard actually binds. On July 16, 2026 FERC made this schedule binding in Docket RD26-7-000, directing NERC to file both the standards and computational-load registration criteria by December 31, 2026. The deadline is therefore settled; only its content remains open.  LIKELY 
• Mandatory ride-through obligations on load. ERCOT’s NOGRR282 and NPRR1308 impose frequency and voltage ride-through on Large Electronic Loads — the first such requirements placed on demand in North America — alongside dynamic modeling and mandatory registration under NPRR1325. Facilities that had cleared energization approval or completed their interconnection study by November 14, 2025 are exempt, a population ERCOT identifies as a continuing reliability exposure.  SETTLED 
• Queue and forecast integrity. ERCOT's Batch Zero queue screen (live July 11, 2026), the $50k/MW financial security requirement, and a PUCT duplicate-request disclosure rule due by December 2026. Addressing speculative queue entries could substantially reduce future cost-allocation disputes, since costs are socialized today on the basis of load that may never exist.  SETTLED 

3. What to watch over the next 12–24 months#

• Year-end 2026. The first NERC large-load reliability standard and the PUCT duplicate-disclosure rule — both required to convert guidance into obligation. Under FERC's July 16, 2026 order in RD26-7-000 the NERC filing (standards and registration criteria) is now a compliance deadline, with a Phase II work plan due March 1, 2027.  LIKELY 
• Across 2027. FERC's large-load rulemaking maturing from its advance notice (ANOPR) to a proposed and then final rule, and the co-location and flexible-load service terms that follow.  EMERGING 
• 2028–2029. The years NERC's assessment marks as the onset of elevated shortfall risk (MISO ~2028, PJM ~2029). That interval — between the onset of the risk and the readiness of the remedies — defines the period to plan around.  EMERGING 
• An early indicator. Whether any jurisdiction writes a bounded, priced, predictable curtailment product — a stated annual hour cap, limits on event duration and frequency, a notice window, and compensation. A jurisdiction that does so first should attract a disproportionate share of investment, for two reasons established later in this report: time to energization, rather than electricity price, has become the principal siting constraint for many hyperscale data-center developers (Section 2), and developers are already accepting curtailment terms bilaterally in exchange for earlier connection (Section 6). A tariffed version converts a negotiation available only to the largest firms into a published product any developer can price.  EMERGING 

The Assessment boxes throughout carry the author's reasoned judgment rather than neutral summary or cited fact — each labeled and confidence-tagged so it reads apart from the record. Points tagged Emerging in particular remain contestable, and reasonable analysts weighing the same evidence may land differently.