Load Growth
vs. Grid Reality:
The 2026 Numbers

The load-growth debate of 2026 is no longer a contest of forecasts. It is a contest of documents, and the documents disagree: the announced pipeline, the state planning forecast, and the signed interconnection agreement now describe the same future in numbers that differ by an order of magnitude or more.

Section 01The decade the forecast broke

For nearly two decades, from the mid-2000s to the early 2020s, U.S. electricity consumption was essentially flat. Efficiency improvements and the structural shift away from energy-intensive manufacturing offset economic growth almost exactly, year after year, and an entire generation of planning practice was built on that flatness.¹ Interconnection procedures, distribution planning cycles, and resource plans were all tuned to a world in which new demand arrived slowly, predictably, and in small increments.

That world ended in the data before it ended in the institutions. The U.S. Energy Information Administration reports that national consumption set an all-time high in 2024 and forecasts new records in 2025 and 2026. The growth concentrates where large sites live: EIA's 2020 through 2026 outlook has commercial-sector sales, the category that contains data centers, growing at an average of 2.6 percent per year, and industrial sales at 2.1 percent, against 1.7 percent for the system overall.¹ Those percentages look modest on paper. On a national base measured in the trillions of kilowatt-hours, each point represents capacity that has to be planned, permitted, built, and paid for, and because the growth clusters in a handful of customer classes and places, the stress lands on specific circuits and corridors rather than on the national average.

The continent's reliability assessor then quantified how fast expectations were moving. The North American Electric Reliability Corporation's 2025 Long-Term Reliability Assessment, published in January 2026, forecasts summer peak demand growing by 224 gigawatts over the next ten years. One year earlier, the 2024 edition of the same assessment forecast 132 gigawatts. That is 92 gigawatts added to the ten-year outlook in a single annual revision, an increase of roughly 70 percent, with winter peak growth forecast higher still at 246 gigawatts. NERC attributes most of the increase to data centers and other large loads, and it pairs the forecast with a warning that resource additions and supporting infrastructure are lagging the demand now visible in utility submissions.²

A practitioner should read the revision itself as the finding. When a reliability body raises a ten-year forecast by roughly 70 percent in twelve months, one of two things is true. Either demand genuinely accelerated faster than any institution anticipated, or the inputs feeding the forecast, which are utility-reported pipelines built from customer service requests, contain expectations that are not all going to materialize. The 2026 evidence says both are partly true, and the tension between them, not either headline number, is the actual planning environment.

Section 02One pipeline, four stages

National forecasts aggregate; contracts itemize. The clearest place in the country to watch announced demand meet executed capacity is a single utility's investor disclosure, because a presentation furnished to the U.S. Securities and Exchange Commission carries a discipline that a press release does not. Pacific Gas and Electric Company's second-quarter 2026 earnings presentation is, for this purpose, the reference document of the year.³

The presentation reports a data-center pipeline of 12,710 megawatts as of June 2026, swelled by intake from the company's 2026 cluster study and roughly two and a half times the figure reported for March 2026 in the same disclosure series.³ The pipeline is presented in stages, and the stages are the story, because each stage is a claim of a different strength.

The arithmetic, derived from the same disclosure, is the market evidence this paper exists to put on the record. Capacity with executed agreements is about 4 percent of the announced pipeline. Adding construction in progress brings contractually or physically committed capacity to roughly 630 megawatts, about 5 percent. For every megawatt with a signature, there are roughly 26 announced megawatts waiting on one. And the utility's own stated expectation is that about 1.8 gigawatts of the pipeline is in service by 2030, approximately one-seventh of the announced figure.³

None of this is a criticism of the utility. Staged disclosure is what honest reporting of an uncertain pipeline looks like, and publishing the stages at all is a service to the market. The point cuts in both directions. A pipeline that grows this fast in a single quarter says the demand interest is real, large, and accelerating. An executed column at 4 percent says the near-term physical grid is committed to a small fraction of it. Both statements are true at once, and confusion begins the moment a reader quotes one column as if it were the other.

A pipeline megawatt is a statement of interest. An executed megawatt is an obligation. The 2026 numbers measure the distance between the two.

Section 03Three institutions, three numbers

Now set three published California numbers side by side. First, 12.7 gigawatts: one utility's announced data-center pipeline, from its investor disclosure.³ Second, 1.8 gigawatts: the California Energy Commission's forecast, as of January 2026, of data-center load growth across the entire California ISO grid by 2030, as recorded in the ISO's large-load issue paper, which also carries a 4.9 gigawatt figure for 2040.⁴ Third, 0.49 gigawatts: the capacity holding executed interconnection construction agreements within that same utility pipeline.³ Three serious institutions, three purposes, three answers separated by successive multiples.

The numbers disagree because they measure different things: interest, expectation, and obligation. A service request costs relatively little to file and binds no one, so a developer siting one facility may hold positions in several service territories and several states simultaneously. The same eventual data center can therefore appear in multiple utility pipelines at once and, in the end, be built in none of them. Planning forecasters know this, which is why the state's grid-wide 2030 planning figure is a fraction of a single utility's announced pipeline: a planning forecast discounts announcements for the probability of realization, while a pipeline disclosure counts requests and a contract ledger counts signatures. None of the three is dishonest. Each answers a different question, and each misleads only when quoted as the answer to a different one.

The honest residual is that nobody, including the institutions publishing these figures, knows the realization rate. NERC's one-year revision shows forecasters chasing utility submissions upward.² The executed column shows contracts lagging far behind announcements.³ The eventual truth lands somewhere between, unevenly, territory by territory. For an owner planning a specific site, the spread itself is the datum worth watching: the wider the gap between announced and executed capacity in a region, the more crowded the queue ahead of any new request, and the less an early-stage position is worth in schedule terms.

Section 04The referees move

The institutions that govern the grid spent 2026 acknowledging the gap in their own documents, which is itself evidence of how structural it has become.

In June 2026, the Federal Energy Regulatory Commission announced targeted action to speed the integration of large loads onto the bulk power system. Rather than opening a single national rulemaking, the Commission issued show cause orders to each of the organized-market operators under its jurisdiction, the California ISO among them, directing each to demonstrate that its existing tariff provisions for large-load interconnection and co-location arrangements remain just and reasonable, or to propose revisions, with initial responses due on a schedule measured in weeks.⁵ Read plainly, the federal regulator has concluded that the frameworks connecting large loads may no longer fit the era, and it reached for the fastest procedural instrument available to it.

California's grid institutions were already moving in parallel. The California ISO opened a large-load interconnection and energization initiative with an issue paper in January 2026, noting that utilities are receiving an increasing number of large-load interconnection and service applications, and that co-located load and generation are today studied in separate processes: the load through the transmission owner's process, the generation through the ISO's own.⁴ In May 2026, the ISO's board approved the 2025-2026 transmission plan: 38 projects with an estimated cost of $6.7 billion at full buildout over the next decade, against state forecasts of roughly 15 gigawatts of load growth by 2035 and 20 gigawatts by 2040, with associated resource capacity needs exceeding 74 gigawatts by 2035.⁶

An owner should read the institutional response honestly, in both directions. The case for encouragement: by regulatory standards this is fast. Show cause orders operate in weeks where rulemakings operate in years; flexible-connection constructs that trade firmness or curtailment for speed can land inside a corporate deadline window; and a transmission plan that leans on reconductoring adds real capacity without decade-long greenfield construction. The case for sobriety: tariffs are revised in months to years, transmission is built in years to decades, and none of it changes what is physically available to a site carrying a 2027 or 2028 energization deadline. Process reform mostly narrows the queue for the next cohort of requests. It rarely rescues the current one.

Section 05What the gap means for one site

A market statistic is not a site fact. The 26-to-1 spread between announced and executed capacity describes the crowd, not any individual request, and the first diagnostic for an owner is positional: which stage of the funnel does your own request occupy, and what document evidences it. A dated, written capacity and timeline statement from the utility is the benchmark. A recollection of an account representative's optimism is not a document, and in a queue this crowded the difference between the two is worth a board-level conversation.

The position cuts both ways. An owner holding an executed agreement or a documented near-term energization date holds an asset the 2026 numbers say is scarce; the rational course is to defend it, perform its milestones precisely, and decline to abandon queue position on generalized market fear. An owner whose request sits in the early stages of a crowded pipeline should draw the opposite conclusion: published queue membership is weak evidence of a date, and the planning burden shifts toward alternatives that do not depend on winning a 26-to-1 race on schedule.

Those alternatives are a menu, not a product, and each entry carries an honest case against it alongside its case for.

• Flexible or curtailable service can connect materially sooner where offered, because it asks less of the constrained system. Its firmness is reduced by design, and the curtailment terms are the economics; the date is only half the offer.
• Phased energization delivers partial capacity earlier for loads that genuinely ramp. The later phases are a negotiation, not an entitlement, and they inherit the queue's risk.
• Storage as a bridge suits short-duration gaps and demand management well. It shifts energy within a day rather than creating it, and recharge depends on whatever service already exists at the meter.
• On-site generation is a family, not a machine. Reciprocating engines typically carry the lowest capital cost among common choices and a deep service ecosystem, and face demanding combustion permitting in California's stricter air districts, plus noise and scheduled-outage burdens. Combustion turbines and microturbines are compact and reward sites with real thermal demand, while part-load efficiency deserves scrutiny and the permit remains a combustion permit. Fuel cells operate without combustion, quietly and at high electrical efficiency, which in several districts lightens air permitting; they carry higher capital cost, exposure to gas prices, and stack-replacement lifecycle items. Linear generators are fuel-flexible and modular, and as a newer equipment class they shift the diligence burden to fleet operating history and service depth. Solar paired with storage buys no fuel and can be decisive for daytime-weighted loads, while its capacity factor rarely carries continuous load alone and land or roof area binds early.
• Relocation to already-powered sites is a real path now that powered land carries explicit market value. The honest ledger includes people, logistics, entitlements, and time, not just energy.
• Waiting, with the delay priced, is the baseline every other path must beat. Unpriced delay is still the most common hidden cost in this field.

The discipline that connects them is the one this series exists to describe: price every credible path for the specific site, with sourced and dated inputs, before selecting any technology. A recommendation prepared by a party compensated only when one of these entries wins is a proposal, and proposals deserve to be read as proposals.

Section 06Six rules for reading the 2026 numbers

The evidence assembled above supports a short reading discipline. It costs nothing to apply and, in our experience, it changes conclusions more often than any model does.

1. Announced, planned, executed, and energized are four different words. Most 2026 controversy about load growth dissolves into one of these categories being quoted as another. Insist on knowing which column a number comes from before letting it move capital.
2. Date every number. A utility pipeline grew roughly two and a half times in one quarter, and a national ten-year forecast rose roughly 70 percent in one year.^2,3 In a market moving at that speed, an undated figure is expired on arrival.
3. Ask what the counting unit is. Requests can count one eventual facility several times across territories and states. Contracts cannot. That asymmetry, more than any bad faith, explains most of the spread between headline numbers.
4. Treat revision velocity as two-sided. The same feedback loop that added 92 gigawatts to a forecast in a year can subtract it. A serious site plan holds up in the boom case and in the correction case, because the 2026 documents cannot tell you which one arrives.
5. Separate market statistics from site facts. The only numbers that belong in a capital decision are written, dated, and specific to your meter: the utility's documented capacity and timeline for your request, your measured load, your air district's actual treatment of your candidate equipment.
6. Watch the executed column. Announced capacity tells you about sentiment. Executed and under-construction capacity tells you what the grid's near-term future actually contains. In every disclosure that separates the two, the second column is where reality is legible.

The 2026 disclosures did the market a genuine service: they made the gap measurable at last. For the market as a whole, reconciling a 224-gigawatt expectation with a 490-megawatt executed ledger is likely to take years and to land unevenly. For one site, the reconciliation happens on the day capital is committed, and it happens on the documents in hand. An owner who reads the columns correctly, prices the full menu, and buys the decision before the technology is not exposed to the argument at all.