The U.S. Energy Information Administration’s October Short-Term Energy Outlook contains a pair of numbers that should stop any power buyer mid-sentence. Wholesale electricity prices in PJM, the grid operator spanning 13 states and the District of Columbia, are forecast to rise 41 percent in 2026 compared with 2025 — the largest increase across any region EIA tracks — and then fall 7 percent in 2027. No other region moves as sharply in either direction. Taken together, the two numbers describe a market that surged and is expected to settle, but they describe almost nothing about what it costs to buy power in PJM on a given day.
That is the point of this piece. An annual average is a planning input. It is a poor procurement signal. The hours underneath the averages are where the money is, and PJM’s own record for the first half of 2026 shows how wide the gap between the average and the experience can get.
The outlook in full
Start with what EIA actually published. The October outlook, released October 6, 2026 with model inputs finalized October 1, anticipates wholesale electricity prices averaging $52 per megawatt-hour in 2026 across the 11 hubs EIA tracks, 11 percent higher than in 2025, easing to $49 per MWh on average in 2027. EIA attributes the 2026 elevation mainly to weather: winter storm Fern and high summer temperatures, particularly in July, pushed prices up compared with last year across most of the country.
PJM is the outlier at the top. The three regions with the highest wholesale prices this year — PJM, ISO New England, and NYISO — are all expected to see declines in 2027, ranging from 7 percent at PJM to 14 percent at NYISO. EIA’s stated reason for the decline is instructive: the forecast assumes winter temperatures return to near average after relatively cold weather last winter. In other words, a meaningful part of the 2027 improvement is not a structural change in supply or demand. It is a weather assumption. If the coming winter runs cold again, that part of the forecast unwinds first.
The other end of the table makes the same point in reverse. Northwest Mid-Columbia, in the Pacific Northwest, saw the largest decline for 2026, with prices falling 23 percent from 2025 levels. CAISO, meanwhile, is forecast to rise 19 percent to $30 per MWh in 2027 after reaching historic lows this year. Two regions, opposite directions, same national average. A buyer who plans off the $52 national figure without asking which hub, which hours, and which season is pricing a market that exists at no single location.
What an annual average averages away
To see why, recall the structure laid out in this desk’s earlier explainers. Wholesale electricity is not one price but thousands of locational prices every five minutes, plus forward curves stretching years ahead. The day-ahead market schedules most of the next day’s power hour by hour; the real-time market re-optimizes every five minutes against actual conditions. An annual average collapses all of that — every hour, every zone, every weather event — into a single number.
That collapse hides at least three things that matter more than the average itself. The first is shape: power bought at 6 p.m. on a July heat-wave day is a different product, at a radically different price, from power bought at 3 a.m. in April, and a buyer whose load peaks with the system pays disproportionately for the expensive hours. The second is geography: transmission constraints mean prices in one PJM zone can diverge sharply from prices a few hundred miles away, even within the same hour. The third is duration: a price that spikes for 20 five-minute intervals and then collapses leaves barely a trace on an annual average, but it leaves a very real mark on the bill of anyone exposed during those intervals.
EIA is candid about the limits of its own figures. Wholesale prices, the outlook notes, are just one component of residential electricity prices, which also include transmission and distribution upgrades and capacity market auction prices in some regions, among other expenses. The same caveat applies with more force to commercial and industrial buyers: the wholesale average is an input to the bill, not the bill. Hedging, capacity allocation, transmission charges, and the buyer’s own load shape all intervene between the hub average and the invoice.
The first half of 2026, hour by hour
PJM’s independent market monitor, Monitoring Analytics, supplies the hourly record that the annual average conceals. Its State of the Market report for January through June 2026 puts the total cost of wholesale power in PJM at $114.50 per MWh, up 50.3 percent from $76.16 in the same months of 2025. Energy accounted for 63.5 percent of that total, transmission 17.4 percent, and capacity 17.1 percent. Those half-year averages already differ from EIA’s full-year forecast, as expected: a half year ending in June weights winter storm Fern and early summer heavily, while the full-year figure dilutes them with autumn.
The more revealing monitor figures are the ones that describe distribution rather than level. The average spread between day-ahead and real-time prices in the first half of 2026 was -$1.22 per MWh, against +$1.17 in the same months of 2025. Read naively, that looks like convergence improving: the two markets pricing each other more tightly, the gap nearly vanishing. Read properly, it is a warning about averages. A spread near zero can mean every hour priced consistently — or it can mean large positive and negative deviations canceling each other out across thousands of hours.
The second figure resolves the ambiguity. Shortage pricing fired in 102 five-minute intervals on 23 separate days in the first half of 2026. That is the operating reserve demand curve doing the job described in this desk’s piece on why power prices spike: as operating reserves thin toward minimum requirements, the operator adds an administratively determined adder to energy prices, small when reserves are comfortable and rising steeply as they thin. One hundred and two intervals is a small fraction of the roughly 52,000 five-minute intervals in a half year. It is also, potentially, where an unhedged buyer’s year is decided. The intervals cluster on the days when the system is genuinely stressed — the days when demand overshoots forecasts, plants trip, or fuel tightens — and a buyer exposed in real time during those intervals pays scarcity prices while the annual average serenely reports that day-ahead and real-time prices converged.
Capacity did the moving
The monitor’s breakdown also corrects a natural misreading of the STEO headline. A 41 percent rise in wholesale prices sounds like an energy story — fuel costs, weather-driven demand, expensive marginal units. In the first half of 2026, the largest proportional move was not in energy at all. It was in capacity.
The capacity component of PJM’s wholesale cost rose from $6.39 per MWh in the first half of 2025 to $19.61 per MWh in the first half of 2026, a 207 percent increase. Data center load, included in PJM’s capacity auctions, added $11.11 per MWh to the total wholesale cost in the half year, a 9.7 percent increase — and that adder arrived almost entirely through the capacity market. The monitor is explicit that the $11.11 figure covers the capacity market only; it excludes any data center effect on energy prices and transmission costs.
This matters for how buyers should read the STEO. Capacity is procured years ahead through PJM’s Reliability Pricing Model auctions, and it is allocated to load-serving entities on their customers’ contributions to system peak, not spread evenly across every kilowatt-hour. A buyer whose load coincides with system peaks therefore carries more of the capacity cost than a buyer using the same energy off-peak — a distinction no annual average can capture. And capacity, unlike energy, is a commitment: once an auction clears, the payments flow for the delivery year whether or not the forecast load behind them materializes. That is why this desk’s earlier monitor piece treated the capacity forecast as a separate question from the energy price. The STEO’s 41 percent is an energy-price forecast; the capacity shock of the last three auctions is a parallel cost stream running alongside it, and the two should not be conflated into a single “PJM is up 41 percent” shorthand.
It also matters for the 2027 forecast. EIA’s expected 7 percent decline at PJM is a wholesale energy-price expectation driven substantially by a weather normalization assumption. It is not a forecast that capacity costs will fall. Buyers who read the 2027 decline as relief on their total bill, without separating the energy component from the capacity component, will be disappointed on the capacity line even if EIA’s energy forecast proves exactly right.
Why procurement must be hourly
So what should a buyer, a utility planner, or a large customer actually do with an outlook like this? The honest answer is to treat it as the starting point of a procurement analysis, not the analysis itself.
First, decompose the average. Ask how much of the forecast move is energy and how much is capacity, transmission, and ancillary services, and on what allocation basis each component reaches your bill. A flat per-MWh adder applied to a peaky load will misstate the cost in both directions: it understates exposure in the hours that set the capacity obligation and overstates it in the hours that do not.
Second, price the shape, not the strip. A forward contract for around-the-clock power averages away the same information the annual average does. The relevant questions are what the buyer’s load looks like hour by hour against system peaks, how much of that load is hedged forward versus exposed to day-ahead and real-time prices, and what the cost of that residual exposure is during the 23 shortage days rather than the average day. A hedge that covers 90 percent of expected load can still leave ruinous exposure if the unhedged 10 percent coincides with scarcity intervals — the lesson of every retail failure this desk has chronicled, and the reason hedging discipline is existential rather than administrative.
Third, separate the forecast from the assumption. EIA’s 2027 decline rests in part on winter temperatures returning to near average. That is a reasonable central assumption and a fragile one: a single cold winter, a gas pipeline constraint during extreme cold, or a repeat of the July heat that drove this year’s prices would each push the outcome away from the average in the same direction — upward. Sensitivity analysis around the weather assumption is worth more than precision about the central number. The same applies to EIA’s demand expectations: commercial-sector demand, which includes data centers, is forecast to grow 2.8 percent in 2027, with industrial demand up 2.7 percent and residential demand mostly unchanged. Those are sector aggregates, not a data center forecast and not a PJM forecast, and they should not be relabeled as either.
Fourth, watch what would change the forecast rather than the forecast itself. The variables that moved 2026 — winter storm severity, July temperatures, natural gas prices — are the variables to monitor into 2027. The STEO’s timeline is a reminder of how quickly the picture ages: inputs were finalized October 1, and the outlook does not account for later market events.
Reading the outlook like an analyst
Return to the two headline numbers with those disciplines in hand. PJM up 41 percent in 2026, down 7 percent in 2027: the first number is dominated by weather events that have already happened, including winter storm Fern and a hot July; the second is a forecast that assumes weather normalizes. Neither number describes what any particular buyer paid or will pay, because neither accounts for load shape, capacity allocation, hedging, or the 102 intervals when shortage pricing fired. The national $52 average is even further removed: it blends PJM’s surge with Mid-Columbia’s 23 percent decline and CAISO’s historic lows into a figure that corresponds to no actual hub.
None of this makes the outlook useless. It is most useful when read as EIA intends it: a consistent set of regional expectations against which to benchmark one’s own procurement. The error is not in publishing an average. The error is in buying against it — in signing a contract, setting a budget, or siting a load as though the average were the price. In power markets, the average is never the price. The price is the hour, the zone, and the interval, and the institutions that understand that distinction are the ones that survive the years when the average moves 41 percent in either direction.
When the next outlook arrives, ask four questions before reading the headline. What moved: energy, capacity, or both? What drove it: weather, fuel, or load growth — and which of those is an assumption rather than an observation? What does the distribution look like underneath the average: how many intervals, on how many days, at what scarcity prices? And finally, what does your own load look like in those hours, rather than in the average hour? The outlook can answer the first question and gesture at the second. The third and fourth are where the money is, and they are the buyer’s own work to do.