The Rebuild: Power, Equipment and Crews

The Argument in Brief

The United States is trying to add data centers, generating plants, grid and some factories at the same time, after a long stretch in which electricity consumption was essentially flat for nearly two decades. Our working argument was that physical items with long queues bind: gas turbines, large transformers and switchgear, interconnection slots and skilled electrical crews. The companies that sell the scarcest items collect backlog first, then pricing, then margin.

Testing changed that in three places. The factory leg is missing: manufacturing construction spending was $170.7 billion at a seasonally adjusted annual rate in August 2026, 19.2% below a year earlier, so the rebuild is a data center and power-system story. Scarcity is strongest in turbines and large grid equipment and hardest to document in labor, where official data show steady growth but no measured shortage. And scarcity does not guarantee margin: Eaton’s largest segment saw trailing-twelve-month organic orders rise 41% while its quarterly operating margin fell, because input costs rose.

Demand: Two Flat Decades, Then Growth

EIA’s October 2026 Short-Term Energy Outlook puts U.S. electricity consumption at 4,195 billion kilowatt-hours in 2025, 4,288 billion in 2026 and 4,356 billion in 2027. The commercial sector, which includes data centers, goes from 1,493 billion in 2025 to 1,592 billion in 2027. Industrial use goes from 1,042 billion to 1,083 billion over the same years. EIA expects 2027 growth of 2.8% in the commercial sector and 2.7% in the industrial sector, with residential use roughly unchanged.

Two cautions apply. A hot summer lifted third-quarter 2026 consumption 4% above the prior-year quarter, so some of the 2026 number is weather. And the growth rate is meaningful but not explosive. In May 2025 EIA expected consumption to grow 1.7% a year from 2020 through 2026, with commercial growing 2.6% a year and industrial 2.1% a year. The national curve bends upward. It does not break.

Data centers are the part with the widest forecast range. Lawrence Berkeley National Laboratory (LBNL) estimated that data centers used 176 terawatt-hours in 2023, about 4.4% of U.S. electricity, up from 58 terawatt-hours in 2014. Its 2024 report projected 325 to 580 terawatt-hours by 2028, or 6.7% to 12% of the national total. The June 2026 update moved the horizon to 2030 and set a reference case of 649 terawatt-hours, or 11.8%, inside a scenario range of 9.5% to 15.3%. Its single-assumption sensitivities run from 578 terawatt-hours with fewer IT installations to 782 with higher server idle power and utilization, and the compounded bounds are 521 to 843 terawatt-hours. The lab builds these from equipment shipments and per-device power, so they are bottom-up estimates and still wide.

Grid operators plan to the high side. PJM, the largest U.S. regional grid, forecast in January 2026 a 2036 summer peak of about 222,000 megawatts, nearly 66,000 megawatts above today, a growth rate of about 3.6% a year. PJM also applies adjustments for large loads that trim near-term peaks, including 0.7% off the summer 2026 peak. The market has already priced the tightness: EIA expects 2026 wholesale power prices to average $52 per megawatt-hour, 11% above 2025, with PJM up 41%.

For equipment makers, the forecast spread matters less than the lead times. When a transformer takes years to deliver, a utility cannot wait to learn which forecast was right.

Utilities Are Spending to Match

The Edison Electric Institute (EEI), which tracks investor-owned utilities, projected in October 2025 that its members would invest nearly $208 billion in 2025. In June 2026 it projected $239 billion for 2026. These are projections from two report vintages, so the step up is indicative rather than a clean year-over-year comparison.

Dollars overstate units. Power transformer prices are up 77% since 2019 according to Wood Mackenzie data, and medium-voltage switchgear is up 50% since 2021. Part of utility capex growth is the same work costing more. That is good for suppliers and a cost problem for ratepayers, which is one reason regulators and the political process are a risk to the cycle later in this report.

The contractors who build for utilities show the same pull. Quanta Services reported a total backlog of $63.7 billion at September 30, 2026, of which $38.4 billion is expected within twelve months.

Generation: The Turbine Queue

The clearest scarcity is in large gas turbines. GE Vernova’s gas backlog plus slot reservations stood at 122 gigawatts at September 30, 2026, up from 116 gigawatts a quarter earlier. Management expects at least 130 gigawatts by year-end. Within the 122 gigawatts, firm backlog rose from 53 to 64 gigawatts while reservations fell from 63 to 58 gigawatts, which means reservations are converting into contracts.

Delivery dates tell the same story. GE Vernova said on its second-quarter call that it books reservations four to five years out, is on track to be more than halfway contracted for 2031, and needed more time before it could describe contracting for 2032. By the third-quarter call in October, management said it had sold multiple slots for 2032 delivery. At Mitsubishi Power, an Asia Pacific executive said in October that delivery is around 2031 and that 2028 and 2029 slots are almost fully booked, with global orders this year projected at 70 to 80 gigawatts.

Wood Mackenzie’s April 2026 analysis frames the gap. It put global gas turbine orders at about 110 gigawatts at the end of 2025 against annual manufacturing capacity of 60 to 70 gigawatts. It reported turbine prices up 195% since 2019 and forecast $600 per kilowatt by the end of 2027. It also estimated that the turbine is 20% to 30% of the cost of a combined-cycle plant. That last number explains why buyers pay: a large price increase on a minority of project cost is tolerable, and a missed delivery slot is not.

Two facts temper the AI framing. GE Vernova’s customers are about 80% traditional customers such as utilities and about 20% data centers, per the Utility Dive account of the second-quarter call. Turbine demand is therefore a utility resource-adequacy story first. And the turbines sold today are not what is being built today. EIA’s developers report planning 86 gigawatts of additions in 2026, of which 43.4 gigawatts is solar, 24 gigawatts is batteries, 11.8 gigawatts is wind and 6.3 gigawatts is gas, after 53 gigawatts in 2025. The gas fleet that today’s orders will create arrives from 2029 onward. Between now and then, the power supply is solar, storage and existing plants.

Supply is responding slowly. GE Vernova plans to lift annual turbine capacity from 20 gigawatts to 30 gigawatts by 2030. Mitsubishi Heavy Industries said in 2025 that it aimed to double production capacity within two years, and we have not seen confirmation that it has done so.

Wires and Boxes: Transformers and Switchgear

Grid equipment is the second bottleneck, and it is older. Wood Mackenzie’s second-quarter 2025 survey found average lead times of 128 weeks for power transformers and 144 weeks for generator step-up transformers, with high-voltage switchgear at 44 weeks. The firm estimated 2025 supply deficits of about 30% for power transformers and 10% for distribution transformers, and demand for generator step-up transformers had risen 274% since 2019.

That survey is now more than a year old, and we did not find a newer primary dataset. Later reporting points the same way. PwC analysts put lead times for high-capacity units at as long as four years. A construction firm’s own project data show 52 to 80 weeks for medium-voltage switchgear, though that is one company’s sample and not a national survey. Treat the exact weeks as approximate and the direction as settled.

Manufacturers are adding capacity, with delivery dates that set the clock for relief. Wood Mackenzie counted about $1.8 billion of announced North American transformer expansions. Hitachi Energy’s $457 million South Boston plant is targeted for 2028, Siemens Energy’s $150 million Charlotte plant for early 2027, and Eaton’s $340 million South Carolina plant for 2027. If those dates hold, the tightest period for large transformers runs through roughly 2027 to 2028, and the benefit goes to the suppliers already in the queue.

Inputs add a cost layer. Eaton’s Electrical Americas segment reported an operating margin of 27.5% in the second quarter of 2026 versus 29.5% a year earlier, and the company attributes a 470 basis point decline in the first-half margin to commodity inflation, partly offset by volume. Backlog and orders can outrun price when copper and steel move.

Interconnection and Permitting

A plant or load that has equipment still needs a grid connection. LBNL’s 2026 interconnection study found that the median project built in 2025 took 61 months from request to commercial operation, compared with 36 months in 2015 and 22 months in 2008. About 2,061 gigawatts of capacity was actively seeking interconnection at the end of 2025, a 10% decrease for the year, and only about 19% of projects requesting interconnection from 2000 to 2020 had reached operation by the end of 2025.

The gas line in that queue is the one to watch. Active natural gas capacity rose 86% to 253 gigawatts while solar, storage and wind all declined. Developers are filing for gas in volume, and most of those filings will not be built, since the same study shows low completion rates. A large queue is a measure of intent. Turbine slots, interconnection studies and equipment are what convert it, and each is limited.

The time dimension supports the scarcity argument from a different side. A five-year path from request to operation means that a data center developer who wants power in 2030 must secure equipment and a grid position now. That behavior is what produces slot reservations, advance deposits and long-dated backlog at the equipment makers.

Crews

Labor is the input we can document least well. The Bureau of Labor Statistics (BLS) counts 821,000 electricians in 2025 and projects 9% growth from 2025 to 2035, an addition of 75,900 jobs, with about 72,700 openings a year once retirements and transfers are counted. Median pay was $63,190 in May 2025. Most electricians complete a four- or five-year apprenticeship with about 2,000 paid hours a year. For line installers and repairers, BLS reports 131,900 jobs in 2025, 10% projected growth, 10,900 openings a year and a median of $95,320. That group includes telecommunications linemen as well as electric power workers, so it overstates the pool for utilities.

What this shows is a trade that grows faster than average and needs years to train. What it does not show is a measured shortage. BLS projections are a ten-year trend, and we did not find a public series for vacancies or wage premiums on utility and data center projects. The evidence from the companies is qualitative: Quanta lists a potential shortage of and increased costs with respect to skilled employees among its risks and credits a skilled workforce that moves across end markets. Its guidance for 2026 revenue is $41.3 billion to $41.6 billion.

The scarcity of crews works differently from the scarcity of turbines. A turbine maker sells a scarce product at a higher price. A contractor with scarce labor sells hours that cost more to produce. The contractor’s advantage is that it has the crews at all, and that advantage shows up first as backlog and share, and only later, if at all, as margin.

Where the Construction Money Is Going

Census data supply the cleanest check on the “everything at once” premise. Total construction spending was $2,203.1 billion in August 2026, 1.7% below August 2025. Inside that total, power construction was $186.0 billion, up 8.5%. Office construction, the category that carries most data center building, was $150.7 billion, up 24.6%. Manufacturing was $170.7 billion, down 19.2% from $211.1 billion.

Two readings follow. The first is that the factory wave has crested. The release gives no cause, and we have not tested one. The second is that the total pie is shrinking, with residential building down 4.8%, so power and data center work is taking a larger share of a flat construction economy. That may ease some trade competition, though electrical specialties do not transfer easily from housing.

EIA’s forecast has industrial electricity use growing 2.7% in 2027, and Eaton reports weakness in residential and industrial end markets in the second quarter of 2026 offset by data center strength. Both are consistent with a cycle in which data centers and utilities are the buyers and factories are a minor one.

Policy: Tariffs, Credits and the Grid Queue

Three policy settings shape supply and cost.

Tariffs raise input costs. Under a proclamation issued on April 2, 2026 and effective April 6, 2026, Section 232 duties apply to the full customs value of covered steel, aluminum and copper articles. Derivatives made of more than 15% of these metals pay 25%, while certain grid and industrial equipment is capped at 15% through 2027. We could not confirm from primary text which transformer and switchgear codes sit in that annex. The cap’s expiry after 2027 is a date to watch, as it would raise import costs just when the new domestic plants are due.

Tax credits shape what gets built before the gas fleet arrives. The One Big Beautiful Bill Act, signed July 4, 2025, requires wind and solar projects that begin construction after July 4, 2026 to be placed in service by December 31, 2027 to claim the 45Y and 48E credits. Projects that began on or before that date are exempt from the 2027 deadline, and storage follows the existing phase-out schedule. The start-of-construction window has now closed. Our inference is that the solar and wind pipeline for the late 2020s is largely set by projects already begun, which leaves gas, storage and nuclear to fill additional demand.

Permitting and interconnection reform has the largest uncertainty. LBNL notes that interconnection process reforms are underway and that several regions processed record numbers of interconnection agreements in 2025. Faster approvals would shorten the 61-month median and could weaken the scarcity premium on slots. They would also raise demand for the equipment those projects need.

Who Captures the Scarcity

The three segments of the chain show different degrees of pricing visibility.

Turbine makers have the clearest evidence. GE Vernova said third-quarter margins expanded on higher price, volume and productivity, and that Gas Power margin gains were primarily driven by favorable price and higher volume. The company’s total backlog was $197 billion, up $20 billion in the quarter. Slot reservations let the company schedule capacity years ahead, which is the visibility the argument predicts.

Grid equipment sellers have strong volume and a more mixed price record. Eaton’s Electrical Americas backlog was $15,175 million at June 30, 2026, up 33% from a year earlier, with a book-to-bill ratio of 1.3. Yet segment margin fell. GE Vernova’s Electrification backlog was $42.2 billion, up 60% year over year, including $5 billion from Prolec GE, an acquisition, so the organic comparison is smaller.

Contractors have the most backlog per unit of risk and the least obvious pricing power. Quanta’s $63.7 billion backlog gives revenue visibility, and its scarce input is crews. Whether that converts to margin depends on contract structure, which we have not examined in this report.

The ordering is by how hard the scarce item is to replicate. A gas turbine slot is hard to replicate in under a decade. A transformer slot is hard to replicate in two to four years. A crew can be built in four to five years of apprenticeship, with the cost borne by the contractor.

What Would End the Cycle

The cycle ends if demand falls short, supply catches up, or cost allocation breaks politically. Each has a measurable marker.

Demand could fall short. LBNL’s own lower-installation case gives 578 terawatt-hours in 2030 against a 649 terawatt-hour reference, so a modest miss sits inside the lab’s own range and would not by itself end the cycle. The largest single driver is data center capital spending by the biggest cloud companies. We could not retrieve a primary third-quarter 2026 capital spending figure for them in time for this report, so this link is the least verified in our chain. A sustained cut to their spending plans would show first in GE Vernova’s reservation conversion and Eaton’s data center orders.

Supply could catch up. Turbine capacity is planned to rise from 20 to 30 gigawatts a year at GE Vernova by 2030 and Wood Mackenzie sees industry capacity of 60 to 70 gigawatts against about 110 gigawatts of orders, a gap that takes several years to close even if every expansion lands. For transformers the clock is 2027 to 2028, when the announced plants start. A slowdown in the reservation count is the earliest sign of a plateau, and it would show before capacity ends the shortage.

Cost allocation could break. Wholesale prices in PJM are up 41% in 2026, and capex is rising on price as well as volume. If regulators shift cost to large loads or slow approvals, utility spending plans would be the first number to change.

Fuel is a fourth variable. EIA forecasts Henry Hub gas at $3.16 per million Btu in 2027, 9% lower than 2026. Cheap gas supports new gas plants. A spike like January 2026, when Henry Hub averaged $7.72 after Winter Storm Fern, would raise the cost of running them, though it would also raise the value of reliable capacity.

Signposts we would use, with thresholds: GE Vernova’s gas backlog plus reservations below its own 130-gigawatt year-end target; Eaton’s Electrical Americas book-to-bill below 1.0 on the 1.3 reported at June 30, 2026; power construction spending turning negative year over year against the 8.5% gain in August 2026; and Quanta’s twelve-month backlog falling below the $38.4 billion reported at September 30, 2026.

The Companies

GE Vernova (GEV)

GE Vernova sells the gas turbines in the longest queue and a growing line of grid equipment. Its third-quarter 2026 orders were $29.3 billion and revenue was $12.6 billion, up 27%. Gas backlog plus reservations reached 122 gigawatts, and about 80% of customers are traditional customers such as utilities, with about 20% data centers. That mix makes it a bet on power adequacy more than on any one hyperscaler. Read the full GE Vernova memo.

Conversion and timing are less certain. Reservations are not firm orders, and GE Vernova only began selling 2032 delivery slots in the third quarter. Capacity has to rise from 20 to 30 gigawatts a year for the backlog to ship on schedule. Pricing evidence is favorable today, and the test is whether the older, lower-priced slots roll off before the new, higher-priced ones are delivered.

Eaton (ETN)

Eaton sells the switchgear, power distribution and related gear that sit between the grid and the data center or factory floor. Electrical Americas net sales were $3,951 million in the second quarter of 2026, up 18%, with trailing-twelve-month organic orders up 41% and backlog of $15,175 million. Total company backlog was about $24.1 billion at June 30, 2026, versus about $19.8 billion at December 31, 2025, a period that includes the $9.55 billion Boyd Thermal acquisition closed in March. Read the full Eaton memo.

The pressure point is margin. Electrical Americas margin fell two points year over year on commodity inflation even as orders surged, and the company reports weaker residential and industrial demand. Third-quarter 2026 results were not yet in our filing set as of October 30, so the latest hard data are the second-quarter figures.

Quanta Services (PWR)

Quanta builds the lines, substations and interconnections that turn equipment into service. Its third-quarter 2026 revenue was $12.01 billion, total backlog was $63.7 billion and 2026 revenue guidance was raised to $41.3 billion to $41.6 billion. Read the full Quanta Services memo.

Quanta’s constraint is skilled labor, which the company names as a risk. Backlog is high and visible, and converting it without diluting margins depends on whether contractors can hire and train fast enough. Acquisitions during the period, including two completed in September and October, make organic comparisons harder.