In 2021, lumber went from $400 per thousand board feet to $1,600 in 14 months and added roughly $36,000 to the average new US home. The construction industry talked about nothing else. Five years later, the same kind of structural choke point is back, only now the constraint is electricity, the buyer eating the supply is a data center, and the wait times are measured in years rather than months. Hyperscalers are out-bidding builders for transformers, switchgear, and grid capacity. The US has 2,600 GW of generation and storage stuck in interconnection queues. The median wait to power up a project is five years; for data centers in PJM, it has stretched past seven. Power is the new lumber, and unlike lumber, it is not mean-reverting any time soon. Data centers have already won the supply chain war that most contractors did not know was happening.
Construction has a short memory for supply chain shocks. The lumber crisis of 2021 is already remembered as a weird pandemic anomaly: a temporary spike that resolved itself when the sawmills caught up. That memory is half-right and entirely misleading. Lumber resolved because lumber is a relatively simple supply chain. You can build a new sawmill in 18 months, hire workers, and add capacity. The bottleneck cleared because the bottleneck could be cleared.
The 2026 version of the same story does not have that exit. The constraint this time is not lumber, steel, or concrete. It is electricity, and the equipment needed to move electricity from where it is generated to where it is consumed. A decade ago developers optimized for the lowest delivered cost. In 2026, they are optimizing for earliest energization. Power has become the pacing item on construction projects across multiple sectors, and the reason is simple: data centers have moved faster than the grid can accommodate them.
This is the part most construction executives have not fully internalized. On a recent show, Built Technologies general manager Chris DeVito described what he is hearing on the road from developers across the country. Funded projects, ready to break ground, sitting idle. Not waiting on permits, not waiting on materials, not waiting on labor in the traditional sense. Waiting on power. The grid connection that used to be a check-the-box milestone has become the most expensive question on a development pro forma, and the winners of every available megawatt are not who the construction industry expected them to be.
Power has become the binding constraint on a growing share of US construction. The companies winning that constraint are data center operators, and the rest of the market is now competing for the leftovers. This is a supply chain story, not just an energy story, and it is reshaping which projects get built, where, and on what timeline.
The lumber comparison, by the numbers
How a four-month spike rewired an industry for two years
To see why the power story is bigger than the lumber story, it helps to remember exactly how bad lumber actually got. The cash price of softwood lumber averaged $400 per thousand board feet in early 2020. By the second quarter of 2021, it averaged $1,280 per thousand board feet, peaking near $1,600 in May. The National Association of Home Builders estimated that lumber alone added more than $36,000 to the average price of a new single-family home.
That four-month spike was painful. It also ended. By 2022, prices were back in a range buyers could live with. The sawmills caught up because sawmills can catch up. New capacity comes online in roughly 18 to 24 months. Substitution is possible at the margins. The supply curve is not infinitely elastic, but it is not flat either.
Now apply that lens to the power constraint. The median wait time for a US interconnection project to reach commercial operation is approaching five years, with some data center projects facing potential delays of up to 12 years. Nearly 80% of new projects in the queue eventually withdraw, often because the cost of the required grid upgrades destroys the project economics. There is no equivalent of “the sawmills will catch up” here. Building a high-voltage transformer takes 30 months. Permitting and constructing a new high-voltage transmission line can take a decade. The supply curve is not just inelastic; it is structurally slow.
So many people have projects ready to go and they’re just stuck waiting on power. We have these projects, they’re funded, we’re ready to break ground, but we can’t get access to power.Chris DeVito, General Manager, Built Technologies
Why data centers won the supply chain
And why they were always going to
The framing of this article matters. Power did not become scarce. It got expensive and slow because demand exploded, and that demand has a name: data centers. McKinsey estimates AI data center demand will grow 3.5x from 2025 to 2030, reaching 156 GW worldwide. In Texas alone, CenterPoint Energy reported a 700% increase in large-load interconnection requests between late 2023 and late 2024, jumping from 1 GW to 8 GW. Utilities like ComEd, PPL, and Oncor are now seeing more gigawatts of data center applications than their historical maximum peak demand.
This was not a fair fight for the rest of construction. Hyperscalers have three structural advantages over the typical developer queuing for power. First, they can afford to pay. A $2 billion data center campus can absorb a $40 million transformer surcharge that would kill an industrial or commercial project. Second, they move faster. They engage utilities pre-acquisition, sign letters of intent early, and treat interconnection as a Day 1 underwriting question rather than a permitting afterthought. Third, when the grid will not deliver, they go around it.
That last point is the most important. Faced with multi-year waits, hyperscalers are increasingly bypassing the grid entirely. A Cleanview report published in early 2026 identified approximately 50 gigawatts of behind-the-meter data center power capacity announced in 2025 alone. Meta’s Hyperion project in Louisiana involves three approved Entergy gas plants plus seven more proposed, totaling roughly 7.5 GW of generation for a single campus. xAI trucked mobile generators into Memphis to power its facility before a grid connection was available. Oracle’s Stargate campuses in Texas are explicitly designed around behind-the-meter natural gas.
Other developers cannot replicate that strategy at scale. The capital cost of co-locating generation is enormous, the permitting risk is high, and the operational complexity is not what a typical commercial real estate developer is built for. So the data center operator pulls ahead, and the multifamily developer in Northern Virginia gets told the substation is full.
The transformer problem
A $40 million part holding up $2 billion of capex
If there is one single supply chain choke point that captures the new reality, it is the high-voltage transformer. Wood Mackenzie data shows substation transformer lead times rose from roughly 140 weeks in 2023 to 150 weeks in 2025 to over 160 weeks in 2026. Some high-voltage transformer orders are quoting 36 to 48 months. These are not commodity parts. They are custom-engineered, built by a small number of global manufacturers, and the order book is full for years.
The economics are surreal. A high-voltage transformer might represent less than 2% of a data center campus’s total construction cost, but without it the other 98% sits inert. A $2 billion campus can be held up waiting on a $40 million transformer order that was placed too late. The new playbook is to order the transformer before you have entitlement on the land, because the lead time on the equipment is longer than the lead time on everything else combined.
That dynamic has now spilled outside the data center world. Utilities, who need transformers for their own grid upgrades, are competing with developers for the same supply. Industrial projects, semiconductor fabs, EV manufacturing plants, and large institutional campuses are all bidding into the same constrained order book. The transformer queue is the bottleneck behind the bottleneck.
The hidden labor effect
Why power is also pulling skilled trades away from your project
The power constraint is not just about the grid. It is about the people who build the things that move power. Electrical contractors with experience on high-voltage substation work are being absorbed by data center campuses at rates the rest of the construction market cannot match. Hyperscalers pay tech-company premiums for skilled labor and write contracts with completion bonuses that look more like aerospace than commercial construction.
The effect is the same as the data center demand for transformers. The supply of qualified electrical trades does not expand quickly. The US construction industry needs roughly 349,000 net new workers in 2026, with 41% of the current workforce projected to retire by 2031. The trades best positioned to work on grid and substation projects are exactly the ones the AI infrastructure boom is hiring. Smaller projects in adjacent property types end up competing for the leftover capacity, on a slower schedule, at a higher cost.
Owen Drury and Chase Gilbert touched on this on the recent show. The labor pull is not just an immigration enforcement issue, although that is also pinching the numbers. It is also data center work absorbing capacity from everything else. Two compounding effects, same direction, no obvious fix.
Get the signals before the rest of the industry does
Power, deal flow, and the operator stories shaping the next year of construction. One sharp briefing each week.
Join 2500+ ReadersWhat the market is doing about it
Five strategies developers are using to de-risk the power constraint
The smart money has already moved. Developers and contractors who want to keep building are restructuring their pre-construction process around power as a Day 1 question. The patterns are consistent across hyperscalers, large GCs, and the more sophisticated commercial developers.
Lumber 2021 vs. power 2026
The two crises, side by side
| Dimension | Lumber 2021 | Power 2026 |
|---|---|---|
| Price impact | ~300% increase vs pre-pandemic | Virginia electricity up 267% over 5 years |
| Direct project cost added | ~$36,000 per single-family home | Multi-year delays plus $40M+ transformer orders |
| Time to add capacity | 18-24 months (new sawmills) | 3-10+ years (grid upgrades, transformers) |
| Substitution available | Yes (steel, concrete, alt species) | Limited (behind-the-meter generation only) |
| Primary demand driver | Pandemic housing/DIY boom | AI data center buildout |
| Resolution timeline | ~18 months | Indefinite without grid investment |
Who wins and who loses
The cross-sector impact of the data center power grab
The headline is that data centers won, but the cascading effects are more interesting. Chase Gilbert’s read of his pipeline data, drawn from Built Technologies’ visibility across hundreds of US lenders and developers, captures the dispersion well. Not all markets are created equal. Not all property types are created equal. Data centers are ripping. Spec office is the polar opposite. Multifamily is restarting cautiously after a 2022 pencils-down period. Industrial and manufacturing project growth surged 68.1% year-over-year in 2025 per the Bridgit Workforce Benchmark Report, often into the same constrained power markets where data centers are competing.
- Data centers and AI infrastructure. The cause of the problem and also, paradoxically, the most affected sector. Hyperscalers are absorbing supply chain capacity and labor, but they are also the ones running into transformer shortages and rate disputes. Behind-the-meter is the workaround.
- Industrial and advanced manufacturing. Semiconductor fabs, EV plants, and battery factories need similar grid profiles to data centers and are competing for the same capacity. Site selection has become a power-availability exercise first, land second.
- Multifamily and commercial real estate. Caught in the middle. Smaller per-project power demand, but lower margins make new substation surcharges harder to absorb. Some markets have effectively become unbuildable until the grid catches up.
- Infrastructure and utility work. The big winner over the next decade. Grid expansion is now a strategic federal and state priority, and the contractors who can deliver high-voltage transmission, substations, and behind-the-meter generation will see sustained demand.
The bigger picture
This is the productivity gap finally hitting infrastructure
Construction has been told for years that its productivity problem is a tech adoption problem. That framing has always been incomplete. The deeper issue is that the physical infrastructure underneath modern construction, especially the grid, has been chronically underinvested in for decades. US construction productivity sits roughly where it did in 1950 while every other industry has tripled, and a meaningful share of that gap is structural: the rails, wires, and substations that make new construction possible were not built on a curve that anticipated 156 GW of AI workload demand.
The data center power crisis is the first crisis where that infrastructure debt becomes an active drag on construction itself. It is not going to be the last. EV charging demand, electrification of heating, onshoring of manufacturing, all push in the same direction. The grid was always going to be the constraint. Data centers just got there first and with the deepest pockets.
For construction executives, that means power planning is no longer a utility company problem. It is a pre-construction problem, a pursuit problem, and a portfolio problem. The firms treating it as such are the ones still building. Roughly one-third of the announced 2026 data center pipeline is actually under active construction, with the rest stalled by exactly the constraints described here. The other two-thirds are not lost forever; they are queued. And the queue is everyone’s queue now.
Lumber resolved itself in 18 months. Power will not. That is the headline. The construction industry that internalizes this early, and structures its pre-construction process around energization milestones rather than treating power as a check-the-box, will pull away from the one that does not.
Frequently asked questions
AI data center demand has exploded faster than the US grid can accommodate it. The US has roughly 2,600 GW of generation and storage projects sitting in interconnection queues, with a median wait of approximately five years to commercial operation. Data center projects in regions like PJM are facing wait times of seven years or more. Demand for skilled labor and long-lead equipment has surged alongside, pulling capacity away from other construction sectors. (Source)
Lumber prices rose roughly 300% in 14 months in 2020-2021, peaking near $1,600 per thousand board feet and adding around $36,000 to the average new single-family home per NAHB data. The crisis resolved within roughly 18 months as sawmill capacity caught up. The power constraint is structurally worse: transformer lead times stretch to 160+ weeks, new transmission lines take a decade to permit and build, and substitution is limited to expensive behind-the-meter generation. The supply curve cannot mean-revert on the same timeline. (Source)
High-voltage transformers are custom-engineered, manufactured by a small number of global suppliers, and now require 30 to 48 months of lead time. They can represent less than 2% of a data center campus’s total cost, but without them the rest of the build sits inert. A $2 billion campus can be held up by a $40 million transformer order placed too late. Utilities and developers are now competing for the same constrained supply, which has spilled into rate disputes and project cancellations. (Source)
Yes. Hyperscalers can pay tech-company premiums for transformers, electrical contractors, and grid capacity, and where they cannot get grid power they are co-locating their own generation. Roughly 50 GW of behind-the-meter data center capacity was announced in 2025 alone. The result is that other developers, particularly in multifamily and mid-market commercial, are competing for residual capacity, on slower timelines, and at higher cost. Some markets have effectively become unbuildable in the near term. (Source)
Electrical contractors and trades with high-voltage substation experience are being absorbed by data center campuses at rates the rest of the construction market cannot match. The US construction industry needs roughly 349,000 net new workers in 2026 while 41% of the current workforce is projected to retire by 2031. The combination of retirement, immigration enforcement, and data center demand is squeezing labor availability for adjacent commercial and infrastructure projects, pushing schedules and costs in the wrong direction. (Source)
Treat power as a Day 1 underwriting variable, not a check-the-box milestone. Five practical moves: site-select based on grid capacity and queue position, not just land; order long-lead equipment like transformers before entitlement is complete; engage the utility before the architect; consider behind-the-meter generation where economics support it; and build phased energization into the project schedule rather than assuming full nameplate capacity on day one. The firms doing this are still building; the firms treating power as standard pre-con are not. (Source)
Not on a comparable timeline. Lumber resolved because sawmill capacity can be added in 18 to 24 months. Power generation, transmission, and high-voltage equipment supply expand over multi-year and in some cases decadal timelines. AI workload demand is forecast to keep growing through 2030 per McKinsey, electrification of heating and transport will push in the same direction, and federal and state policy reforms are only beginning to address the queue dysfunction. Some regional relief is possible by late decade. Industry-wide normalization is not. (Source)
Related Articles
Data Centers Can’t Get Power from the Grid. So They’re Building Their Own Power Plants.
Big Tech Promised $650 Billion in Data Centers. Most of It Isn’t Being Built
Construction’s Productivity Problem: 75 Years of Data That Should Bother You
Construction Is Spending Billions on the Wrong Problem
Enki AI: Grid Interconnection Delays 2026, A Threat to US Energy
Data Center Knowledge: Why AI Data Center Projects Face Years of Delays After Approval
Camus Energy: Why Does It Take So Long to Connect a Data Center to the Grid?
LVI Associates: Power Ready or Left Behind, Data Center Site Selection in 2026
RMI: The Interconnection Queue Continues to Be a Barrier to American Economic Competitiveness
