Power Generation Trends: Demand Has Outrun the Grid, and That Changes What You Can Build

The defining trend in power generation is not a technology. It is that demand has grown faster than the grid or the equipment supply chain can respond.

Three numbers carry the whole argument. ERCOT is tracking roughly 474 gigawatts of large-load interconnection requests, around 90 percent of them data centres. PJM interconnection waits now approach seven years for large campuses in Northern Virginia. And gas turbine delivery slots for 2026 and 2027 are largely sold out.

Against that, a data centre build cycle runs 12 to 24 months and an interconnection takes 36 to 84 months. The gap between those two numbers is why roughly 101 gigawatts of behind-the-meter gas generation has been announced in the United States, and why one Tennessee project energised 200 MW of load in 122 days using mobile generation rather than waiting.

This guide covers the demand, the bottleneck, the equipment position, and what an operator who needs power before 2030 should actually do.

Currency note: every figure below carries its date and source. Figures in this category move quarterly, and this page should be reviewed against current grid operator and OEM disclosures before any capital decision.

Demand: How Much, How Fast

US electricity demand is growing at a rate not seen since the 1980s, and data centres are the largest single driver.

US data centre electricity consumption reached an estimated 176 terawatt-hours in 2025, and is forecast to grow to between 420 and 728 TWh by 2030, depending on interconnection success and the pace of AI adoption. Rabobank's research puts the incremental addition at roughly 300 TWh by 2030, equivalent to the combined annual consumption of New York and California, and describes it as the largest five-year surge to the US grid since the 1980s.

The capital behind it

The five largest hyperscalers are expected to spend between $745 billion and $775 billion on capital expenditure in 2026 alone. That spending is what converts a forecast into interconnection requests.

How fast the forecasts moved

The speed of revision matters as much as the level. World Resources Institute reports that Grid Strategies found the total of five-year forward summer peak demand growth forecasts published by US utilities rose from 38 GW in 2023 to 128 GW in 2024, more than tripling in a single year.

Rystad Energy has identified over 100 GW of US data centre demand coming online between 2024 and 2035, roughly ten times New York City's 2023 summer peak. EPRI has estimated that data centres could consume between 4.6 and 9.1 percent of all US electricity.

Is it overstated?

Partly, and the correction is already visible. Against 16 GW of announced 2026 data centre capacity, only 5 GW is currently under construction. Announcements are not projects, and a proportion of any pipeline never energises.

That does not resolve the constraint. Even the discounted number exceeds what the grid and the equipment supply chain can deliver, which is why the queue and the lead times are the operative facts rather than the forecast.

Demand is not confined to data centres either. Industrial electrification, onshoring of manufacturing and electrified transport all add load to the same networks. For electrification in oil and gas specifically, see our guide to field electrification and site power. For how AI load is reshaping generation requirements, see our analysis of AI and data centre power needs.

The Interconnection Bottleneck

Interconnection has replaced chip supply as the primary bottleneck for AI data centre deployment, and the waits are now measured in years rather than months.

An interconnection queue is the ordered process by which a new generator or large load applies to connect to the transmission system, has its impact studied, and receives an agreement defining the required network upgrades and who pays for them. A large-load request is the load-side equivalent of a generation interconnection request, and several grid operators now run separate queues for each.

The current position

ERCOT is tracking approximately 474 GW of large-load requests, around 90 percent of them data centres, per ERCOT Senate testimony in July 2026. That figure was approximately 356 GW as recently as March 2026.

PJM interconnection waits now approach seven years for large AI campuses in Northern Virginia. Lawrence Berkeley National Laboratory research finds that interconnection wait times have more than doubled over the past fifteen years, with projects now averaging five years in queue before reaching commercial operation. In primary data centre markets, average grid connection waits exceed four years.

Grid interconnection timelines of 36 to 84 months are structurally incompatible with data centre build cycles of 12 to 24 months. That single sentence is the trend.

A campus joining the Northern Virginia queue in May 2026 cannot realistically expect to draw utility power before 2030 to 2033, regardless of how fast the building is constructed or how much capital is committed.

Why it got slower

Three forces compounded. Demand, as data centres, electrification and new manufacturing flooded utilities with large-load requests simultaneously. Equipment, because the network upgrades interconnection requires face the same lead times as the customer's own substation. System headroom, because the regions developers want have less spare transmission capacity than the incoming demand requires, making upgrades the rule rather than the exception.

PJM reported in January 2026 that 29 percent of project milestone change requests in its queue were driven by permitting delays and a further 23 percent by supply chain constraints.

Reform and intervention

FERC Order 2023 reformed generator interconnection from a serial first-come first-served process to a first-ready first-served cluster study framework, with higher commercial readiness deposits and withdrawal penalties.

ERCOT approved a Batch Zero framework in June 2026, grouping large-load projects of 75 MW and above into coordinated study batches, with roughly 205 GW of large load eligible.

And intervention can be abrupt. On 3 August 2026, Governor Abbott halted all new Texas data centre grid connections pending a comprehensive PUCT and ERCOT audit.

Treat queue position as an asset with political risk attached. File early, because position is chronological and the queue only lengthens, and do not build a schedule that assumes a study completes on its first estimate.

Equipment Supply

Gas turbine delivery slots for 2026 and 2027 are largely sold out, and the constraint now extends to transformers, breakers and conductor.

GE Vernova's chief executive confirmed on the company's Q1 earnings call that slots for delivery in 2026 and 2027 are largely sold out. Approximately 60 percent of the gas turbine supply chain is dedicated to the US market, of which data centres are expected to receive around a third of output beyond 2027.

Those figures imply an upper limit of roughly 11.4 GW of gas turbines available to US data centres in 2028, after existing orders are delivered.

The consequence is already measurable

Analysis of the ERCOT market projects that Texas data centre buildout will be limited to 44 GW by 2035, approximately 24 percent below the ERCOT outlook, specifically because of the shortage of gas-powered turbines for self-supply.

Equipment availability is now setting the ceiling on load growth, not capital, not permitting and not demand.

Beyond turbines

Power transformers, breakers and large conductor face lead times of 90 to 130 weeks, roughly two to two and a half years. These are the same components network upgrades require, which is why equipment scarcity lengthens the interconnection queue as well as the private build.

What to do about it

Reserve slots before design freeze. Where an in-service date is fixed, securing a manufacturing position ahead of full configuration is a legitimate sequence and increasingly the only one that works.

Consider relocated and refurbished assets. Low-hour and surplus units can mobilise far faster than new-build equipment, which is decisive where the deficit is immediate. Condition assessment, remaining life and warranty position must be established before commitment, and insurers price relocated equipment differently from new.

Order the balance of plant in parallel. A turbine that arrives without its transformer is not generation.

For the procurement discipline this requires, including expediting against manufacturing milestones, see our guide to the industrial procurement process.

When does supply catch up

Not before the end of the decade on current disclosures. Manufacturing capacity expansions take years to commission and the order book ahead of them is already several years deep.

The Behind-the-Meter Response

Roughly 101 gigawatts of behind-the-meter natural gas generation has been announced in the United States, because waiting for the grid now costs more than building your own power.

Behind-the-meter means generation connected on the customer's side of the utility revenue meter, serving the site's own load rather than exporting to the grid.

What the market actually chose

Announced behind-the-meter capacity splits by technology as follows:

Gas turbines dominate at scale, reciprocating engines take a significant minority, and roughly 20 GW of announced turbine capacity has not disclosed whether it is heavy-duty, industrial or aeroderivative.

For how those technologies compare on capacity, efficiency, start time and cost, see our guide to power generation equipment compared.

How far the shift has gone

Bloom Energy's 2026 data centre survey found that expectations for fully on-site-powered facilities increased sharply over the preceding six months, and that more than a third of data centres are expected to use 100 percent on-site power by 2030. Roughly one in five campuses is expected to exceed 1 GW by 2030.

The economic test

The question is not grid or generation in the abstract. It is whether the cost of waiting exceeds the cost of owning power.

For a facility whose revenue starts when it energises, four years of delay has a calculable cost, and on an AI campus it frequently exceeds the entire capital cost of on-site generation.

For an industrial facility, the comparison is against avoided retail cost including demand charges rather than against wholesale price. See our guide to captive power for industrial facilities, and for full plant economics our guide to thermal power plant cost.

If gas prices rise

Fuel is the largest lifetime cost of a thermal plant, and a turbine plant's economics are more sensitive to fuel price than an engine plant's. Model across a price range and identify the crossover at which grid supply beats self-generation, then decide whether that crossover sits inside a plausible band.

Evaluating a proposal

Require the energisation date with its assumptions visible, the equipment delivery position with a purchase order reference, the permitting path with the applicable rule named, and the fuel supply arrangement. A proposal missing any of the four is a capability statement rather than a schedule.

For microgrid configurations and decentralised architecture, see our guide to microgrids and decentralisation.

What Actually Gets Built Fast

A Tennessee project deployed 200 megawatts of data centre load in 122 days using mobile natural gas generators, and that is the current benchmark for speed.

Four months from decision to energised load, against four to seven years in an interconnection queue.

Why mobile and modular wins on schedule

Factory-built, trailer-mounted or skid-mounted units arrive commissioned rather than constructed. Civil works are minimal. Multiple units install in parallel rather than in sequence. And capacity scales by adding units rather than by resizing a single machine.

The trade-off is efficiency and unit cost. Mobile and aeroderivative capacity costs more per kW and burns more fuel per MWh than a permanent combined cycle plant. On a project where revenue starts on energisation, that trade is usually correct.

The bridge-then-permanent pattern

The pattern now common on constrained sites is temporary or mobile capacity first, permanent plant or grid connection later, with the interconnection application filed on day one and running in parallel.

That sequence gets load energised while preserving the queue position, and it treats the two paths as complementary rather than as alternatives.

For rating classes, deployment models and contract structures on temporary capacity, see our guide to temporary and mobile power solutions.

What evidence proves a delivery claim

A named project of comparable scale, a stated elapsed time from instruction to energisation, and an owner who will take a call. A delivery claim without all three is a target.

Storage and Firming

Battery storage now serves a different purpose than it did five years ago: it firms variable generation and covers short-duration peaks, rather than replacing capacity.

Firming means converting variable output into a dependable profile the system can rely on. Duration is how many hours a storage system can discharge at rated power, and it is the metric that determines what a battery can actually do.

Lithium iron phosphate systems achieve approximately 95 percent round-trip efficiency and dominate recent utility-scale awards. Turnkey system cost has reached $110 to $117 per kWh in competitive markets.

What storage does and does not solve

Storage responds in milliseconds, has no emissions at the point of use and shifts energy across hours. It cannot cover a multi-day event, and a four-hour system is a four-hour system regardless of how much power it can deliver.

Against the current demand problem, storage firms and shifts. It does not add energy. A facility short of 200 MW of continuous supply needs generation, and storage alongside it improves the quality of what that generation delivers.

For sizing, chemistry selection, NFPA 855 compliance and cost, see our guide to battery energy storage systems.

Renewables and the Generation Mix

Renewables reached roughly 34 percent of global electricity in 2025 and overtook coal for the first time in over a century, and that shift continues independently of the demand surge.

Fossil fuels supplied approximately 55 percent of global electricity in 2025, renewables about 34 percent, and nuclear around 9 percent. Coal fell below a third of global generation for the first time on record.

Why renewables alone do not answer the demand problem

Three reasons, none of them about cost.

Shape. Solar and wind produce when the resource is available, and an AI data centre load is flat and continuous. Matching one to the other requires firming capacity, storage, or both.

The same queue. Renewable projects sit in the same interconnection queues as everything else and face the same waits.

Curtailment. Where transmission is constrained, output is reduced below what the resource could produce, which is a network problem rather than a generation one.

Renewables are growing quickly and they are not a fast answer to a load that needs to energise in eighteen months. The projects being announced to serve that load are overwhelmingly gas, which is a statement about schedule rather than about policy.

Permitting and Emissions

Adding generation triggers air permitting, and for new gas turbines in the United States the governing standard now depends on both size and capacity factor.

40 CFR Part 60 Subpart KKKKa applies to stationary combustion turbines constructed, modified or reconstructed after 13 December 2024. It sets NOx limits by turbine category and, for units above 850 MMBtu/hr heat input operating above a 45 percent capacity factor, requires selective catalytic reduction, which injects ammonia or urea across a catalyst to convert NOx into nitrogen and water.

Reciprocating engines are regulated under separate standards depending on engine type, size, fuel and whether the unit is new or existing.

What this changes practically

Permitting duration belongs in the schedule alongside equipment lead time, because on a fast-track project it is frequently the second-longest item.

Self-generation also moves emissions from Scope 2, purchased electricity, to Scope 1, direct combustion. Whether that is a net increase depends on the carbon intensity of the grid you would otherwise have drawn from, and it should be stated explicitly rather than discovered at the first reporting cycle.

If permitting blocks on-site generation

Some jurisdictions restrict new combustion capacity in non-attainment areas or impose emissions offsets. Establish the position before selecting the site, because the permitting envelope can eliminate the fastest technical option entirely.

What This Means by Sector

The constraint is common. What changes is which part of it binds first.

Data centres

Time to power outranks cost per MWh, and the interconnection wait is frequently longer than the entire development cycle. Behind-the-meter generation, bridging capacity and early queue filing in parallel are now standard practice rather than contingency. For redundancy requirements on the resulting infrastructure, see our guide to data centre power redundancy.

Manufacturing and industrial

Onshoring and electrification add load to networks already congested by data centre requests. Industrial facilities generally lack the capital intensity to outbid a hyperscaler for a queue position, which makes captive generation and demand management more attractive at a smaller scale.

Oil and gas

Field electrification, methane compliance and electric fracturing create power demand at sites with no service. The constraint there is not the queue but the absence of any network at all.

Healthcare

Load growth is modest, and the reliability requirement is absolute. The binding constraint is equipment lead time on standby and emergency systems rather than interconnection.

Metals, mining and process industries

Very large continuous loads concentrated in a few drives, frequently in remote locations. Where grid supply is available it is often already constrained, and where it is not, the comparison is against delivered fuel rather than a tariff.

Utilities

Serving the load while the queue lengthens, the equipment is unavailable and regulators intervene. Utility five-year peak forecasts more than tripling in a single year is the clearest evidence of how fast the planning problem changed.

Outside the United States

The demand pattern is similar and the mechanics differ. Connection regimes, grid codes, permitting and equipment allocation all vary, and with 60 percent of the gas turbine supply chain dedicated to the US market, availability outside North America is a separate question that must be asked directly of the OEM.

What To Do Now

Six actions, in this order, for any organisation that needs meaningful power before 2030.

File the interconnection application immediately. Queue position is chronological and the queue only lengthens. Filing costs a fraction of what a lost position costs, and it preserves the option even if you never use it.

Establish your real energisation date. Ask the utility for the current study timeline in your specific region rather than relying on a published average. A five-year national median is not your number.

Quantify the cost of waiting. Revenue deferred, contractual exposure, and opportunity cost. That figure is the budget available for the alternative, and on many projects it is larger than the alternative costs.

Secure equipment position before design freeze. With 2026 and 2027 turbine slots largely sold out and transformers at 90 to 130 weeks, availability now precedes configuration rather than following it.

Plan the bridge and the permanent solution together. Mobile or temporary capacity to energise, permanent plant or grid connection behind it, with the queue application running throughout.

Confirm the permitting envelope early. Air permitting can eliminate the fastest technical option, and finding that out after equipment commitment is expensive.

What to require from a partner

A named project of comparable scale with an owner who will take a call. A stated schedule to energisation with assumptions visible. Current equipment availability with purchase order evidence, not a catalogue. Permitting experience in your jurisdiction. And clarity on whether they will operate the asset afterwards and on what availability terms.

Capital planning and staffing

Behind-the-meter generation is capitalised and depreciated, which moves the decision into the capital cycle rather than the operating budget. It also creates an operating and maintenance requirement the facility did not previously have, which is a role and a competency rather than a line item. For what that scope contains, see our guides to rotating equipment maintenance and gas turbine outage planning.

What Prismecs Does

Prismecs delivers fast-track power generation projects, supplies ready-to-ship equipment, and operates the assets afterwards, which is the combination the current constraint requires.

Delivered project scope includes eight TM2500 dual-fuel units totalling 260 MW at Birr, Switzerland, delivered as a fast-track reserve plant online in six months with a new 220 kV interconnection on a compact site; four TM2500 units totalling 110 MW at Duqm, Oman, kept grid-ready with resident O&M crews, CMMS and parts support; an LM2500XPRESS plant at Miaoli, Taiwan delivered in ten months; three LM6000PC units adding 150 MW of fast-start reserve; an LM6000 fleet decommissioned in Norway, transported and recommissioned at a new site; and DC-coupled battery energy storage for solar and hybrid projects.

The relocated Norway fleet is the point worth isolating. Where new-build slots are sold out, a relocated and recommissioned asset is frequently the only route to a near-term in-service date, and it requires the ability to decommission, transport and recommission rather than simply to supply.

Apply this article's criteria to any partner, including us. Ask for a named project with an elapsed time from instruction to energisation. Ask what equipment is actually available and against what purchase order. Ask what permitting path is assumed. Ask who operates it afterwards.

To discuss equipment availability or a fast-track power project, send your required capacity, target energisation date, site location, fuel position and current interconnection status to sales@prismecs.com or call +1 (888) 774-7632.

Frequently Asked Questions

What is the biggest trend in power generation right now?

Demand has outrun supply. ERCOT is tracking approximately 474 GW of large-load interconnection requests, around 90 percent of them data centres. PJM waits approach seven years for large Northern Virginia campuses. Gas turbine delivery slots for 2026 and 2027 are largely sold out. The result is roughly 101 GW of announced behind-the-meter gas generation in the United States, built specifically to bypass the queue.

How much electricity do US data centres consume?

An estimated 176 terawatt-hours in 2025, forecast to reach between 420 and 728 TWh by 2030 depending on interconnection success and AI adoption. That is roughly 300 TWh of incremental demand, equivalent to the combined annual consumption of New York and California, and the largest five-year surge to the US grid since the 1980s. EPRI has estimated data centres at 4.6 to 9.1 percent of all US electricity.

How long is the interconnection wait now?

Lawrence Berkeley National Laboratory research finds interconnection wait times have more than doubled over fifteen years, with projects averaging five years in queue before commercial operation. In primary data centre markets average waits exceed four years, and PJM approaches seven years for large Northern Virginia campuses. Grid interconnection timelines of 36 to 84 months are structurally incompatible with data centre build cycles of 12 to 24 months.

Why did interconnection get so much slower?

Three compounding forces. Demand, as data centres, electrification and new manufacturing flooded utilities with large-load requests simultaneously. Equipment, because required network upgrades face the same 90 to 130 week lead times as customer substations. And system headroom, because the regions developers want have less spare transmission capacity than incoming demand requires. PJM reported 29 percent of milestone change requests driven by permitting and 23 percent by supply chain.

Are gas turbines actually unavailable?

Delivery slots for 2026 and 2027 are largely sold out according to GE Vernova's Q1 earnings disclosure. Approximately 60 percent of the gas turbine supply chain is dedicated to the US market, with data centres expected to take a third of output beyond 2027, implying an upper limit of around 11.4 GW available to US data centres in 2028. Transformers, breakers and conductor run 90 to 130 weeks.

How much behind-the-meter generation has been announced?

Approximately 101 GW in the United States as of 2026, split by technology as roughly 70 GW gas turbine, 18 GW reciprocating engine, 13 GW with undisclosed equipment class and 3 GW mixed. Bloom Energy's 2026 survey found more than a third of data centres expect to use 100 percent on-site power by 2030, with about one in five campuses exceeding 1 GW.

How fast can on-site power actually be deployed?

The current benchmark is 200 megawatts of data centre load energised in 122 days using mobile natural gas generators, on a Tennessee project. That is roughly four months against four to seven years in an interconnection queue. Factory-built mobile and skid-mounted units arrive commissioned rather than constructed, install in parallel, and scale by adding units rather than resizing a single machine.

Should I wait for the grid or build my own power?

The test is whether the cost of waiting exceeds the cost of owning power. Quantify revenue deferred, contractual exposure and opportunity cost across the real energisation date, then compare it to the capital and operating cost of generation. On projects where revenue starts at energisation, four years of delay frequently exceeds the entire capital cost of on-site capacity.

Should I still file for interconnection if I build my own generation?

Yes, and file immediately. Queue position is chronological and the queue only lengthens, so filing preserves an option that costs a fraction of what a lost position costs. The pattern now common on constrained sites is bridging or mobile capacity to energise load, permanent plant behind it, and the interconnection application running in parallel throughout.

Can renewables meet the new demand?

Not on the timescale the demand requires, for three reasons unrelated to cost. Shape, because solar and wind produce when the resource allows while AI load is flat and continuous. The same queue, because renewable projects wait alongside everything else. And curtailment where transmission is constrained. Renewables reached roughly 34 percent of global electricity in 2025 and overtook coal, and the capacity announced to serve new AI load is overwhelmingly gas.

What role does battery storage play in this?

Firming and short-duration peak coverage, not capacity replacement. Lithium iron phosphate systems achieve around 95 percent round-trip efficiency at $110 to $117 per kWh turnkey in competitive markets. Storage responds in milliseconds and shifts energy across hours, but a four-hour system cannot cover a multi-day event. A facility short of continuous supply needs generation, with storage improving what that generation delivers.

What emissions rules apply to new on-site gas generation?

40 CFR Part 60 Subpart KKKKa applies to stationary combustion turbines constructed, modified or reconstructed after 13 December 2024, setting NOx limits by category and requiring selective catalytic reduction for units above 850 MMBtu/hr operating above a 45 percent capacity factor. Reciprocating engines fall under separate standards. Permitting duration belongs in the schedule alongside equipment lead time.

Is the data centre demand forecast a bubble?

Partly overstated and not resolved by that. Against 16 GW of announced 2026 capacity, only 5 GW is currently under construction, and announcements are not projects. But even the discounted pipeline exceeds what the grid and equipment supply chain can deliver, which is why queue length and lead times are the operative facts rather than the forecast itself.

What should I do first if I need power before 2030?

File the interconnection application immediately to secure queue position. Establish your real energisation date from the utility rather than a national average. Quantify the cost of waiting, which becomes your budget for alternatives. Secure equipment position before design freeze, since availability now precedes configuration. Plan the bridge and the permanent solution together. And confirm the permitting envelope before committing to equipment.

Tags: Power Generation Trends Data Center Power Demand Interconnection Queue Behind the Meter Generation Equipment Lead Times