Energy-first campus development means securing firm power on a contracted date before committing capital to land, shell, or fit-out. It leads site selection because US data center power demand is projected to rise from 31 GW in 2025 to 66 GW in 2027 (Goldman Sachs Research, US data center power demand, May 2026) while large-load grid connections average 4.4 years (Cushman & Wakefield, Global Data Center Market Comparison, May 2026).
What does “energy-first” mean, and why did power replace location as the source of site value?
For two decades a data center site was valued on fiber, proximity to users, tax treatment, and land cost; power was assumed. That has inverted. Power replaced location as the source of site value, roughly 50 GW of behind-the-meter gas generation was announced in 2025, and secured turbine slots now trade as an asset at about three times 2019 prices (StackedAI analysis, Energy-First market review, Jul 2026).
Energy-first development reverses the order of operations. The developer first secures a firm power path with a date attached: a utility service agreement with a defined energization date, a permitted on-site plant with equipment slots in hand, or both. Only then does capital go into land, shell, and fit-out.
The reason is financial: a data center earns nothing until energized, and every month between capital commitment and energization is carried at the project’s cost of capital. Power availability has been the top challenge cited by data center investors for the third consecutive year (CBRE, 2026 NA Data Center Investor Intentions Survey, Apr 2026). A site with dated, firm power is worth more than a better-located site without it, the premise of StackedAI’s Tier-2 brownfield conversion thesis, where an existing interconnect is often the most valuable thing on the property.
How bad is the grid constraint for data centers in 2026?
Demand: US data center power demand is projected to rise from 31 GW in 2025 to 41 GW in 2026 and 66 GW in 2027, and only 50 to 60% of scheduled capacity comes online on time (Goldman Sachs Research, US data center power demand, May 2026). US peak load is projected to grow 166 GW by 2030, roughly 90 GW of it data centers (Grid Strategies, National Load Growth Report 2025, Dec 2025), with ERCOT growing about 10% per year (EIA, Today in Energy, Mar 2026).
Supply: the median generation project completed in 2025 took more than 5 years from request to operation, and only 13% of requests filed between 2000 and 2020 ever reached operation (LBNL, Queued Up: 2026 Edition, May 2026). More than 2.2 TW sits in queues (RMI, Interconnection reform for AI data centers, Mar 2026).
Loads: new large-load requests average 4.4 years to power delivery (Cushman & Wakefield, Global Data Center Market Comparison, May 2026), AI campus interconnection runs 24, 36, or 48+ months by market (CBRE, 2026 US Outlook: Data Centers, Jan 2026), and ComEd territory in Chicago faces delays until 2032 or later (CBRE, NA Data Center Trends H2 2025: Chicago, Feb 2026).
Policy now moves with physics: after Texas paused new data center development on August 3, 2026, the EIA cut its Texas 2027 load growth forecast from 14% to 6% (EIA, Short-Term Energy Outlook, Aug 2026). When the fastest market pauses, a queue-independent power path gains value everywhere.
What behind-the-meter power options can a data center campus actually deploy?
Behind-the-meter (BTM) power is generation on the customer’s side of the utility meter, serving the facility directly. The dominant architecture is grid plus BTM bridge: on-site generation carries load from day one and shifts to peaking or demand response once the utility delivers.
| Option | Role | Time to power | Permitting posture | Fit for 7–25 MW |
|---|---|---|---|---|
| Utility grid | Long-term primary supply | 4.4 yrs avg (Cushman & Wakefield, May 2026); ComEd 2032+ (CBRE, Feb 2026) | Tariff and large-load rules | End state; rarely on the required date |
| Gas turbines | Prime power, 100 MW+ | Slots sold out through 2030–31 (StackedAI analysis, Jul 2026) | Full air permit, often Title V scale | Too large per unit; poor M+1 granularity |
| Reciprocating gas engines | Prime power and bridge below 100 MW | From ~1 MW, marketed as bypassing 3–7 yr queues (vendor claim, unverified; StackedAI analysis, Jul 2026) | Engine-scale permit; NSPS applies | Best fit; unit size matches load steps |
| Fuel cells | Prime power where emissions matter | Bloom Q2 2026 revenue +165.5% (Bloom Energy, Jul 2026) | Lighter combustion burden | Viable; modular, higher capital per kW |
| Small modular reactors | Long-dated firm supply | No dated data center deployments in cited sources | Federal licensing | Option value, not a bridge |
| Battery storage (BESS) | Ride-through, load shaping, reserve | Fast; paired with engines | Fire code review | Reduces spare-unit count |
| Diesel generators | Emergency backup, black start | Fast | Emergency permits limit run hours | Backup only; never prime power |
Diesel stays emergency-only despite cheap gas: gensets start in seconds and need no pipeline, but permits restrict them to outages plus limited testing hours, on-site fuel is finite, and diesel fuel cost per kWh is far above pipeline gas. The turbine versus engine choice is developed in gas turbine vs reciprocating engine for data center power.
What is M+1 redundancy for on-site generation, and how does it differ from N+1?
M is the number of generating units required to carry the critical load at rated output. M+1 holds one spare unit above that number, so any single unit can be out for scheduled maintenance, or trip unexpectedly, while the remaining units carry full load. M+2 holds two spares.
N+1 applies the same logic on the distribution side: N UPS modules, chillers, or paths plus one spare. The two sides fail differently: a UPS module is either in service or not, while a gas engine or turbine has an OEM maintenance calendar that removes it from service for days or weeks at planned intervals. A plant that cannot carry load with one unit down for service is not firm supply, which is why maintainability, not fault tolerance alone, drives M+1.
The arithmetic favors smaller units. A 20 MW critical load on 5 MW engines needs M = 4 and M+1 = 5 units, 25 MW installed and 20% spare; on 10 MW machines it needs M = 2 and M+1 = 3 units, 30 MW installed and 50% spare. Battery storage can shrink the spare further by carrying load through a trip while a standby unit starts. The worked example is in M+1 redundancy for on-site power.
Why are turbines and transformers the bottleneck, and what does that do to cost?
The supply chain sets the pace of BTM development in 2026. Turbine prices are expected to reach about $600 per kW by the end of 2027, up 195% from 2019, with orders of roughly 100 GW against manufacturing capacity of 60 to 70 GW per year (Wood Mackenzie via Bloomberg, Gas turbine prices surge, Apr 2026), and GE Vernova’s backlog reached 116 GW in Q2 2026 (GE Vernova, Q2 2026 results, Jul 2026). Transformers gate the grid and BTM paths alike, with average lead times of 120 weeks in 2024 (Wood Mackenzie, Power transformer lead times, Apr 2024); large power transformers and medium-voltage switchgear are the critical-path items on any campus schedule (StackedAI analysis, Energy-First BTM cost model, Jul 2026).
| Item | Figure | Source |
|---|---|---|
| Gas turbine price | ~$600/kW by end-2027; +195% vs 2019 | Wood Mackenzie via Bloomberg, Apr 2026 |
| Turbine orders vs capacity | ~100 GW ordered vs 60–70 GW/yr | Wood Mackenzie via Bloomberg, Apr 2026 |
| GE Vernova turbine backlog | 116 GW (Q2 2026); ≥125 GW by YE2026 | GE Vernova, Jul 2026 |
| Turbine slot resale premium | ~3x 2019 prices | StackedAI analysis, Energy-First market review, Jul 2026 |
| Turbine slot availability | Sold out through 2030–31 | StackedAI analysis, Tier-2 conversion timelines, Jul 2026 |
| Power transformer lead time | 120 wks avg (2024); 80–210 wks large units | Wood Mackenzie, Apr 2024 |
What does behind-the-meter gas power cost per kWh compared with the utility?
Using Henry Hub and Houston Ship Channel gas at roughly $2.9 to $3.0 per MMBtu and a heat rate of about 7,400 Btu/kWh (about 46% thermal efficiency), fuel cost is about 2.2 cents per kWh; with operations, maintenance, and consumables, all-in operating cost at high utilization is roughly 3.5 to 5 cents per kWh (StackedAI analysis, Energy-First BTM cost model, Jul 2026). The derivation is 7,400 Btu/kWh multiplied by $3.0 per million Btu, or $0.0222 per kWh.
StackedAI’s Tier-2 benchmarks put blended utility rates at about 9.25 cents per kWh in Northern Virginia, 7.15 in Dallas-Fort Worth, 7.77 in Salt Lake City, and 8.5 to 9.5 in Columbus (StackedAI analysis, Tier-2 market benchmarks, Apr 2026). On operating cost alone, BTM gas undercuts all of them.
That comparison flatters BTM: the 3.5 to 5 cent figure excludes capital recovery, permitting, land, and diesel backup, while the utility rate includes the utility’s capital, delivery, and reserve margin. BTM gas is competitive on a levelized basis at high utilization and cheap gas, and decisively better on time to revenue; the full build-up is in behind-the-meter power cost for data centers.
What tariffs and air permits gate a BTM project?
AEP Ohio’s data center tariff requires a minimum take of 85% of contracted capacity for 12 years; Georgia’s Public Service Commission and Oregon’s Schedule 96 have their own templates; and Texas Senate Bill 6 imposes curtailability on large loads during grid emergencies (StackedAI analysis, large-load tariff template review, Jul 2026). Minimum-take terms turn a grid connection into a long-dated fixed obligation, and curtailability means the grid is not firm during scarcity; both push developers toward on-site generation.
Air permitting is the gating item for the plant and usually runs longer than engine lead times. Conceptually, three regimes apply. Title V is the federal operating permit for major sources; whether a plant crosses that line depends on potential to emit, a function of unit count, run hours, and controls. New Source Review is the preconstruction program: Prevention of Significant Deterioration in attainment areas, nonattainment NSR with offsets and stricter controls elsewhere. New Source Performance Standards set unit-class limits for stationary turbines and reciprocating engines. Source aggregation rules can combine adjacent generation under common control into one source. Thresholds vary by pollutant and jurisdiction; the permit strategy is set with the unit count.
Does behind-the-meter power make a site more or less financeable?
It depends on whether lenders treat the generation as an asset or a risk, which turns on four items: firm gas transportation with tenor matching the debt; an OEM long-term service agreement; a permit for prime rather than emergency-only operation; and a credit tenant behind the revenue.
The debt market prices those items directly. Top-credit projects price at spreads in the low 200 basis points with leverage up to 85% loan-to-cost, while non-credit tenants pay 200 to 300 basis points more and lever at 70 to 80% (JLL, NA Data Center Report Midyear 2026, Aug 2026). StackedAI’s Tier-2 archetype is 60/40 debt-to-equity at 7.5% with an anchor tenant and 55/45 at 9% without (StackedAI analysis, Tier-2 conversion underwriting framework, Jul 2026). BTM that shortens the path to an anchor lease improves terms; BTM that adds merchant fuel exposure and permit risk without a tenant worsens them. Sponsor modeling of that trade-off is covered in how private equity underwrites data center power risk.
How does a 7–25 MW Tier-2 site use BTM as a bridge?
StackedAI’s buy box is 7 to 25 MW sites in secondary US markets with a conversion timeline of 12 to 18 months versus 36 or more for greenfield (StackedAI analysis, Tier-2 conversion underwriting framework and conversion timelines, Jul 2026). At that scale BTM is a bridge, not a campus plant.
The sequence: retain the existing utility interconnect, usually the most valuable asset on a brownfield site; file for the incremental load; and install reciprocating gas engines with battery storage in M+1 to carry the increment until the utility delivers. Prime-power engine and BESS packages from about 1 MW are marketed as bypassing the 3 to 7 year queue, a vendor claim StackedAI has not independently verified (StackedAI analysis, Energy-First market review, Jul 2026). When the utility arrives, the engines become peaking and curtailment capacity, which tariffs such as Texas SB6 reward. This is the speed-to-power lane where smaller operators beat hyperscalers (StackedAI analysis, Tier-2 conversion underwriting framework, Jul 2026). The cost trade-off against a new build is in brownfield vs greenfield data center cost per MW.
Key terms
- Behind-the-meter (BTM) power: generation on the customer’s side of the utility meter that serves the facility directly.
- M+1 redundancy: a generation standard holding one spare unit above the number (M) required to carry critical load.
- N+1 redundancy: the distribution-side standard holding one spare component above the number (N) required.
- Heat rate: fuel energy consumed per unit of electricity produced, in Btu per kWh; lower is more efficient.
- Title V permit: the federal operating permit required for major stationary sources of air pollutants.
- Minimum-take tariff: a large-load tariff requiring payment for a fixed share of contracted capacity for a set term regardless of use.
How StackedAI applies this
StackedAI screens every Tier-2 acquisition target on its power path before its real estate, ranking sites by the date on which firm power can be contracted. For sites inside the 7 to 25 MW buy box, the firm models a reciprocating-engine-plus-storage bridge in M+1 against the utility’s delivery date and tariff, then tests financeability against current debt terms, benchmarking fuel, heat rate, and utility rates per engagement. The analyses feed the work on the advisory services page and the structures on data center transaction structures.
Frequently asked questions
How long does a grid connection take for a data center in 2026?
New large-load requests average 4.4 years to power delivery (Cushman & Wakefield, May 2026), the median generator interconnection exceeds 5 years (LBNL, May 2026), and ComEd territory in Chicago faces delays until 2032 or later (CBRE, Feb 2026). Timelines vary by market; StackedAI benchmarks them per engagement.
What is M+1 redundancy for on-site generation?
M is the number of generating units required to carry the critical load. M+1 adds one spare unit so any single unit can be out for scheduled maintenance, or trip unexpectedly, while the plant still carries full load. It is the generation-side counterpart to N+1 on the distribution side, and maintainability drives it.
What does behind-the-meter gas power cost per kWh?
With gas at roughly $2.9 to $3.0 per MMBtu and a heat rate of about 7,400 Btu/kWh, fuel alone is about 2.2 cents per kWh, and all-in operating cost at high utilization is roughly 3.5 to 5 cents per kWh (StackedAI analysis, Jul 2026). That excludes capital recovery, permitting, land, and backup.
Why are gas turbines so hard to get?
Orders of about 100 GW are running against manufacturing capacity of 60 to 70 GW per year, pushing prices toward $600 per kW by end-2027, up 195% from 2019 (Wood Mackenzie via Bloomberg, Apr 2026). GE Vernova’s backlog reached 116 GW in Q2 2026, and secured delivery slots trade at roughly three times 2019 prices.
Does behind-the-meter power make a data center site more financeable?
It can, when the plant is permitted, fuel supply is contracted, O&M sits under an OEM service agreement, and a credit tenant is in place; lenders then treat the generation as an asset that shortens time to revenue. Unpermitted plants, interruptible gas, or unproven technology push a project toward wider spreads and lower leverage.
Is diesel a viable primary power source for a data center?
No. Diesel generators remain the standard emergency backup because they start fast and need no pipeline, but their permits restrict them to emergency operation plus limited testing hours, fuel storage is finite, and fuel cost per kWh is far above pipeline gas. Prime power comes from gas engines, turbines, or fuel cells.
Sources
- Goldman Sachs Research, US data center power demand projected to double by 2027, May 2026, https://www.goldmansachs.com/insights/articles/us-data-center-power-demand-projected-to-double-by-2027
- Lawrence Berkeley National Laboratory (LBNL), Queued Up: 2026 Edition, May 2026, https://emp.lbl.gov/publications/queued-2026-edition-characteristics
- RMI, Interconnection reform for AI data centers and generator queues, Mar 2026, https://rmi.org/resources/interconnection-reform-ai-data-centers-generator-queues/
- Grid Strategies, National Load Growth Report 2025, Dec 2025, https://gridstrategiesllc.com/wp-content/uploads/Grid-Strategies-National-Load-Growth-Report-2025.pdf
- EIA, Today in Energy, Mar 2026, https://www.eia.gov/todayinenergy/detail.php?id=67344
- EIA, Short-Term Energy Outlook, Aug 2026, https://www.eia.gov/outlooks/steo/
- Cushman & Wakefield, 2026 Global Data Center Market Comparison, May 2026, https://ir.cushmanwakefield.com/news/press-release-details/2026/Dallas-Texas-Ranked-No–1-Primary-Data-Market-in-the-World-as-AI-Demand-Power-Constraints-and-Regulation-Reshape-CRE-Strategy/default.aspx
- JLL, North America Data Center Report Midyear 2026, Aug 2026, https://www.jll.com/en-us/insights/market-dynamics/north-america-data-centers
- CBRE, 2026 US Real Estate Market Outlook: Data Centers, Jan 2026, https://www.cbre.com/insights/books/us-real-estate-market-outlook-2026/data-centers
- CBRE, North America Data Center Trends H2 2025: Chicago, Feb 2026, https://www.cbre.com/insights/books/north-america-data-center-trends-h2-2025/chicago-data-center-market
- CBRE, 2026 North American Data Center Investor Intentions Survey, Apr 2026, https://www.cbre.com/insights/briefs/2026-north-american-data-center-investor-intentions-survey
- Wood Mackenzie via Bloomberg, Gas turbine prices surge, Apr 2026, https://www.bloomberg.com/news/articles/2026-04-01/gas-turbine-prices-surge-crimping-efforts-to-power-data-centers
- GE Vernova, Second Quarter 2026 Financial Results, Jul 2026, https://www.gevernova.com/news/press-releases/ge-vernova-reports-second-quarter-2026-financial-results-raises-2026-financial
- Bloom Energy, Second Quarter 2026 Financial Results, Jul 2026, https://investor.bloomenergy.com/press-releases/press-release-details/2026/Bloom-Energy-Reports-Record-Second-Quarter-2026-Financial-Results-and-Raises-Full-Year-2026-Guidance/default.aspx
- Wood Mackenzie, Supply shortages and an inflexible market give rise to high power transformer lead times, Apr 2024, https://www.woodmac.com/news/opinion/supply-shortages-and-an-inflexible-market-give-rise-to-high-power-transformer-lead-times/
- StackedAI analysis, Energy-First BTM cost model, Jul 2026 (internal)
- StackedAI analysis, Energy-First market review, Jul 2026 (internal)
- StackedAI analysis, Tier-2 market benchmarks, Apr 2026 (internal)
- StackedAI analysis, Tier-2 conversion timelines, Jul 2026 (internal)
- StackedAI analysis, Tier-2 conversion underwriting framework, Jul 2026 (internal)
- StackedAI analysis, large-load tariff template review, Jul 2026 (internal)