Behind-the-meter versus grid-tied site economics: the numbers in 2026

The question every US site developer is asked at some point in 2026 is: what does the power actually cost on this site, and over what tenor. The answer everyone wants is one number. The answer that matters is six numbers — and which one binds depends on the offtaker's clock. This is the numerical companion to Time to power: the math behind the assertion that the binding variable is no longer $/MWh, it is the product of $/MWh and the certainty that those MWh arrive when the load needs them.

Numbers here are indicative ranges drawn from Lazard's LCOE+ June 2025, LevelTen's Q1 2026 PPA Price Index, and a sample of disclosed BTM project deals. They are correct as orders of magnitude, not as quotes for any specific site. The pricing for any individual project will depend on resource quality, site cost basis, offtaker credit, tenor, and structure.

What grid-tied PPAs actually cost in mid-2026

LevelTen's Q1 2026 PPA Price Index landed at average $64.49/MWh for solar and $79.40/MWh for wind — both at record levels, both still rising. Solar PPA prices rose 4.7% in Q1 2026 and roughly 13% year over year. Wind rose nearly 8% in Q1 and about 24% year over year. Those are weighted national averages.

The regional split tells the operational story.

These are clearing prices for long-tenor (12–15 year) corporate PPAs against utility-scale projects already moving toward begin-construction. They are not what a 2026-vintage start gets, because OBBA's wind and solar phase-out caps new pipeline that does not begin construction by 5 July 2026 (see Cell 1.3). The clearing prices on the next vintage of wind and solar PPAs — for projects that fall outside the credit window — will be materially higher.

Practical implication for a site developer: if your offtaker is shopping a grid-tied PPA in 2026, the price they will be benchmarking against is rising fast, the project they will be benchmarking against may not exist by the time they need it, and the time to power on the underlying generation is still queue-bound (3–7 years per Cell 4.3).

What BTM costs in mid-2026

Behind-the-meter economics split by technology. Three technologies dominate current deployment for data centre and large industrial load.

Natural gas reciprocating engines (BTM)

The fastest path to firm power on site. Deployment timeline: 12–24 months once gas connection and air permit are secured.

Indicative all-in $/MWh: $80–130/MWh for new-build BTM gas reciprocating, depending on gas basis, utilisation, and project scale. Lazard's June 2025 LCOE+ pegs gas combined cycle at $48–109/MWh and gas peaking at $149–251/MWh. BTM reciprocating typically lands between the two — built to run base-load behind a 24/7 load, not as a peaker, so utilisation is high; but capex per kW and O&M are higher than utility-scale CCGT.

Recent project markers:

• INNIO + VoltaGrid + Oracle: 2.3 GW across 92 × 25 MW power packs.

• Wärtsilä Ohio: 412 MW from 40 × 34SG spark-ignited engines.

• Wärtsilä US data centre engine deployment total: past 1.6 GW.

• Nscale Monarch Compute (West Virginia): 2 GW of Caterpillar G3500 reciprocating engines.

• Cleanview tracks 46 US BTM data centre projects representing 56 GW of planned capacity.

Natural gas turbines (BTM, simple and combined cycle)

Slower than reciprocating engines for the equipment lead time (24–36 months from order in current market conditions; aeroderivative units faster). All-in $/MWh: $70–110/MWh for combined-cycle BTM at scale; $90–130/MWh for simple-cycle aeroderivatives. Better thermal efficiency than reciprocating engines, worse modularity.

Solar + storage (BTM, co-located)

24–36 months to first power for typical configurations. All-in firm-equivalent $/MWh: $90–160/MWh, depending on solar resource, storage sizing, and the share of load the system is asked to firm. Storage costs continued to decline in 2025 — Lazard noted "notable declines" in standalone and hybrid BESS LCOS.

The trap on solar+storage for data centre BTM is that the firmness calculation is non-trivial. A system sized to meet 80% of a 24/7 load at 90% reliability is materially more expensive than the LCOE headline implies, because the back-end of the duration curve carries the cost. Worth modelling explicitly, not estimating from LCOE.

Advanced nuclear / SMR (BTM)

Time to power: 5–8 years for first units; project economics still pre-commercial for most SMR vendors. Indicative $/MWh: $80–140 for projects benefitting from §45U / §45Y eligibility, materially higher without. The OBBA retained nuclear credit eligibility, which is the only reason these numbers are even in the conversation.

For most 2026-vintage site developer decisions, SMR is a 2028+ deployment option, not a near-term option. Worth including in long-horizon planning, not in the next quarter's site economics decision.

Geothermal (BTM and grid-tied)

Where the resource exists, very attractive: $60–110/MWh range with high capacity factors and firm output. Lazard noted limited public data for new-build geothermal, so ranges are wider than reported. Time to power: 36–60 months for next-generation closed-loop or enhanced systems; faster for sites that can use existing reservoir data.

Site-specific. Most US developers will not have a geothermal-suitable site; those that do should treat it as a structural advantage.

The comparison matrix

Three observations on the matrix.

First, the $/MWh range on BTM is structurally above the grid-tied range for energy-only comparisons. This is correct and expected. The headline LCOE comparison consistently favours grid-tied because grid-tied amortises capex across a larger asset base. The grid-tied number stops mattering when the grid path cannot deliver inside the offtaker's window.

Second, the capacity factor column matters more than the $/MWh column for 24/7 load. A 90% capacity-factor BTM gas system at $100/MWh delivers more useful firm power than a 25% capacity-factor solar PPA at $50/MWh — even before you account for the cost of the firming you'd need to build around the solar PPA to make it serve a 24/7 load.

Third, the regulatory column will reorder the matrix over the next 36 months as the IRS publishes Material Assistance Cost Ratio safe harbour tables, as wind and solar pipelines fall outside the credit window, and as nuclear and geothermal credit certainty is tested against actual project economics.

The variable people miss: time-adjusted cost

The number that determines which path wins for a specific site is not LCOE. It is time-adjusted cost to load — the present value of the power, discounted by the probability that it arrives in the window the offtaker can transact on.

Stylised example. A 200 MW data centre offtaker has a 24-month deployment window. Two paths on the same site:

Path A — grid-tied solar PPA at $55/MWh, 15-year tenor, 3-year time to power. Headline LCOE: lowest in the matrix. Real economics: the offtaker either delays its build by 12 months (cost: $50–100M in lost revenue, depending on workload) or finds another power solution and the developer's site option lapses unused.

Path B — BTM gas reciprocating at $105/MWh, 15-year tenor, 18-month time to power. Headline LCOE: roughly 90% higher. Real economics: the offtaker's data centre comes online when planned. Lifetime cost differential vs. Path A on a static basis: ~$50/MWh × 200 MW × 8,760 hr × 0.85 capacity factor × 15 years ≈ $1.1 billion. Time-shift value to the offtaker (revenue brought forward by 12 months): plausibly $500M–$1B depending on load economics.

The two paths are roughly comparable for a 24-month-window offtaker. For a 36-month-window offtaker, Path A wins. For a 12-month-window offtaker, Path A doesn't exist as an option.

This is the calculation that should be running on every site under control in 2026. Most developer pitch decks are running it implicitly, in one direction, against a single offtaker profile, with no scenario stress. The discipline is to run it for the credible offtaker profiles your site can actually attract, and to underwrite the site option against the path the highest-probability matched offtaker actually needs.

What the matching layer does to this calculation

A site that can credibly support both paths (grid and BTM) is worth more than a site locked into one. The matching layer's job is to:

1. Identify the offtaker profile whose deployment window and price tolerance actually fit the site's available paths.

2. Pre-qualify the technology counterparty on each path against bankable performance data, OBBA cohort eligibility, and PFE position.

3. Pre-qualify the capital structure on each path against the offtaker's credit and tenor expectations.

4. Run the time-adjusted cost calculation explicitly, showing the offtaker which path wins under their constraints — not which path looks best on a static LCOE chart.

A developer running through a matching layer arrives at the offtaker with a path-selected, counterparty-pre-cleared, capital-structured deal. A developer running through traditional channels arrives at the offtaker with three LOIs and an LCOE spreadsheet, and watches two of the three die for reasons that show up at lender review.

What to do with this matrix this quarter

If you run a US site portfolio:

Pull each live option through the matrix. For each site, classify the credible technology paths, time-to-power on each, and indicative $/MWh range. The exercise takes a day per portfolio of ten sites. The output is a reorder of which options to extend, which to let lapse, and which to push into the matching gate now.

Run the time-adjusted cost calculation on the top three sites. Use the offtaker profile you've actually been approaching, not a generic hyperscaler. The number that matters is whether your site beats the alternative the offtaker has in hand — not whether it beats a national LCOE average.

Identify which OBBA cohort your most likely technology path falls into. If your live options sit on solar+storage with a 2028+ in-service date, the credit math has changed since 2024. If your options sit on firm-power (geothermal, gas, BTM storage), the credit math has improved. Either way, the underwriting model needs a 2026 input, not a 2024 input.

Bring the path-classified options into a matching conversation. The application gate is at sitepower.ai/apply/site-developers (or /nz for Australasian sites). No developer-side fee. The matching either closes or it doesn't, fast.