Behind-the-meter gas microgrids: a firm-power route around the interconnection queue

TL;DR

• The behind-the-meter gas microgrid is a firm-power route around the 4-7 year interconnection queue for AI datacenters. Enchanted Rock is the canonical operator.
• The class underwrites on two sources: a firm-power premium (ISO capacity + ancillary + firming savings, $150-280/kW-year), and a queue-avoidance option worth billions to hyperscalers racing frontier compute to market.
• Meta El Paso (~366 MW, utility bridge-power deal with EPE) and Microsoft San Jose (60 MW Phase 1 RNG-backed microgrid) are the flagship deployments, with roughly 1,500 MW aggregated across announced, contracted, and operational sites across other US hyperscaler-adjacent locations (operational share materially smaller than the aggregate; underlying project detail in Enchanted Rock S-1 disclosures).
• Installed capex $1,200-1,800 per kW. Levered IRR 11-19% depending on ISO. EaaS structures let hyperscalers avoid capex while contracting for 15 years.
• The bear case is emissions permit tightening (CA already strict, VA and IL tightening on 24-month lag) and RNG feedstock scarcity. Both matter through 2028.

01Why now

Interconnection queues in PJM, ERCOT, MISO, and CAISO stretched to 4-7 years for gigawatt-scale campuses between 2024 and 2026. Meanwhile, hyperscaler AI compute demand accelerated on 18-month silicon refresh cycles that cannot wait.

Behind-the-meter gas microgrids emerged as one of the few architectures that can deliver substantial firm MW inside 12-24 months without joining the conventional bulk interconnection queue. Meta's ~366 MW Enchanted Rock deployment in El Paso and Microsoft's San Jose RNG microgrid are the flagship deals. Roughly 1,500 MW of announced, deployed, or contracted capacity across US hyperscaler and data-centre applications is in the market through mid-2026 (Enchanted Rock disclosures).

Standby diesel gensets ran for decades as ride-through insurance, priced at roughly $50 to $100 per kW-year in wholesale reserve markets. The economic case for continuous or extended-duration BTM gas required three things to align: gas turbine and reciprocating engine economics at hall scale, RNG availability as an emissions-permit strategy, and an EaaS contract structure that let hyperscalers avoid on-balance-sheet capex. All three fell into place between 2022 and 2024.

02Canonical boundary

Architecture taxonomy

The generation architectures below share hardware but differ on economic function.

BTM prime power. Supply firm MW behind the meter as the primary generation source for the site load.

Bridge power. Get the customer through the interconnection gap while permanent grid infrastructure completes. Utility or customer ownership; front-of-meter or BTM configuration.

Grid-support microgrid. Supply the site plus provide dispatchable grid services under ISO or utility parallel-operation agreements.

Resiliency microgrid. Backup and continuity only. Traditional standby duty.

This essay covers everything above except the resiliency-only case, where the generation asset's economic role is limited to backup duty.

The three clocks

The class thesis reduces to a diagnostic across three schedules.

Compute clock. When the hyperscaler needs firm MW at the site to bring frontier compute revenue online.

Grid clock. When the utility can deliver equivalent firm interconnected bulk-grid capacity to the site.

Generation clock. When an alternative generation architecture can be permitted, procured, and commissioned on site.

The economic opportunity opens when the compute clock beats the grid clock. The architecture whose generation clock closes that gap fastest captures the largest share of the rent. BTM gas wins when its generation clock beats the grid clock. It loses when permitting, fuel, and equipment lead times push the generation clock past the grid clock.

03The two-source thesis

The BTM gas microgrid earns value from two distinct sources. A firm-power premium payable by ISO markets and utilities, and a queue-avoidance option worth billions to a hyperscaler that gets frontier compute to market earlier.

The firm-power premium is direct and monetisable. An ISO-registered, dispatchable MW earns capacity-market revenue in PJM (RPM auction), ISO-NE, and NYISO, while ERCOT monetises equivalent scarcity value through ORDC and ancillary services; a forward capacity auction sits elsewhere. Capacity or scarcity-equivalent revenue runs $2-6 per kW-month depending on ISO. For a 100 MW asset, that alone produces $2.4-7.2M of annual gross revenue. Ancillary services (frequency regulation, spinning and non-spinning reserves) add another $12-48/kW-year in markets where the asset can register. Utility firming savings from avoiding demand charges during grid peaks add another $6-14/kW-month. The resulting underwriting range for monetisable firm-power value at the operator level is roughly $150-280 per kW-year, dominated by the EaaS take rate that captures most of the underlying value stack. The listed capacity, ancillary, and firming components are alternative or partially overlapping monetisation routes captured within that range; additive streams above it would double-count.

The queue-avoidance value is where the class economics compound. For a hyperscaler racing frontier compute to market, connecting 100-300 MW of AI capacity 24 months earlier is worth billions in terminal enterprise value. At $2 per GPU-hour and 1,500 GPUs per MW, one avoided year on 100 MW represents roughly $2.6B of theoretical gross GPU-revenue capacity before utilisation, power, cooling, networking, software, and other operating costs. Enterprise value is a small share of that after operating costs and financing. The hyperscaler does not need to share anything close to $2.6B with the microgrid operator for the queue-avoidance option to support a substantial EaaS premium. Enough of it does to sustain EaaS take rates that would otherwise look aggressive. The typical Enchanted Rock EaaS structure charges $150-300 per kW-year, which is three to five times what a traditional standby-generator fleet commands.

The firm-power premium is what appears on the market-clearing statement. The queue-avoidance value is what supports the EaaS uplift over standby rates. Removing either source collapses the equity thesis.

04How the EaaS archetype captures the value

Enchanted Rock and comparable operators use an Energy-as-a-Service structure that solves three problems at once for the hyperscaler.

The hyperscaler avoids on-balance-sheet capex. Enchanted Rock owns the physical assets outright, financed through a mix of its own capital, infrastructure equity partners, and project-level debt. The hyperscaler signs a 15-year service agreement that looks like a utility bill, not a capital purchase. Site delivery becomes an operating expense.

The hyperscaler avoids operational responsibility. Enchanted Rock handles turbine maintenance, fuel procurement, ISO market registration, permit compliance, and grid-parallel operating agreements. The hyperscaler receives firm capacity at the meter and reserves priority rights over its own load during grid events.

Enchanted Rock captures the ISO revenue streams and shares a portion with the hyperscaler. The specific revenue-share arrangement varies by deal. A typical structure gives Enchanted Rock the operator margin on ISO market participation and gives the hyperscaler a bill credit or fee discount tied to dispatch performance. The economics work because Enchanted Rock aggregates dispatch across multiple sites and multiple ISOs, capturing scale economics that an in-house hyperscaler team cannot match.

The 15-year contract is what makes the debt financeable. Infrastructure debt providers underwrite Enchanted Rock deals against the contracted cash flow. Merchant ISO revenue is upside on top of the debt-service line. The debt looks like project finance against a hyperscaler counterparty, which is high-quality credit.

Residual risk stays with Enchanted Rock: fuel exposure (particularly RNG price), permit renewal, and market participation rules. The hyperscaler takes counterparty risk on Enchanted Rock as an operator and residual permit risk if the site becomes non-compliant. The split is asymmetric by design. Enchanted Rock is priced for taking on the operating complexity in exchange for the operator margin.

05The vendor stack

A BTM gas microgrid runs on a layered vendor stack. The operator layer is what the hyperscaler contracts with. The layers below capture supply-chain economics and are already priced by public markets and known long-term contracts.

Enchanted Rock is the canonical US operator. The company runs an EaaS structure where it owns the physical assets, contracts long-term with the site owner, and captures dispatch revenue from ISO markets on the site's behalf. PowerSecure (Southern Company subsidiary) is the closest comparable, though it operates more in the traditional utility-adjacent frame. Independent operators and integrated hyperscaler teams occupy the balance.

Caterpillar is a leading supplier of reciprocating gas gensets in US data-centre microgrids, both by installed base and by warranty coverage. Cummins is the second-largest name in the reciprocating layer. Wärtsilä ships large reciprocating engines for the >20 MW site class, and Solar Turbines occupies the continuous-duty industrial-turbine layer at similar scale. Rolls-Royce (via MTU) is a fourth notable name in continuous-duty engines.

Fuel supply is where the emissions story lands. Natural gas is delivered by conventional pipelines (Kinder Morgan, Enterprise Products) at $3-5 per MMBtu. RNG blends come from landfill gas producers (Waste Management, Republic Services), dairy digester cooperatives, and industrial biogas at $12-18 per MMBtu. RNG shifts the emissions arithmetic materially. Meta El Paso runs a natural gas plus RNG blend; Microsoft San Jose runs an RNG-heavy configuration.

EPC and integration goes to large infrastructure firms for >100 MW builds (Bechtel, Fluor, Kiewit) and with specialist microgrid EPCs for smaller sites (CDM Smith, Burns and McDonnell). The EPC layer is where site-specific permitting risk concentrates.

06Named deployments

The commercial architecture matters as much as the physical one. In the operator-owned EaaS model, the operator finances and owns the physical asset, the hyperscaler contracts for service on a 10 to 15 year term, and the operator captures dispatch revenue on the site's behalf. In the utility bridge-power model, a regulated utility purchases the modular generation, the hyperscaler funds a bridge PPA to the utility, the plant supplies front-of-meter prime power until permanent grid infrastructure completes, and the asset can then transition into utility rate base. Both monetise time-to-power. The waterfalls differ.

Meta El Paso is the class-defining bridge-power deployment. The 366 MW McCloud facility near El Paso comprises approximately 813 modular Enchanted Rock gensets running on a natural gas plus RNG blend. Public disclosures (Data Center Dynamics, Argus Media) indicate El Paso Electric (EPE) purchased the Enchanted Rock systems, with Meta paying EPE through an initial five-year bridge PPA during which the plant supplies prime power to Meta's El Paso data center campus while the permanent EPE transmission and substation buildout completes. After grid interconnection, the asset is expected to transition into EPE utility service or rate-base treatment. EPE operates in the Western Interconnection (WECC), and El Paso is the one major Texas city outside ERCOT, so market economics for this specific project run through EPE tariffs and WECC rules; ORDC scarcity settlements apply in a different market. Public reporting puts the generation-unit cost at roughly $473M, or about $1,290/kW of installed capacity.

The 813-engine configuration at McCloud is a schedule optimisation. Engineering efficiency alone would favour a smaller number of larger units. A conventional 300 to 400 MW simple-cycle or combined-cycle plant brings a single-order turbine dependency, an extended EPC schedule, and a construction path that resembles conventional utility generation. A modular multi-hundred-engine deployment substitutes procurement optionality (units drawn from pre-booked OEM production slots), staged commissioning (partial capacity online while later blocks arrive), and field-assembly simplicity for a lower headline efficiency and a higher $/kW capex.

The modular architecture pays that premium to buy schedule. Under 2026 supply-chain conditions, a conventional 300 to 400 MW combined-cycle plant brings a 36 to 48 month deployment window. A modular multi-hundred-engine deployment can bring first blocks online in 12 to 18 months and reach full capacity in 24 to 30 months. On $/MWh generation efficiency, the modular architecture trails. On NPV of time-to-power, it wins by a substantial margin when the customer values early compute revenue.

Procurement pattern defines the class more than fuel choice does. The economics of a modular engine farm can trail a combined-cycle plant on generation efficiency and still dominate on time-to-power NPV. That inequality justifies choosing hundreds of small engines over a monolithic plant of equivalent nameplate.

Microsoft San Jose is the second flagship. Its structure is closer to the operator-owned EaaS archetype. The site combines Enchanted Rock and U.S. Energy in a partnership structure, with RNG-heavy fuel to preserve California air quality permitting margin. Public documentation identifies a 60 MW Phase 1 microgrid supporting a larger Microsoft campus load. The site covers grid outages and California Base Interruptible Program events.

Beyond the two flagships, Enchanted Rock has disclosed approximately 1,500 MW of aggregate capacity across hyperscaler and data-centre applications, at various stages: operational, under construction, contracted, and announced. The operational share is materially smaller than the aggregate. Amazon has deployed a mix of Bloom Energy solid-oxide fuel cells and gas standby at select sites, though not at the microgrid scale of Meta or Microsoft. Google is evaluating multiple vendors but has not disclosed at-scale BTM gas deployments through mid-2026.

Meta El Paso structure specifics. EPE-owned systems, with Enchanted Rock as OEM-partner supplier. Meta pays EPE through an initial five-year bridge PPA sized to cover generation costs during the initial period. The project incorporates RNG alongside natural gas, with the blend varying according to operating and permitting requirements. On grid interconnection completion, the plant is expected to become EPE-controlled dispatchable capacity available under EPE tariff structures, with revenue subsequently flowing to EPE rate base; a hyperscaler-EaaS operator sits apart.

The reported ~$473M generation-unit cost ($1,290 per kW installed) falls within the broader 2025-2026 range of $1,200 to $1,800 per kW for modular gas-genset deployments at scale in the current supply-chain environment.

What Meta El Paso tells the market. Speed-to-power is valuable enough that a regulated utility, and not only a hyperscaler, will deploy modular gas generation as a bridge around the interconnection timeline. The bridge-power archetype extends the class thesis: the scarcity rent from the gap between when firm MW is needed and when the grid can deliver it can be monetised through utility ownership plus later rate-base transition, in addition to operator-owned EaaS. That widens the addressable structure for the class. Grid-parallel operating agreements and utility ownership pathways are the enabling regulatory steps.

07Revenue mix and value stack

The Enchanted Rock EaaS structure earns from six distinct revenue streams and one non-marketable value stack. The revenue streams are what appear on the operator's clearing statement. The value stack accrues to the hyperscaler and shows up as EaaS uplift over traditional standby rates.

The largest revenue stream is the EaaS service fee, typically $150-300 per kW-year on a headline basis. Capacity market payments (RPM in PJM, comparable in ISO-NE and NYISO) add $36-72 per kW-year in ISOs where the asset can register. Firming savings from avoided utility demand charges add $72-168 per kW-year. Ancillary services participation runs $12-48 per kW-year depending on ISO. Demand response and bilateral firming contracts with the hyperscaler add smaller but real streams. These ranges run non-additive. They represent alternative or overlapping manifestations of the site's value stack; realised operator revenue depends on market participation rights, contract allocation, and dispatch. The EaaS fee is the contractual monetisation mechanism. ISO, utility, and dispatch revenues are underlying sources of value that may accrue to the operator, the hyperscaler, or both depending on the contract.

The value stack accrues alongside. An underwriting range for queue acceleration is $500M-2B per site to a hyperscaler with frontier compute urgency. The range is scenario-based. Market-observed valuations stay undisclosed at that granularity. Resilience against grid outages avoids downtime costs measured in millions per hour. Environmental optionality (RNG credit trading, emission compliance) creates additional site-level value that does not clear through wholesale markets.

Value at the site separates into three categories. Power value covers capacity, energy, ancillary services, and demand response. Infrastructure value covers avoided transmission, avoided interconnection delay, and avoided substation timing. Compute value covers the earlier availability of productive GPU capacity to the hyperscaler. The three route to different balance sheets. The $2.6B illustrative figure elsewhere in this piece represents economic value potentially unlocked for the compute customer. It is a compute-customer number. The value of the power asset itself is materially smaller.

08Comparative underwriting

BTM gas microgrids, hall-level BESS, and software-defined load flex (Emerald AI and adjacent) all target the same grid-flexibility wallet. They differ on capex profile, queue-avoidance capability, and bankability.

BTM gas microgrids offer full firm capacity and bypass the bulk interconnection queue, subject to grid-parallel permit outcomes plus site-level gas infrastructure and environmental review. The trade is $1.2-1.8M per MW installed capex, real fuel exposure, and rising emissions permit risk. Contract terms run 10-15 years.

Hall BESS offers duration-limited firm capacity (2-4 hours typically), reduces the firm-import requirement subject to utility and ISO accreditation, and brings zero fuel exposure. Capex runs $1.0-1.5M per MW with the Section 48 storage ITC providing a 30% offset. Emissions permit risk is minimal.

Emerald-style load flex offers zero capex, no firm capacity, and potentially substantial reduction in required firm interconnection subject to regulator accreditation of flexible load as a firm-capacity substitute. Bankability is the primary uncertainty. Contract terms are shorter, typically 3-5 years at pilot scale.

At a hyperscaler site the three often coexist. BTM gas provides firm base capacity, Hall BESS covers dispatchable duration between arbitrage windows, and load flex provides software-defined MW during grid stress. The comparative analysis moves past zero-sum. But underwriting exposure to each varies by architecture.

Cross-architecture comparison matrix

09ROI case study

A reference deal at 100 MW installed capacity, $1,500/kW capex, 15-year EaaS contract at $200/kW-year headline fee produces the following ISO-specific economics.

These are underwriting ranges. They stand apart from observed transaction returns. The ranges are constructed from the stated capex, EaaS, fuel, leverage, tax and market-revenue assumptions. Actual project returns stay undisclosed at sufficient granularity to establish a market benchmark.

PJM offers the highest levered IRR range (14-19%) because the RPM capacity auction pays reliably plus the RegD frequency regulation market is well-established. ERCOT trails at 12-16% because there is no forward capacity market, only ORDC scarcity payments and ancillary services participation. CAISO is at 11-15% because of California-specific permit constraints and the requirement for an RNG-heavy fuel mix or alternative permitting strategy to preserve air quality margin, which raises fuel cost.

Actual site returns depend on specific contract structure, gas price and RNG blend, hyperscaler pull-through on ISO markets, and how permits resolve.

10The fuel arithmetic

Fuel supply is where the emissions story lands. Natural gas is delivered by conventional pipelines at $3-5 per MMBtu at 2026 prices. RNG comes from landfill gas producers, dairy digester cooperatives, and industrial biogas at $12-18 per MMBtu.

The emissions arithmetic makes RNG worth the premium in tightening-permit states. Natural gas produces roughly 530 g of CO2 per kWh generated. RNG produces roughly 0-100 g per kWh depending on feedstock, with landfill gas being the lowest and industrial biogas the highest. A 25% RNG blend on a natural gas base produces roughly 400 g CO2 per kWh, or 25% below unblended natural gas. A 50% blend drops emissions to roughly 265 g. Emissions values are illustrative generation-cycle estimates; actual values depend on prime mover, heat rate, methane accounting boundary, and RNG feedstock pathway.

In California, continuous-duty BTM generation faces materially tighter air-quality constraints. Projects may require RNG-heavy configurations or alternative permitting strategies to remain commercially viable. Virginia is trending toward similar treatment as PJM data center concentration rises (analyst view based on current PUC filings). Illinois is assumed to follow within 24 months in our base case, on the same underwriting view.

The RNG price premium adds roughly $30-60 per MWh generated to fuel cost relative to unblended natural gas. That is material. But the alternative is capex-intensive site conversion or class-excluded operation. RNG feedstock availability is the binding constraint. Price is secondary. Landfill gas capture and dairy digester supply grow at 4-6% per year against class deployments growing at 20-30% per year through 2028. The gap between supply and demand tightens.

Enchanted Rock has been aggressive on RNG feedstock lockups, signing multi-year offtake agreements with Waste Management and dairy cooperatives. Competitors that arrive late may find RNG scarcity is the binding constraint even before permit tightening materialises.

11Gas pipeline as the second queue

A 200 MW reciprocating engine plant at continuous duty burns roughly 40,000 to 60,000 MMBtu per day at typical heat rates. That volume does not move through a standard distribution main. It requires either an interstate pipeline tap through FERC Section 7 rules, or an intrastate lateral through state PUC jurisdiction, plus compressor station capacity that holds its own build queue.

The following pipeline figures are underwriting ranges, not class-wide observed averages. Firm transportation contracts price at $0.20 to $0.60 per MMBtu depending on region and season. Interruptible transport prices at roughly half that but brings winter-peak curtailment risk precisely when hyperscaler loads also spike. Continuous-duty BTM operation cannot rely on interruptible gas. That constraint flows straight through to the base-case fuel cost stack in Section 10.

Lateral pipeline construction takes 18 to 36 months from FERC filing to in-service depending on route length and environmental review. Compressor station upgrades take 12 to 24 months. If gas interconnection for a specific site cannot be secured within the electric-queue-avoidance timeline that motivated the BTM decision, the underlying thesis breaks. Bypassing the electric queue while sitting in the gas queue fails as queue avoidance.

Canonical operators mitigate this through site selection. Enchanted Rock, Bloom Energy, and Caterpillar EPC counterparties cluster deployments in Permian, Haynesville, and Marcellus proximate areas where high-pressure infrastructure already exists. Sites immediately proximate to high-pressure mainline infrastructure (in the Texas and Louisiana Gulf-Coast basins, typically within a few miles of a large-diameter mainline) can materially compress the gas-delivery timeline, in some cases toward the ~12-month range. In the Eastern Interconnection Northeast and California, that constraint tightens materially and can foreclose siting altogether.

Gas pipeline queue avoidance is a site-selection question. Technology alone will not solve it. Verify firm transport contracts and pipeline delivery capacity for the specific latitude and longitude before crediting queue-avoidance value in the base case. A vendor topology score without a bound pipeline capacity number falls short of a live deal.

12Prime-mover supply chain

Operator-side reports indicate 24 to 36 month lead times for some high-volume Caterpillar G3520H, Cummins QSK95, Wärtsilä 50SG, and Solar Turbines Titan 130 configurations as of Q3 2026. Actual availability varies sharply by OEM, model, order size, and factory slot. Aggregate reciprocating engine orderbooks across the top three OEMs are indicatively reported to have roughly doubled from 2023 volumes. Continuous-duty gas turbines above 20 MW face similar or longer waits.

BTM gas microgrids commissioning in 2027 and 2028 therefore depend on hardware slots locked in during 2024 to 2025. Operators without pre-committed factory positions cannot bring incremental capacity online for two to three years regardless of financial commitment. Speed-to-power depends on factory allocation more than on project-finance readiness.

Canonical operators mitigate through three mechanisms. Advance batch orders: Enchanted Rock is reported to hold standing purchase orders spanning multiple Caterpillar factory windows, allowing site-level assignment after contract close. Pooled procurement: pre-booked engine slots can be redirected to specific EaaS deployments as contracts sign, decoupling factory allocation from site commissioning. Secondary market and refurbished units for shorter-duration bridging capacity, typically at lower efficiency and higher heat rate but faster in-service.

The competitive implication reads across the class. OEM factory allocation is now a real barrier to entry. New entrants without pre-committed factory positions cannot compete on scale in 2027 to 2028 windows. That tightens the effective supply of firm speed-to-power beyond what the underlying technology curve would suggest, and it concentrates the scarcity rent among operators who secured OEM capacity early.

13Regulatory sensitivity

The state-level ranking that follows is an underwriting synthesis. States listed have not adopted class-wide BTM-gas restrictions as of 2026.

Regulatory sensitivity plays out across emissions permit strictness at the state level, RNG mandate or credit program strength, and PUC treatment of BTM parallel operation with the utility.

Our base-case underwriting treats Texas as the most permissive major market for BTM gas. Texas has generally been more permissive toward grid-parallel BTM generation than the major constrained markets. Emissions permitting is loose relative to California. The state does not have an RNG mandate, though RNG availability is emerging.

Our base case assumes tightening permit timelines in Virginia and Illinois over the next 12 to 24 months. Both states are seeing PJM data center concentration and increasing attention to distributed emissions. Neither has issued class-wide restrictions yet, but permit timelines are lengthening over a 12 to 24 month period.

California is strict. Air quality permits require RNG-heavy configurations or alternative permitting strategies for continuous-duty operation. The California BIP (Base Interruptible Program) provides a structured demand response revenue stream. The LCFS program provides some RNG credit value. But the permit strictness raises capex per kW by 15-25% relative to Texas.

New York is tightening on a similar trajectory to Illinois. Georgia, Ohio, and Arizona remain permissive but with normal state variance.

Local AQMD detail and SCR capex

State-level treatment sets the ceiling; local Air Quality Management District rules set the floor. In California's South Coast AQMD, Bay Area AQMD, and San Joaquin Valley APCD, non-attainment classifications limit non-emergency run-hours to roughly 200 to 500 per year without Best Available Control Technology installation. That can effectively foreclose continuous-duty BTM operation unless the project meets applicable emissions limits through Selective Catalytic Reduction with urea injection or another permitted control strategy, which adds $100 to $200 per kW to installed capex and introduces ammonia storage plus daily reagent logistics.

Texas Commission on Environmental Quality allows more permissive continuous operation in most zones. Harris County non-attainment areas near Houston require SCR for permitted continuous run-hours above 500 per year. Illinois EPA operates similarly with tighter constraints in the Chicago non-attainment zone, driven by O3 and PM2.5 designations under 40 CFR Part 81.

The permitting stack layers directly onto the capex figures earlier in this essay. A base install of $1,200 per kW in a permissive Texas zone becomes $1,350 to $1,400 per kW in a moderately constrained zone, and reaches $1,600 per kW plus operating complexity in South Coast AQMD. Underwriting capex should stratify by specific AQMD. State-level ranges hide zone-level dispersion. Two projects on opposite sides of a non-attainment boundary can diverge by $300 to $400 per kW of installed cost for identical hardware.

14What breaks the class

The class economics can break under the following triggers, each with a distinct mechanism and probability under our base-case underwriting.

1/ Federal EPA rule tightening on distributed gas generation. Class-wide capex and permit costs would rise 30-60%. The mechanism runs through the New Source Performance Standards (NSPS) and could arrive as a rulemaking through 2027-2028. This is the highest-impact bear case.

2/ RNG feedstock scarcity as availability tightens. RNG blend share is what preserves California and increasingly Virginia permitting margin. If landfill gas and dairy digester feedstock cannot scale with class deployments, blend shares drop and fuel exposure rises. This is a supply-chain risk that shows up first in California, then in tightening-permit states.

3/ Nuclear PPAs at competitive economics. If Constellation, Vistra, and others deliver nuclear PPAs at $60-75 per MWh to hyperscalers at commercial scale, the firm-power premium compresses across the whole class. Enchanted Rock EaaS rates would need to fall by roughly 20-30% to remain competitive.

4/ Utility ratebase moves to preempt BTM. Some state PUCs (potentially Arizona, potentially Virginia) may move to bring hyperscaler firm-power provision inside utility ratebase. This shifts contracting freedom and captures the class rent for the utility.

5/ Queue reform compresses interconnection timelines. The common-mode risk hits every flexibility architecture, including hall-scale storage and load flexibility. If FERC and ISOs collectively bring gigawatt-scale queues from 6-7 years down to 2-3 years, the queue-avoidance value collapses and the EaaS uplift disappears.

15The nuclear PPA threat, quantified

The nuclear-PPA competitive threat is fundamentally a queue-avoidance question. Levelised cost is secondary. The real test is whether nuclear can deliver firm power to a hyperscaler without requiring the hyperscaler to wait for new grid capacity. Nuclear-fed hyperscaler PPAs currently contract for grid-connected delivery. A hyperscaler still needs interconnection to receive nuclear-fed power, which reintroduces the queue problem the BTM gas class was built to solve. That structural constraint could still change through 2029-2032 if nuclear-BTM structures become commercial.

Constellation Energy's Three Mile Island (Crane Clean Energy Center) restart contracted with Microsoft on terms reported in the approximately $85 to $110 per MWh range over a 20-year period, with the specific rate depending on inclusion of adjustments and adder disclosures. Vistra's Comanche Peak PPA discussions with hyperscalers have been reported by public analysts (renewables researchers, sell-side) at $105-120 per MWh for the reported AWS transaction (1,200 MW, deliveries Q4 2027). Some earlier reporting cited lower ranges tied to different term structures. New small modular reactor construction from Kairos, X-energy, and others is projected to deliver first megawatts in 2029-2030 at unit costs that would require regulated support to reach $80-100 per MWh.

BTM gas microgrids currently deliver firm MW at an all-in cost of roughly $110-140 per MWh, accounting for capex amortisation, fuel, EaaS operator margin, and dispatch revenue offsets. The delta is $20-50 per MWh in favour of nuclear at commercial scale.

The nuclear threat has not materialised at commercial scale yet. Nuclear PPAs currently contract for grid-connected delivery. Behind-the-meter delivery remains outside the current commercial scope, and a hyperscaler still needs interconnection to receive nuclear-fed power, which reintroduces the queue problem the BTM gas class was built to solve. Nuclear capacity is also limited: Constellation, Vistra, Duke, and Southern together represent the vast majority of the US nuclear fleet, and their capacity is already partially subscribed.

If nuclear-fed hyperscaler PPAs at commercial scale become standard by 2028-2029, the queue-avoidance component of the BTM gas thesis compresses first. The firm-power premium survives longer because ISO capacity payments accrue regardless of the marginal generation source. But EaaS take rates would need to fall by roughly 20-30% to remain competitive with a nuclear-backed grid-connected alternative. That materially compresses operator margins.

All-in delivered nuclear economics

A $75 to $85 per MWh nuclear PPA is a wholesale price at the plant bus. The delivered cost to a data center meter adds transmission service charges, congestion premiums during system stress, capacity charges under ISO settlement rules, and line losses that scale with distance from the generating asset.

In PJM, transmission service charges run $8 to $18 per MWh depending on zone and load-serving entity. Congestion premiums in constrained areas including northern Virginia, central Pennsylvania, and portions of ComEd territory can add $10 to $30 per MWh during peak hours. Capacity charges under Base Residual Auction and Capacity Interconnection Rights rules add $15 to $25 per MWh depending on capacity market clearing. Line losses run 2 to 4 percent of the delivered price.

Under this illustrative PJM delivered-cost stack, the delivered nuclear cost to a data-center meter in a moderately constrained zone reaches approximately $125 to $145 per MWh. Specific PPA settlement mechanics vary materially by counterparty and contract structure. In ERCOT, the equivalent stack for a Comanche Peak or South Texas Project PPA reaches $95 to $120 per MWh delivered given tighter congestion but higher transmission service costs to Dallas-Fort Worth and Houston loads. In CAISO, wheeling and congestion push the number higher still.

The competitive read tightens the BTM gas thesis; loosening framing understates it. On delivered $/MWh, BTM gas at $110 to $140 is inside the delivered-nuclear range in constrained ISOs. The queue-avoidance advantage adds to a cost position that is already competitive on delivered energy, with speed to power as an added edge. That reframes the nuclear PPA competitive threat: nuclear beats BTM gas on carbon and on baseload durability past 2032, but does not necessarily beat it on delivered cost during the 2026 to 2030 scarcity window that motivates the class.

16The call

The BTM gas microgrid is a real emerging asset class through 2028. Named at-scale deployments demonstrate the technical and commercial architecture works at hyperscaler grade. Enchanted Rock is the best-positioned canonical operator. The vendor stack (Caterpillar and Cummins on the engine side, established RNG suppliers, tier-one EPCs) supports scale.

The class captures a specific value proposition that other flexibility architectures cannot match. Firm capacity clears through the ISO capacity market for registered assets. Assets outside ISO capacity constructs, such as Meta El Paso in WECC, clear the equivalent value through utility PPA or hyperscaler contract. A substantial reduction in incremental grid-capacity dependence for the firm MW supplied behind the meter is payable by hyperscalers through EaaS uplift. Batteries capture partial firm capacity (duration-limited) and reduce firm-import requirements subject to accreditation. Software captures no firm capacity and can reduce firm interconnection needs subject to regulator recognition of flexible load. The combined capture explains the Monte Carlo result. In the companion model, "wins" means the architecture captures the highest share of a single site's modelled flexibility value after capex, accreditation, and contract structure. It is a site-level ranking. Market share is a separate underwriting question. In the model, Enchanted Rock wins 96-100% of Base Case draws. Specifically, within the Flexibility Right Monte Carlo model, running on constructed inputs from stated assumptions; observed market data stays absent, three architectures compete on accredited MW, capex, contract duration, and value capture. The result reflects constructed model inputs. Observed transaction data stays undisclosed at that granularity. The EaaS take rate runs three to five times traditional standby rates because it reflects the value stack size, while engine hardware itself is commoditised.

The remaining underwriting question is what breaks the value stack. Under our base-case assumptions, emissions permit tightening compresses the addressable market by state on a roughly 24 to 36 month lag, and queue reform compresses the queue-avoidance value across the board on a roughly 4 to 7 year lag. Everything else is secondary to those two common-mode risks at the class level.

Enchanted Rock and its private-operator peers keep most of the EaaS uplift. Public equity offers upstream buildout exposure through Caterpillar (engines), Southern Company via PowerSecure (integration), and Waste Management (RNG feedstock). No listed equity captures the operator rent directly.

An institutional investor who wants direct exposure to the class needs either an Enchanted Rock secondary or a co-investment alongside an infrastructure sponsor. Access matters more than the specific take rate assumption or the specific ISO the deal is written for.

BTM gas competes with hall-scale storage and load flexibility for the same grid-flexibility wallet. All three show up at hyperscaler sites in different combinations and produce different equity outcomes. Underwriting turns on which architecture captures the biggest share of the interconnection-scarcity rent after accreditation, financing, enforcement, and value-sharing. The mechanics are who moves scarce grid capacity from utility to customer, and who bears the residual risk when the scarcity resolves. Through 2028, BTM gas takes the largest share.