Behind the Meter or in Front of It:
Who the Project Serves Decides Its Rules

Strip away the technology debate and every on-site power project in California must first answer a structural question: does it exist to serve the load behind your own meter, or to sell into the grid in front of it? That single answer assigns the project its regulator, its interconnection door, its revenue benchmark, and most of its risk, before any machine is chosen.

Section 01The fork comes before the technology

The phrases sound like siting jargon. They are actually legal identities. A behind-the-meter project sits on the customer side of the retail meter, operates in parallel with the utility system, and delivers its output to the host's own load; the grid simply sees a smaller purchase. A front-of-meter project connects to the distribution or transmission system in its own right and sells what it makes, as wholesale energy, capacity commitments, environmental attributes, or all three, to buyers who take delivery through the grid.

Nothing about the machine decides which identity applies. The same reciprocating engine, gas turbine, fuel cell, solar array, or battery can be built on either side of the meter. What changes is who buys the output, and the American regulatory system is organized around exactly that question. Self-supply and retail-side arrangements are state matters. Sales for resale into wholesale markets are federal matters. In California the split is written directly into the interconnection tariffs: the state's Electric Rule 21 governs interconnection of net-metered facilities, non-export facilities, and qualifying facilities selling to the host utility at avoided cost, and it does not apply to facilities intending to participate in wholesale markets, which interconnect instead under a utility's federal wholesale distribution tariff or the grid operator's tariff.¹

That is the whole paper in one sentence: who the project serves decides which rulebook it lives under, and the rulebooks differ in queue, tariff, permit, and economics. Owners who choose a technology first and discover the identity later tend to discover it as schedule.

Section 02Two doors into the grid

Consider the interconnection doors separately, because they do not resemble each other.

The behind-the-meter door is a distribution-level review under state jurisdiction. The engineering question is bounded: can this equipment operate in parallel with the local circuit without degrading safety, reliability, or power quality for anyone else on it. A project configured never to export, enforced with protective functions and control settings rather than promises, narrows the question further, because power that cannot flow onto the wires cannot disturb them. The review runs on screens first, with deeper study only where a screen fails, and it is measured against your circuit rather than against a statewide queue.¹ It is not instant, and it is not automatic. It is bounded, and bounded is the property a deadline cares about.

The front-of-meter door is the grid operator's interconnection queue for transmission-connected projects, or the utility's federal wholesale distribution tariff at distribution voltage. The queue runs on annual cluster windows: applications enter together, are studied together, and share allocated network-upgrade costs. A project that misses a window waits for the next one; the next transmission cluster window in California is scheduled to open October 1, 2026, per the grid operator's board briefing of July 2025.²

The scale of that queue is the defining fact of the export path. The 2023 application window, Cluster 15, drew 541 interconnection requests totaling 347 gigawatts, and with it the active queue exceeded three times the capacity California needs to meet its 2045 policy requirements, by the grid operator's own account.² Federal regulators responded with Order No. 2023 in July 2023, replacing serial first-come, first-served studies with first-ready, first-served cluster studies, raising deposits and site-control requirements, and adding penalties for late studies.³ The California grid operator layered its own intake reforms on top, approved in stages through 2024 and 2025. The reforms are doing what they were designed to do: Cluster 15 shrank to 255 resubmitted requests and then to 145 projects totaling 68 gigawatts proceeding into studies.² Read that honestly. The export door now demands demonstrated site control, real deposits, and readiness evidence simply to stand in line, because the line had grown to three times the size of the state's need.

The behind-the-meter door is narrower in scope, but it is yours. No cluster, no annual window, no allocation of a stranger's upgrade costs. For an owner whose actual deadline is a production line rather than a market position, that difference is usually decisive before the economics are even opened.

Section 03Which price each kilowatt-hour competes against

The second structural difference is the benchmark. A self-supply kilowatt-hour displaces a purchase at the bundled retail rate. An exported kilowatt-hour sells at the wholesale price, or at a contract price negotiated in its shadow. Those two numbers are not close.

In May 2026, industrial customers in California paid an average of 20.2 cents per kilowatt-hour at retail, and commercial customers 24.1 cents.⁴ In the fourth quarter of 2025, the day-ahead wholesale market averaged 42 dollars per megawatt-hour, about 4.2 cents per kilowatt-hour, roughly five percent lower than the same quarter a year earlier.⁵ The comparison is directional, not a project pro forma: the bundled retail rate includes transmission, distribution, and public-program components that a wholesale seller neither pays nor collects. But the direction is the decision. A behind-the-meter kilowatt-hour competes against a benchmark near twenty cents; a front-of-meter kilowatt-hour competes against a benchmark near four. No technology's cost curve is indifferent to a gap that wide.

The retail side deserves its own discipline, because the twenty-cent benchmark is not captured automatically. Some bill components are designed to be nonbypassable, and a careful model values the project against the components it actually avoids rather than the printed average rate. Demand-based charges have their own logic: an on-site machine reduces billed demand only to the extent it is running when the billing peak occurs, so its maintenance schedule and outage behavior belong inside the demand-charge line, not in a footnote. Utilities also commonly apply standby provisions to customers operating parallel generation, and those tariff-specific terms belong in the model from the first draft.

One more retail-side fact disciplines sizing. Exports from new customer-generation systems interconnecting since April 2023 in the investor-owned utilities' territories are credited under the net billing tariff adopted in Commission decision D.22-12-056, at values derived from the state's avoided-cost methodology: usually below the retail rate, occasionally above it in a narrow band of late-summer evening hours.⁶ The practical consequence is that the last increment of an oversized behind-the-meter machine earns the weakest rate available to it. Self-supply economics are strongest exactly up to the load, and they degrade past it.

The export side carries the mirror-image caveats, and an honest framework prints them with equal weight. Merchant energy revenue moves with the market; the market monitor's quarterly reports document the movement.⁵ Capacity revenue exists, but it generally requires deliverability status, which is precisely the studied, upgrade-laden part of the queue; one-third of the Cluster 15 projects advancing through the reformed process elected energy-only status, trading capacity value for speed and lighter upgrade exposure.² A power purchase agreement can move price risk onto a counterparty, provided the project can find a creditworthy one and hold it through the queue's calendar.

Section 04Permits follow the purpose

Air permitting is meter-blind. A combustion machine, an engine or a turbine, needs its air-district permit whether it serves the host or the market, with the same control-technology expectations and, in some districts, the same offset exposure. Non-combustion classes, fuel cells, solar, and batteries, largely avoid that particular gate on either side of the meter while carrying their own siting questions: gas service and footprint for fuel cells, area for solar, fire-code setbacks and safety review for batteries. The fork does not change what you permit. It changes whose land-use process you are standing in.

A behind-the-meter project is typically built inside an existing industrial fence, within existing entitlements, on land the host already controls. The practical critical path is usually the air permit for combustion equipment and the utility's parallel-operation review, not zoning.

A front-of-meter project stands on its own land. That means land control, discretionary land-use approvals, and the environmental review that discretionary approvals generally trigger in California, plus a generation tie, interconnection facilities, and, at transmission scale, the market infrastructure of a functioning power plant: telemetry, scheduling, settlement, and a party responsible for each. None of this is a criticism of export projects. It is the ordinary cost of being a power plant. The mistake is inheriting that cost accidentally.

Export is not a feature added to a self-supply project. It is a different project, with a different regulator, a different queue, and a different buyer.

Section 05Why the industrial answer usually lands behind the meter

Put the pieces together and the pattern that emerges from most industrial cases is not ideology. It is arithmetic plus jurisdiction.

• The buyer already exists. The host's own load takes every kilowatt-hour at retail-benchmark value, with no counterparty search, no marketing, and no market-price exposure on the revenue line.
• The benchmark is several times higher. Roughly twenty cents against roughly four, before project costs, and still a wide gap after subtracting the components self-supply does not avoid.^4,5
• The door is bounded. A non-export review against your own circuit, rather than an annual cluster window shared with hundreds of speculative gigawatts, is the difference between a project schedule and a market phenomenon.²

The honest counterweights deserve equal print. A self-supply project concentrates its fate in the host: if the plant closes, the load shrinks, or the process changes, the project's only buyer changes with it, and no market stands behind it. The machine becomes part of operations, with fuel, maintenance, staffing, and outage consequences the host now owns. The capital comes from somewhere, either the host's balance sheet or a third-party structure that reintroduces a counterparty with rights. And the utility relationship does not vanish behind the meter: standby terms, interconnection agreements, and nonbypassable components remain. Behind the meter is usually the stronger industrial answer. It is not a free one.

Federal tax treatment, for its part, is close to neutral on the fork. Under current law as of August 2026, the investment tax credit for qualifying clean-energy property is 30 percent under sections 48 and 48E of the Internal Revenue Code, on either side of the meter; statutory adders exist but each must be individually qualified, never assumed, and ownership structure determines who can actually use the credit.⁷ That analysis belongs to qualified tax counsel, not to a white paper's assumption.

Section 06When export logic actually applies

Technology neutrality includes structural neutrality. There are cases where the export identity is the right one, and pretending otherwise would repeat the equipment seller's error in reverse. Four cases survive scrutiny regularly.

1. The resource structurally exceeds the load. A byproduct fuel stream larger than on-site demand, land whose highest use is generation, an interconnection position that is itself the asset. When the site's endowment rather than the site's load is the thing of value, the market is the natural buyer and the queue is the price of reaching it.
2. The owner is genuinely entering the power business. Wholesale participation means market interfaces, credit support, asset management, and regulatory standing. Some industrial owners build that capability deliberately and do it well. The failure mode is acquiring a second business by accident, one afterthought at a time.
3. Grid-scale storage. Storage in front of the meter sells market products: energy arbitrage, ancillary services, capacity. Storage behind the meter sells a different product: demand management, ride-through, coincidence with billing peaks. Both are legitimate, and they are different projects with different sizing logic and different doors. The battery is the cleanest demonstration that the fork, not the technology, sets the rules.
4. A contracted buyer already exists. If a creditworthy offtaker's committed terms beat the self-supply value after honest accounting for queue time and upgrade exposure, the export case stands. Test it against signature-quality interest, not against a price forecast.

Then there is the hybrid temptation, which deserves its own warning. "We will size it larger and sell the excess" sounds like prudence and usually is not. Crossing from non-export to export can move the project from the state track to the federal one; the interconnection identity changes, and the schedule changes with it.¹ Meanwhile the marginal revenue on those excess kilowatt-hours is the weakest line on the board: avoided-cost-based export credits on the retail side,⁶ or the wholesale price in the market.⁵ The disciplined structure for most industrial sites is the opposite: size to the load, configure non-export, keep the identity clean, and let any export ambition stand or fall later as its own project.

Section 07The fork test

Six questions settle the fork for most sites, on paper, before any technology is scored. They are answerable in about a week with the site's bills, the real deadline, and some honesty.

1. Who consumes the first kilowatt-hour?If the answer is the host's own load, you are describing self-supply, whatever the proposal's cover page calls it.
2. Which benchmark does each kilowatt-hour compete against?Write down the site's actually avoidable retail components and the current wholesale benchmark, both dated. The gap between them is most of the economics.
3. Which door does this configuration walk through?State distribution review or federal queue, and what the schedule becomes if the configuration flips later. The door is chosen by the design, not by preference.
4. Does the self-supply case survive with exports valued at zero?If the model needs export revenue to close, it is not a self-supply project. Judge it as an export project, with the export project's queue and buyers.
5. Does the export case survive without the host?The symmetric test: an export plant justified partly by on-site use should stand as a market project on its own numbers.
6. Who signs, and for which business?A meter holder buying firmer supply is one signature. An entity entering the wholesale power business is another. A board should know which document is in front of it.

Section 08Decide the fork on paper

Most of the expensive failures in this field are not technology failures. They are identity failures discovered late: a self-supply project that drifted into export ambitions and found itself in a federal queue behind hundreds of gigawatts of strangers, or an export concept underwritten at a retail benchmark its kilowatt-hours were never going to touch. Both are avoidable in the first week, with the fork decided explicitly and in writing.

That is where an owner's-side study earns its fee: not by preferring a side, but by pricing both honestly for one specific site and letting the load, the deadline, and the documents decide. In our study work the fork is settled before any machine is scored, because every number that follows inherits from it. Who the project serves decides its rules. Decide that first, and the rest of the decision becomes tractable.