Who owns ERCOT’s power, and who owes Texas?

ERCOT coordinates plants it doesn’t own

The elementary answer is already settled. ERCOT is the grid and market operator; it owns no generating or transmission assets. The Texas Comptroller says so plainly. Asking who “owns ERCOT” confuses governance of an institution with title to the equipment it coordinates. LeadIQ’s ERCOT profile nevertheless lists “Owner: E. C.” and another supposed owner entry while correctly describing the organization as a nonprofit. That directory field misstates the ownership of the grid operator. How widely readers believe it remains unmeasured. LeadIQ profile. Texas Comptroller explainer.

The concentration story is also partly written. Modo Energy’s October 22, 2026 analysis puts Vistra, Constellation and NRG at 61% of thermal capacity, but 28% of all capacity including batteries. It documents the much broader group building renewables and storage. Those findings deserve attribution; this investigation does not present them as new. Modo’s Q3 2026 fleet and development analysis.

The academic work reviewed includes qualitative research on utility resilience and studies of household harms and load shedding. Those are relevant to how Uri unfolded; they are not a plant-level causal estimate of ownership type. Utility resilience study; household burdens study.

The gap is in the claims beneath those names. A whole-plant parent map, a unit-level fractional ownership map, and a map of who can sell the output answer different questions. EIA supplies the physical census and some joint-ownership interests. State capacity filings and corporate disclosures help connect project companies to controlling parents. Loan agreements add the creditor’s rights: who can demand repayment, seize collateral or call a guarantee.

The accompanying research joins those rights, keeps their dates, and records where the chain cannot be completed. It also tests an inconvenient possibility for the obvious story: ownership can become more concentrated while the frequency of a pivotal supplier declines.

One plant can have several owners—and several kinds of control

“Ultimate owner” therefore needs a stopping rule. For a listed company, the useful control endpoint is the public parent, with public shareholders collectively behind it. For a municipality or cooperative, it is the city or member-owned system. For a private fund, the manager, fund and portfolio company have different roles; identifying the manager does not identify every pension fund or other limited partner supplying the money.

The fleet ledger uses disclosed unit fractions and traces each claim to a supported control parent where possible. It does not multiply every ownership percentage through trusts, listed companies and funds to manufacture a complete beneficial-ownership census. Where operator identity is only a proxy for ownership, the row says so.

The concentration change is concentrated in 2026

Vistra, Constellation and NRG account for approximately 59.3% of operating-classified thermal summer capacity in the EIA grid-supply sectors. Adding municipally owned CPS Energy raises the share to 70.3%. That fourth owner is a reminder that Texas’s generation fleet has long included public and cooperative ownership.

The denominator is 74,205.2 MW: gas and other thermal generators in the electric-utility and independent-power-producer sectors, including nuclear and coal. It excludes industrial/commercial generation from this particular concentration measure, while keeping those plants in the census. It is not all “dispatchable” capacity: hydro and batteries can respond too, subject to different constraints. Nor is an EIA operating classification a guarantee that a generator is available in a given hour.

The historical groups are Vistra, NRG and Calpine; today Calpine belongs to Constellation. The comparison uses the leading mapped groups at each date. Projecting today’s corporate families backward would change it. Unresolved historical interests prevent treating the early ranking as an exhaustive beneficial-ownership result.

The more revealing calculation holds the old fleet still. On the 1,105 generators at 193 plants in the common 2025–current thermal cohort, using unchanged 2025 summer ratings, updating ownership increases the share from 50.37% to 59.12%. The current physical inventory takes it only another 0.18 percentage point, to 59.30%.

The ownership change is reconciled plant by plant: Constellation’s existing fleet joining Calpine adds 4,626.3 MW to the combined group; NRG’s three acquisitions from LS Power add 2,040 MW; Vistra’s Altura purchase adds 485 MW; Fusco’s sale and Calpine’s exit from Gregory remove 760.478 MW. Net: 6,390.822 MW, or 8.75 percentage points of the fixed denominator. This calculation attributes the measured ownership change to transactions. Whether mergers caused new generating capacity requires a causal study.

The chronology includes transactions that a year-end dataset misses: Constellation closed Calpine on January 7, NRG closed its LS Power transaction on January 30, Fusco changed hands on October 14, and Vistra closed Cogentrix on October 30. The ten-plant reconciliation includes each source URL. Constellation closing; NRG closing; Fusco divestiture; Vistra closing.

The result is sensitive to the question, but its direction survives reasonable checks. Including out-of-service and standby units makes the current thermal share 57.9%, versus 50.0% in 2021. Including industrial/commercial thermal capacity makes it 55.3%. Restricting to gas makes it 55.6%. The named Big Three’s share of all operating capacity including storage is 25.8%. Those measures must not be substituted for each other.

ERCOT’s physical inventory has an additional complication. ERCOT’s resource forecast identifies switchable Antelope and Elk stations that EIA assigns to Southwest Power Pool, and availability deductions at other switchable plants. Those records are preserved as inventory exceptions; physical capacity in the main series is not the same as firm capacity reserved for ERCOT. ERCOT May 2026 resource forecast.

Joint interests also matter. Constellation holds 44% of South Texas Project, CPS Energy 40%, and Austin Energy 16%. A whole-plant label under Constellation does not describe that fractional allocation. The ledger keeps this split and the disclosed finance-lease treatment at Pasadena. Source: EIA owner schedules, PUCT generating-capacity filings, company filings and the linked parent crosswalk. EIA annual data; EIA monthly data.

The opposite hypothesis survives

Capacity share measures ownership. The ability and incentive to raise prices also depend on each hour’s demand and available supply. A supplier is pivotal when the remaining available supply cannot cover the relevant demand. The Independent Market Monitor’s 2025 report puts pivotal-supplier frequency at 10.3% of hours in 2025, compared with 42.5% in 2022 in its revised historical series. That can coexist with a concentrated thermal fleet. 2025 State of the Market Report, pp. 113–116 and footnote 79.

There is an important trap in the comparison. The monitor changed the test and restated earlier years. The old 2024 report’s 63% estimate for 2024 must not be spliced to the new 10.3% estimate for 2025. The internally consistent series in the 2025 report is 42.5%, 36.1%, 23.1% and 10.3% for 2022–2025.

The report finds no physical-withholding competitiveness concern from the largest suppliers, while still raising concerns about economic withholding, evening periods and growing load. These observations end in 2025 and do not measure the competitive effect of the 2026 acquisitions. That effect remains an open empirical question. It does kill the shortcut from “three companies own much of the thermal fleet” to “the market is demonstrably less competitive.” Contract positions, location, available capacity and time of day intervene.

Nine closed loans, seven sponsor groups

The current executed-loan inventory contains nine agreements, covering 5,446 MW and approximately $4.10 billion. The fund was authorized in 2023; the first loan agreement was signed in June 2025. It is inaccurate to date actual lending to 2023. Three further applications remain in due diligence and are excluded from the closed-loan total. PUCT program register.

The table below uses announced project capacity and rounded commitments. These are not seasonal accredited MW, and projected completion is not capacity already available to ERCOT. TH Wharton and Pin Oak Creek are the two operating loan projects at this cutoff, totaling 916 announced MW. Precise borrower entities, execution dates, sources, loan balances and unresolved guarantee terms are in the downloadable ledger.

NRG, Vistra and Constellation receive 49.0% of committed dollars and 54.2% of the announced MW. Customer-owned municipal/cooperative sponsors receive 12.6% of dollars and 12.7% of MW. Whether state lending increases incumbent concentration remains open. Testing that requires comparable capacity measures, completion dates, retirements and subsequent acquisitions.

There is another concentration: 52.7% of the committed dollars and 48.9% of MW are in ERCOT’s West load zone, across Basin Ranch, Vistra’s Permian expansion and Pecos. This is exposure of the public loan portfolio to a region. It is not an antitrust market-share estimate or a measured probability of correlated default.

The same interest rate buys different protection

NRG’s second-quarter filing classifies its three Texas Energy Fund loans as recourse debt. The executed Cedar Bayou documents name NRG Energy, Inc. as guarantor. The guarantee is subject to its contractual terms, including a pre-completion route under specified acceleration and contribution provisions that can end the debt guarantee. It is not an unlimited promise without exceptions. NRG debt disclosure; Cedar Bayou parent contribution agreement and guaranty.

Vistra’s disclosure is different. The approximately $583 million loan is to Permian Power I LLC. It is secured by that subsidiary’s assets and equity, but is otherwise nonrecourse to Vistra and the other named group entities. At June 30, $172 million was outstanding. The state has project collateral; it does not have the same disclosed route to the parent balance sheet. Vistra second-quarter debt note.

Both are 3% loans. It does not follow that they have identical risk or identical economic subsidies. The documents identify material differences in the state’s legal protection. They leave default risk and the appropriate loan price unmeasured.

For private projects, the program requires a senior secured claim. Public systems have different permitted structures. The 2025 statute allows municipal securities backed by net revenues on parity with other senior debt, and cooperative loans secured by substantially all electric-system assets on parity with other senior debt. The exact Rayburn loan instrument was not obtained, so that legal option must not be described as its verified executed security package. PUCT loan rule; SB 2268, section 2.

Equity bears losses before senior debt is impaired in the ordinary project capital structure. State security is real protection. Yet taxpayers still bear the risk that recoveries fall short and the economic cost of lending on favorable terms. The state receives interest, principal, security and performance promises; it does not receive the ordinary residual upside of a successful private plant through these loans.

Generation loans sit beside a different public guarantee in the electricity system. The delivery-charge investigation follows storm-restoration bonds backed by adjustable charges on future distribution customers. The project borrower’s repayment obligation and the statutory customer charge give creditors different routes to cash.

“The other 40%” is not necessarily unborrowed shareholder cash

The fund can lend up to 60% of qualifying project cost. That does not make the rest entirely unlevered equity. Kerrville PUB says the remaining approximately 40% of Rock Island’s cost will be financed with tax-exempt revenue bonds. This is public utility borrowing alongside a state loan, with customers ultimately behind the utility. KPUB project financing explanation.

Basin Ranch illustrates borrowing above the project company. Kenon’s annual filing reports that CPV’s then-70% share of required project equity—approximately $470 million—was funded with a $300 million Bank Leumi loan and a $170 million OPC bridge. The Leumi facility later increased to $430 million alongside CPV’s acquisition of the remaining 30% of the project. Texas holds the senior project lien; Leumi’s security includes the account receiving project dividends. Kenon 2025 Form 20-F, pp. 70–71 and 159–160.

That holding-company borrowing does not remove the subordinate equity claim inside the project or put the bank ahead of Texas on the project collateral. It does show why the ownership and financing chains must both be followed. A sponsor’s “equity contribution” describes its position inside the project. The cash can originate in borrowing above that company.

Pecos supplies a second private example. Continental’s third-quarter filing, made October 30, 2026, identifies a 50% economic, noncontrolling equity interest in the venture and a $150 million loan from partner Mercuria used to help fund its contributions. That note is secured by Continental’s venture interests and distributions; guarantees of these obligations are not proof that Continental guarantees the state loan. Continental third-quarter filing, Note 10.

CPV also makes a simple domestic-versus-foreign story untenable. The public filings trace it through OPC Energy and Kenon to a controlling trust structure associated with Idan Ofer, while minority investors participate at several levels. This is a gas project. Foreign utilities, funds and financing claims are not a feature unique to wind and solar.

Cheap credit cannot by itself create a profitable plant

The harder policy question is additionality: which plants will be built because of the loan, or built earlier, larger or with better reliability? A closing announcement cannot answer that counterfactual. The rule permits reimbursement of eligible costs incurred before closing, so the presence of a loan is not proof that investment began only after the state offered it. Loan rule, including eligible costs and performance covenants.

There is a useful example among the pending applicants. In a November 2025 extension request, Rockland’s New Gulf project said forward power prices did not cover increased project costs and that a final investment decision required a move toward long-term contracted offtake. That applicant’s explanation shows why cheaper financing can leave the revenue contract unfinished. Other funded projects need their own financing analysis. New Gulf extension request.

Performance conditions matter too, but they are not an automatic one-strike foreclosure rule. Cedar Bayou’s contract provides corrective-action and reserve-account procedures for failures to meet its performance covenant. The state’s continuing financial and performance assessments are not fully public. Loan-level monitoring disclosures would let the public assess protection more precisely than construction announcements do. Cedar Bayou credit agreement, sections 5.27 and 7.1.3.

The loan and completion-bonus programs also overlap. Five bonus agreements cover 1,438 MW and up to $172.56 million in future conditional payments. Pin Oak Creek and TH Wharton account for 916 MW in both lists. Deduplicating the two lists reduces their combined 6,884 MW to 5,968 MW of unique capacity. None of that converts an awarded maximum bonus into cash already paid. PUCT loan and grant tables.

Wind, solar and batteries expose the limits of a parent label

The public record supports a wider range of renewable and battery sponsors than the thermal Big Three. It does not support treating each sponsor as the sole beneficiary of every project dollar. A project can have a managing sponsor, a minority investor, tax equity, project debt and a contract placing output risk elsewhere.

NextEra’s June 2026 portfolio is unusually instructive. Its Texas entries disclose gross capacity, stated ownership fractions, net capacity and flags for differential-membership-interest financing. Blue Summit III appears with 10% direct NextEra Energy Resources ownership and 40% through XPLR Opco. The remaining interest is not identified by those two rows alone. Some projects marked “100%” also carry differential-membership-interest financing. The table’s “ownership” column is therefore not a complete allocation of cash, tax and control rights. NextEra generation portfolio, June 30, 2026.

Ørsted shows how control and cash rights can diverge. Its audited 2025 schedule classifies Western Trail and Helena as subsidiaries at 20% ownership, but Mockingbird and Sparta as joint operations at 50%. A smaller equity interest can coexist with stronger control. Ørsted annual report, Note 7.4, p. 195.

Its transaction announcement for a four-wind-farm portfolio including Western Trail says Stonepeak will receive 80% of associated cash distributions, while Ørsted keeps operations and specified optionality. The portfolio also raised approximately $700 million in tax equity. Calling the entire wind farm “Ørsted” can describe operating control while obscuring who receives the cash. Ørsted–Stonepeak transaction announcement.

The research package includes 53 Texas rows extracted from the NextEra document, including 50 facility labels, as a documented example of the additional project detail needed. It is not a statewide estimate of foreign ownership or tax-equity participation. A credible percentage for either would need project-level interests, matching dates, and explicit treatment of fund investors and commercial-control contracts.

The census keeps a plant’s reported entity and an evidence status even when its ultimate parent remains unknown. An unresolved project is not silently treated as an independent company; equally, a plausible corporate name match is not promoted to verified ownership.

Uri is a possible test, but not a ready-made ownership experiment

Unit-level outage data are public. ERCOT’s March 12 workbook contains 2,599 event rows, representing 585 station–unit pairs at 280 stations. It also contains 74 zero-MW-reduction rows. The file covers events beginning February 10–19, records an early reporting snapshot, and warns that short outages were not uniformly required to be reported. Raw row counts are not counts of failed plants. ERCOT unit outage workbook.

A valid comparison would need the full contemporaneous fleet, correct February 2021 ownership, non-overlapping outage intervals clipped to the storm, and exposure to weather, technology, age and fuel constraints. Our year-end 2021 concentration baseline is also too late to substitute for February ownership. Detailed resource-specific cause information was withheld from the public responses. An ownership coefficient without those controls could simply measure where and what a company owned. ERCOT responses, question 5.

Rayburn shows why the date and the claim matter. It bought the Sherman plant in June 2023. The Uri file names Panda Sherman Power LLC. Classifying that 2021 performance under today’s cooperative ownership would be wrong. Rayburn’s CEO later described the plant as having performed well while the cooperative suffered as a buyer in the market. That testimony describes the difference between plant performance and a customer’s financial exposure. A causal estimate would need comparable outage and fuel conditions. Rayburn acquisition announcement; CEO interview.

This investigation does not report a public-versus-private Uri performance effect. The released data support constructing the test; they do not justify the result in advance.

How to audit this investigation

The accompanying data package puts the physical generator census, reported ownership claims, dated control-parent crosswalk, calculated concentration, executed state loans, conditional grants, and source links in different files. Plant and generator IDs carry the joins; names remain human-readable labels. Unresolved parents, inferred ownership proxies and missing guarantee terms remain visible fields.

The main ownership measure is capacity allocated using disclosed interests, grouped by control parent. It is not a measure of hourly available supply, plant output, profit, tax benefits or contractual dispatch control. The summer-capacity basis is held consistent across the concentration comparison. All-source ownership and thermal ownership answer different questions; battery power MW and energy MWh must not be mixed.

The loan measure is announced commitment dollars, grouped by current sponsor. It is not cash spent, an estimate of fiscal loss or the present value of the loan subsidy. Outstanding borrowing has its own date. Completion-bonus commitments are deduplicated from loans when counting unique capacity.

High confidence: the nine-loan inventory; the disclosed NRG/Vistra recourse contrast; Basin Ranch’s par, carrying value and financing-rate disclosure; and the loan/grant overlap. Conditional on the stated census and mapping rules: concentration and its change. Incomplete: the beneficial ownership of private funds and tax-equity arrangements across the whole fleet, every loan’s full guarantee terms, and a causal test of owner type during Uri.

The conclusions would change with corrected unit ownership and effective dates, a different defensible capacity denominator, undisclosed guarantees or offtake arrangements, or complete contemporaneous outage and fuel records. Filling those gaps requires the documents behind the corporate names.

The practical accountability question is now twofold: who can decide what happens to each plant, and which balance sheet must pay when it does not work out? Texas’s public loan portfolio makes the second question part of electricity policy even where the first answer remains a private company.

The guarantee matters as much as the owner

State loans deserve comparison by their collateral, parent guarantees and cash repayment terms before a ranking of sponsors. Cheap credit has value even when every principal dollar is repaid. The public’s strongest protection is a claim that still works when the project doesn’t.