A payment and a cheaper kilowatt-hour can both be real. They become one story when the published record is full of kilowatt-hours and thin on receipts. The owners of the model can read that record before the model does. Shipping the model ships the record’s center of mass. Policies defended with that center of mass then move money. Some of the movement is in the documents. Profit, on every account below, is not.
Data window · OpenAI closing cap table 28 Oct 2025 · Borenstein and Davis, National Tax Journal, Jan 2025 · IEA World Energy Investment 2025, values estimated · California Public Advocates Office 22 Aug 2024 · LBNL solar-adopter incomes, installations through 2023 · EU ETS revenue page, through 2025 · RGGI proceeds invested in 2023 · CAISO curtailment through 31 Dec 2024 · Energy Institute Statistical Review 2026, data year 2025 · reconciled 2026·11·04
01The question, kept intact
The economic question is who acquires wealth, assets, market power, and government-guaranteed revenue when an energy policy is adopted, and who pays. Whether the environmental case is sincere is a separate question. A coordinated psychological operation is a separate question again. The recipients of money can be named from the flows. Naming them does not require a plot.
That question has five blanks. A document that fills one blank has not filled the others. The ledger in this inquiry is the list of blanks, filled only where a pinned source fills them.
02Who held the equity on one day
On , OpenAI published the cap table at the closing of its recapitalization. The OpenAI Foundation held of OpenAI Group. Microsoft held . Current and former employees and investors held . The foundation appoints the board of the public-benefit corporation. Those are the company’s figures for that closing. The three stated shares sum to 100, and Microsoft’s share is given as roughly 27 percent. Later financing can dilute them. This page does not replace the closing table with a secondary report from 2026.
One dated cap table is the ownership fact this inquiry will stand on. Cloud companies, chip companies, and foundations also finance the other frontier labs. A single table that ties a named lab shareholder to a named energy subsidy is not among the documents below. The connection this page does trace is about the record, and then about the policies.
03How a corpus picks an answer
The people who pay for the model can see what the published record emphasizes. They know the statistical result of feeding it, in the ordinary sense that a corpus has a center of mass and a model trained on the corpus repeats that center. Five steps are enough. None of them requires a secret instruction.
Greater documentation does not, by itself, mean proportional influence on a model, and an institutional consensus can be true. Rejecting the institutions does not prove a redistribution. The money still has to be traced, and the benefit still has to be measured. The failure this page is built to refuse is narrower: letting the available table decide the question.
04Rules the ledger enforces
A long exchange on this question kept changing the basis underneath the arithmetic. The rules below are the repair. Each one is a refusal the code actually throws.
05Five blanks, five accounts
Each row is one policy system. Each column is one of the blanks in the question. Measured means a pinned document states the quantity and this page repeats it. Partial means the document states an estimate, a projection, or a gross flow. Unmeasured means this page has no pinned figure. Recipient net profit is unmeasured on every row. That is a result of the sources, and the test suite locks it.
| System | Taxpayer cost | Consumer cost | Recipient revenue | Recipient net profit | Public benefit |
|---|
The federal tax-payment side of the household credits is . A Congressional Budget Office distribution for 2022 would fill that cell. The file did not return a table, so the conversation’s “about 70 percent” is left off the page. A credit share toward the top quintile is the distribution of the benefit. Who financed it, in this version, is open.
Pick a measurement. Then ask the ledger to add a second one. The refusal is the point. The two calculators underneath are methods with labeled inputs: a hypothetical wedge, and kilowatt-hours per $100 of a generation cost you type.
06Who received the household credits
Severin Borenstein and Lucas Davis, writing in the National Tax Journal in January 2025, examined U.S. household tax credits for heat pumps, solar panels, electric vehicles, and other clean-energy equipment. Since 2006, households had received . The bottom three income quintiles received of every $100. The top quintile received . The fourth quintile is , the . “About” is the paper’s word. For electric-vehicle credits, the top quintile received of every $100. That bound is open, so the model assigns no remainder to the other quintiles.
A household that receives a $100 credit has usually spent a larger sum of its own money on the equipment. The environmental effect of that purchase is a further measurement. The paper’s shares are the distribution of the tax benefit. They are silent on profit, and they are silent on which quintile’s taxes financed the credit. Several later changes to U.S. household credits are outside the study’s window. This page does not describe the 2026 statute as if the 2025 paper had measured it.
07A bill other customers pay
California’s older net-energy-metering tariffs credit households for electricity they export and reduce the share of system costs those households pay. The Public Advocates Office estimated the annual amount shifted onto customers without rooftop solar at by the end of 2024, from in 2021. The ratio of those two office estimates is . On the same estimate, to of an average non-solar household’s bill, across Pacific Gas and Electric, Southern California Edison, and San Diego Gas and Electric, was going to that compensation. The bar in the ledger stops at the top of the range. A midpoint would pretend the office published a point.
Lawrence Berkeley National Laboratory’s 2024 update, covering installations through 2023, puts the median income of that year’s solar adopters at . The median for all U.S. households was , a difference of , which is and matches the laboratory’s stated gap. The median for all owner-occupied households was , a difference of . Against owner-occupiers in the same state, adopter incomes were about higher. of 2023 adopters had incomes below $100,000, and about were below 120 percent of area median income. The 2010 adopting cohort, measured in present-day incomes, had a median of . Adoption has broadened. Adopters remained richer than households at large.
This is the cleaner upward channel in the set, because the cost is on the electricity bill, where renters and non-adopters pay, and the benefit accrues to households that could acquire the asset. Solar-industry analysts dispute the office’s boundary, including whether electricity a household uses from its own roof is a cost shifted onto others. This page reports the office’s estimate. It does not recompute the counter-estimate, and it does not call either figure a profit.
08Investment is a different transaction
The International Energy Agency’s World Energy Investment 2025 estimated worldwide energy investment at in 2025. was going to renewables, nuclear, grids, storage, low-emissions fuels, efficiency, and electrification. was going to oil, natural gas, and coal. Of every $100 of that estimated total, is the clean bucket and is the fossil bucket. The 2025 values are the agency’s estimates. They are annual capital investment. They are a forecast of spending, and spending is someone else’s gross revenue. They are silent on tax redistribution and on profit.
Inside the clean bucket, solar investment was estimated at , which is of every $100 of that bucket. Battery storage for the power sector was , or of every $100 of the bucket. Adding either line to the $2.2 trillion counts the same dollars twice, and the ledger throws that addition out. Electricity-sector investment, a different cut, was estimated at . The agency describes that cut as some 50 percent higher than spending to bring oil, gas, and coal to market. This page does not recompute the 50 percent against the $1.1 trillion line. Those are different baskets, and the model keeps them apart.
09Two denominators, and energy that was not delivered
The Energy Institute’s Statistical Review, data year 2025, as reported by Carbon Brief on 30 June 2026, puts fossil fuels at of global total energy supply. Ember, a co-author of the review, puts 2025 electricity generation at solar, wind, and nuclear. The first figure is a share of total energy supply. The others are shares of electricity. Fossil supply includes energy that leaves a combustion plant as heat. Electricity from wind and solar is counted as the electricity itself. Subtracting 8.7 from 86.2 would treat those as one denominator. The ledger refuses.
A generation share is also silent on whether the electricity was delivered when it was needed. The California Independent System Operator’s curtailment report for 31 December 2024 sums local economic, local self-schedule, and system economic curtailment to for the year. The U.S. Energy Information Administration’s 28 May 2025 note rounds that quantity to , above 2023, with solar at of the curtailed energy. The increase and the solar share are EIA’s published figures. The megawatt-hour total is the sum of CAISO’s own categories. Curtailed electricity was available. It was not delivered.
Three claims stay separate. Solar and wind can produce competitively priced electricity in many places: the calculator above will turn a dollars-per-megawatt-hour figure into kilowatt-hours per $100, and this page has not pinned a 2026 cost table to feed it. Replacing fossil electricity while keeping reliability, at a lower total cost, is a stronger claim, and curtailment is one reason a generation cost does not settle it. Replacing fossil fuels across transport, heat, industry, and electricity is stronger still. An 86.2 percent supply share is the scale of what remains. A share is not a cost.
10What an auction collects
The European Commission reports that EU Emissions Trading System auctions raised in 2025, and from 2013 through the end of 2025. of the 2025 total went directly to member states. A floor is not a point, so the model will not allocate those euros per $100. Member states must use the revenue, or an equivalent financial value, for climate and energy action. That is a spending rule. It is silent on who bore the permit cost in prices, and silent on what the spending achieved.
The Regional Greenhouse Gas Initiative’s report on 2023 records invested in energy and consumer programs. The press release rounds the same investment to . Energy efficiency took of every $100. From the report’s projections of $2.7 billion in lifetime bill savings and 7.8 million short tons of CO₂ avoided, each $100 corresponds to of projected bill savings and short tons. Those are projections. They do not subtract allowance costs passed into electricity rates. A separate sentence in the release says about 36 percent of investments went to environmental-justice and equity programs. That cut can overlap the program categories. The button in the ledger tries to add it to the 64 percent, and the ledger refuses.
11What this version leaves open
Lazard published the 19th edition of its Levelized Cost of Energy+ report on 13 July 2026. This page did not open the cost table, so no technology is given a pinned dollars-per-megawatt-hour, and no capital-versus-fuel split is printed. The same restraint covers an NREL installed-cost breakdown, an IRENA capacity-factor arithmetic, a historical twenty-year adoption table, an energy-payback interval, and a manufacturing-concentration share. Each of those would answer a neighboring question. Neighboring is not the same question.
Four tests are the ones a replacement claim has to face. This version scores none of them with a number.
The net identity at the top of the page stays . Incremental energy and environmental value, and total incremental economic cost, are not both in hand for any account above. A hypothetical wedge with inputs of $100, $115, and $85 produces an investor advantage of and a social excess of . Those inputs are illustrative. Change them in the ledger and the signs follow the inputs. They are not a finding about any real plant.