Macro / Fiscal arithmetic
Growth or Grind
Net interest is 19.5% of federal receipts. Cheaper computation, useful space services and reliable power can lower that only by becoming taxable domestic income — and the debt identity says how much growth that would take.
By Stratifi Research
November 12, 2026
What would turn cheaper work into a smaller debt burden?
Treasury’s final statement puts net interest at 19.5% of federal receipts in FY2026. The calculation is $1,051.754 billion of net interest divided by $5,403.327 billion of receipts. It describes a budget burden, not a central bank running out of permission to act. The September decision to raise policy rates is an inconvenient fact for any story in which monetary independence has already disappeared. [1,2]
There is a constructive route through the arithmetic. Useful space services can reduce the cost of connecting, locating and coordinating activity. Reliable power can let productive capacity operate rather than wait for a connection. Cheaper computation can make more work economical and help create products that do not yet exist. Those changes can enlarge domestic income and, eventually, the tax base.
The word “eventually” carries a bill. Deployment needs financing. Government may fund early development. Organizations must change the way they work. Interest on existing obligations continues while the new income is still taking shape. The optimistic destination and the uncomfortable path can coexist. Lower useful-work costs remain the constructive destination. Broad deflation is an unscored directional view over the 2027–32 monitoring horizon, not a promised outcome within it or the near-term default. The earlier timing claim still faces the unchanged end-2027 core-CPI test below.
The fiscal promise is conditional. Deployment has to produce additional real output. That output has to become domestic taxable wages or profits, or a verified saving to a public agency. Government must retain a net gain after support and transition costs. Then the debt equation gets a vote. None of those steps is supplied by a market-size forecast.
The question to carry through this piece is simple: where does a real resource saving become taxable domestic income, and who keeps it?
The thesis this piece pursues.
Does growth faster than the interest burden solve it?
It helps. It does not settle the result.
On the final Treasury budget statement, FY2026 outlays were $7,395.800 billion, leaving a $1,992.473 billion deficit against receipts. Subtract net interest and primary borrowing was $940.719 billion. Growth has to contend with that borrowing as well as the inherited stock. These are complete fiscal-year dollar flows, not estimates for a partial year. [1]
There is no maturity deadline hiding in this problem. Treasury reported a weighted average maturity of 69.1 months on November 2, 2026, above its displayed 61.4-month mean from 1980 to the present. An average is not a refinancing calendar. It is evidence against building urgency from an unusually compressed maturity profile, not evidence that financing risk disappears. [15]
The simplified accounting is:
change in debt/GDP = [(r − g)/(1 + g)] × prior debt/GDP
− primary surplus/GDP + stock-flow adjustments/GDP
Here r is net interest divided by beginning budget-financing public debt; g is nominal GDP growth on a matching fiscal-year basis. A primary deficit is a negative primary surplus. Other financing belongs in stock-flow adjustments.
A current accounting example shows why the distinction matters. Treasury’s budget-financing public debt rose from $30,172.402 billion on September 30, 2025 to $32,245.399 billion on September 30, 2026. Divide by fiscal GDP constructed from BEA’s four quarterly annual rates—not a single quarter—and the debt ratio still rose despite favorable growth arithmetic. [1,3]
FY2026: the budget flows behind the 19.5% interest ratio
| FY2026 accounting (constructed) | Result |
|---|---|
| Debt/GDP, prior → current | 98.997155% → 100.030815% |
| Net interest / beginning budget-financing public debt | 3.485815% |
| Nominal fiscal GDP growth | 5.766170% |
| Primary borrowing / fiscal GDP | 2.918273% |
| Other financing / fiscal GDP | .249799% |
Net interest of $1,051.754 billion against receipts of $5,403.327 billion is 19.465% of receipts. Total borrowing less net interest leaves primary borrowing of $940.719 billion. The table applies the debt identity to the same fiscal year: debt/GDP still rose despite nominal growth above the net burden, because primary borrowing and other financing were larger.
Formula: change in debt/GDP = [(r − g)/(1 + g)] × prior debt/GDP − primary surplus/GDP + stock-flow adjustments/GDP. Fiscal GDP is the mean of the four quarterly annual rates Q4 2025–Q3 2026.The debt-ratio construction is 100 × 32,245.399 / 32,235.46575. Fiscal GDP is the mean of Q4 2025 through Q3 2026, including an advance estimate; revisions matter. The interest measure is a net budget burden on a specified stock. Primary borrowing and other financing explain why growth exceeding that burden did not stabilize the ratio. This is accounting, not evidence of a technology dividend. [3]
How much growth would be enough?
There is no single answer independent of fiscal policy and financing.
Take an explicitly hypothetical example, not a calibration of the United States today. Assume debt at 100% of GDP, an average nominal financing rate of 4.5%, a primary deficit of 2.7% of current GDP, and GDP-price inflation of 2.4%. Ignore stock-flow adjustments in this illustration. The constant-ratio condition is:
required nominal growth = (r × d + primary deficit share)/(d − primary deficit share)
= (.045 + .027)/(1 − .027) = 7.399794%
required real growth = 1.07399794/1.024 − 1 = 4.882612%
Those are demanding assumed conditions. They are not an observed national speed limit. Fiscal adjustment changes the requirement. At assumed nominal growth of 6%, the same interest/debt inputs permit a primary deficit of 1.415094% of GDP: ( .06 − .045 ) / 1.06. Faster output and smaller primary borrowing can work together. [4]
This also explains why cheaper products are not automatically debt relief. More useful output raises real income. Lower aggregate price growth alone can lower nominal income growth before existing obligations reprice. More exactly:
1 + nominal GDP growth = (1 + real GDP growth) × (1 + GDP-price inflation)
A benchmark price decline, consumer inflation and the GDP deflator are different measures. Technology can improve welfare and still leave the budget worse off. Conversely, sufficiently strong real output and retained primary gains can improve debt dynamics without requiring lower yields or monetary subordination.
Favorable net-burden arithmetic does not erase primary borrowing
Each line is the stabilization frontier for one assumed effective interest rate: the primary borrowing, as a share of GDP, that leaves the debt ratio unchanged at a given nominal growth rate, with debt at 100% of GDP and no stock-flow adjustment. At r = 4.5% and a 2.7% primary deficit, required nominal growth is 7.40% (4.88% real at 2.4% GDP inflation). At 6% nominal growth the frontier absorbs only 1.42% of GDP. These are analytic nodes, not interpolated history and not a forecast.
Formula: pd_max = d × (g − r)/(1 + g); g_required = (r × d + pd)/(d − pd). Nominal growth: 1 + g = (1 + real growth) × (1 + GDP-price inflation). Marginal yields are not effective budget rates.Where could the additional output come from?
Start with services already doing useful work.
BEA measured $142.526 billion of US space value added in 2023, against a fixed historical GDP vintage of $27,720.7 billion. The share is .514150%: 100 × 142.526 / 27,720.7. That is a real production base, but it is not large enough to justify skipping the downstream-income question. Regular official space accounts were discontinued in March 2026, so a current launch count cannot supply a current sector GDP estimate. [5]
A hypothetical one-time doubling relative to a flat-sector counterfactual adds .514150% to the GDP level, not that much extra growth every year. That scale check pushes the argument toward useful customer output, rather than an ever-larger supplier revenue forecast. [5]
The opportunity becomes more interesting when a satellite service helps a customer produce something additional: connectivity where terrestrial service is costly, better monitoring, less uncertainty in logistics, more resilient sensing. The supplier’s value added and the customer’s additional output can both matter, provided intermediate charges, displaced activity and implementation costs are reconciled. The customer’s entire revenue is not a benefit attributable to space.
Does it matter if government is the customer?
Yes, but not because public output is inherently worthless. Government can finance research, infrastructure and operational learning that private customers would not initially support. It can also commit to expensive capability without producing a fiscal return. The test is useful delivery, subsequent demand and cost per accepted capability, followed by a separate budget ledger.
BEA’s private-industry space category includes government purchases from private suppliers. It therefore cannot identify the privately funded share. An agency’s agreement ceiling is not this year’s production, outlay or tax receipt. A global revenue forecast is not US value added. Those distinctions stop an attractive industry story from quietly becoming an invented fiscal multiplier. [5,6]
The near-window pathway does not require booking orbital-compute income, space power or lunar-resource rents. Those later possibilities need repeat useful delivery and competitive full costs; a paper, permit or prototype is not commercial arrival.
Producer classification does not tell us who paid
| 2023, USD bn | Value |
|---|---|
| Private-industry space value added | 106.924 |
| Government-industry space value added | 35.602 |
| Broad space value added, share of GDP | 0.514% |
| One hypothetical doubling, GDP-level effect | +0.514% |
BEA measured $142.5 billion of broad space value added in 2023, about 0.51% of GDP, with real value added roughly flat since 2015. The private/government split classifies producers, not payers: private-industry value added includes sales to government. A one-time doubling relative to a flat-sector counterfactual adds half a percentage point to the level of GDP, not to its growth rate.
Source: BEA space economy satellite account workbook, Table 1, March 31, 2025 vintage.What does reliable power change?
It can remove a binding complement. A factory or computing facility that cannot obtain reliable delivered power cannot convert its other capital into useful output. Avoiding that constraint may matter more than a modest change in the average electricity bill.
The nearer candidates are existing-site preservation, uprates and restarts. DOE’s October 27, 2026 UPRISE fact sheet identified about 4.7 GW toward a 5 GW goal by 2029. That is an identified pipeline and a target, not an operating increment. A construction permit authorizes construction; it does not establish an operating fleet or the learning curve of repeat units. [7]
To see the scale without inventing a commissioning forecast, assume the full 5 GW is operating at a 90% capacity factor. It would produce 5 × 1,000 × 8,760 × .90 / 1,000,000 = 39.42 TWh per year, about .890842% of the 4,425.029 TWh of utility-scale generation EIA reports for 2025. Both the commissioning and capacity-factor conditions must be visible. [7,8]
Then ask what connects those MWh to income. Is the plant connected? What is the financed delivered cost after fuel and network charges? Does the customer expand useful production, or displace another domestic producer? Are preserved operations being mistaken for wholly new capacity?
The real-growth contribution of power in percentage points is unestimated here. The physical scale calculation is not a GDP coefficient. Nor can a gain in the same factory be counted independently as a nuclear dividend, an AI dividend and a satellite-logistics dividend. These are complements in one production process.
A power pipeline is not an operating fleet
| Item | Value | Status |
|---|---|---|
| Identified potential capacity (DOE UPRISE) | 4.7 GW | Identified; not an operating addition |
| Agency capacity goal, 2029 | 5 GW | Target; not delivered capacity |
| Kemmerer construction permit (NRC CPAR-1), March 9, 2026 | permit | Construction authorized; not operation |
| 4.7 GW at an assumed 90% capacity factor | 37.1 TWh/yr | Conditional; assumes full delivery |
| 5 GW at an assumed 90% capacity factor | 39.4 TWh/yr | Conditional; assumes full delivery |
| Utility-scale generation, all sources, 2025 | 4,425 TWh | Denominator (EIA) |
The identified capacity would add under one percent to annual utility-scale generation even if every megawatt operated at a 90% capacity factor. Preservation, uprates and restarts are the nearer candidates; the real-growth contribution of power is left unestimated here because connection, delivered cost and financing are what turn megawatt-hours into income.
Sources: DOE UPRISE fact sheet (October 27, 2026); NRC construction permit CPAR-1; EIA Electric Power Annual (2025 table).How does cheaper computation become more income?
Through adoption, not by declaration.
A useful task must become cheaper after integration, verification, rework, energy and capital costs. Organizations then need to change workflows, find customers and produce additional accepted output. New products and scientific or organizational invention could make the gains much larger than a task-replacement model suggests. That is the serious upside case.
There is deployed-workflow evidence. NBER’s Generative AI at Work, displayed revision November 2023, finds improved customer-support performance. It supports the mechanism, not national breadth or an economy-wide net-cost coefficient. [9]
The history is also more informative when its denominators survive the chart. Penn Wharton’s September 2025 research includes native observations on earlier digital adoption. Worker use of computers or the internet differs from employment at cloud-adopting firms. A job-posting proxy for smartphone diffusion is not a count of smartphone users. Household ownership and adult use are different again. Those observations can be shown in separate panels; they cannot be pooled into a causal curve from adoption to GDP or taxes. [10]
PWBM’s model projects a 1.5% increase in the GDP/TFP level by 2035, not an extra annual growth rate of that size. The model explicitly leaves out new products, new tasks and possible effects on invention. Its modest result is a reason to demand evidence for a large near-term fiscal claim, not a ceiling on what technology can eventually accomplish. This is a single attributed projection, not a calibrated confidence band or part of an additive sector forecast. [10]
Historical adoption measures do not share one population
Fifty-six published points across nine denominators: workers reporting use at work, and people reporting use outside work, for the PC, the internet, the smartphone, cloud computing and generative AI. The populations differ, so the dots are not one adoption curve and support no speed or causal comparison. Generative AI’s first readings, 26.4% at work and 33.7% outside work in 2024, sit above where the PC and the internet began.
Source: Penn Wharton Budget Model, Figure 3, attributing the Current Population Survey computer and internet use supplements, Pew surveys and Census ABS. Calendar year = anchor year + years since introduction.Cheaper capability reaches GDP through diffusion
One deployed-workflow finding and one published model projection, kept in their own units. NBER’s Generative AI at Work studies 5,179 support agents in one deployment and finds improvement concentrated among less experienced workers. Penn Wharton’s model projects a cumulative 1.5% rise in the TFP level by 2035, arriving as a contribution to annual TFP growth that peaks near 0.2 percentage points in 2032 and fades, not an extra annual growth rate of 1.5%.
Sources: NBER working paper 31161; Penn Wharton Budget Model, Projected Impact of Generative AI on Future Productivity Growth, September 8, 2025, Figure 5.Does Jevons guarantee more revenue or energy demand?
No. Let h be the ratio of the new final service price to the old price, epsilon demand elasticity and e the resource intensity per useful service ratio:
quantity ratio = h^(−epsilon)
nominal spending ratio = h^(1−epsilon)
resource-use ratio = e × h^(−epsilon)
Resource use rises only if epsilon > ln(1/e)/ln(1/h). The familiar threshold of elasticity above one applies only when price and resource intensity fall proportionally with full pass-through. We have not measured the national elasticity for these channels. More tokens can coexist with less spending, more verification work or no additional accepted output. [4]
AI can be useful in fundamentals even if its current architecture is overbuilt. A particular investment cycle failing is different from the capability layer failing to lower useful-work costs. Neither technical success nor a capex bust decides the fiscal outcome without the next links.
More use need not mean more spending
With the price ratio h at 0.5 and resource intensity e at 0.5, quantity rises at every elasticity, but nominal spending and resource use fall unless demand elasticity exceeds one. Resource use rises only if the elasticity exceeds ln(1/e)/ln(1/h), which is 1 here; the familiar Jevons threshold of elasticity above one applies only when resource intensity is unchanged.
Formula: quantity ratio = h^(−ε); nominal spending ratio = h^(1−ε); resource-use ratio = e × h^(−ε). Illustrative identities; no empirical elasticity is measured.Who gets the income, and who retains the budget gain?
Domestic production is not the same thing as a domestic headquarters. The location of workers, production, intellectual-property income and tax residence matters. Wages, profits, losses, credits and investment expensing affect how additional output becomes receipts. Displaced payroll and transition support can offset part of the gain.
Fiscal retention is another decision. If government spends all additional receipts or expands the mission whenever capability gets cheaper, the public benefit may be valuable without reducing borrowing. A lower agency cost per delivered service is not a budget saving if the saved money is immediately committed elsewhere.
The ledger must assign each cost once. If an estimate’s retained-receipts assumption already bundles transfers or tax leakages, subtracting them again understates the result. A loan’s face value is not its immediate subsidy cost. Credit risk, subsidies, recoveries and financing belong on the correct budget basis.
Geopolitics can affect this sequence. Allied customers can expand the market for US-produced services and spread development costs. Reliable energy exports, standards and access can support domestic income. But a foreign procurement commitment is not a transfer to the US Treasury, and local production or larger security commitments can absorb the gain.
Dollar networks can support transaction demand without guaranteeing an equivalent concession on Treasury financing. Federal Reserve research distinguishes dollar convenience from Treasury convenience. It supplies a historical warning against equating usage with cheap sovereign debt, not a current causal funding discount to subtract from the interest bill. Sanctions can serve an objective while inviting substitutes, retaliation or fragmentation. [11]
The real-growth contribution of geopolitical rents in percentage points is unestimated. No separate GDP bonus or funding subsidy is booked. Space exports already counted in domestic production cannot be counted again under a geopolitical heading. From Brussels, expanding security capability can be compatible with sourcing locally. From Beijing, successful substitute technologies could expand global abundance while reducing US rents. That is a live countercase, not a claim about hidden coordination.
What would count as success?
Four different things, which need four different labels.
Annual fiscal relief means the year’s retained incremental receipts and verified public savings exceed that year’s incremental support and transition costs. It does not repay earlier losses.
Cumulative payback adds earlier costs and benefits from a reference starting baseline and includes financing exactly once. A project can have a positive latest year while the accumulated public bill remains unpaid.
Lower debt relative to baseline means a better outcome than the same reference counterfactual. Both paths may still be deteriorating.
CBO’s February 2026 baseline projects a rising debt ratio
Debt held by the public was 99.4% of GDP at the end of fiscal 2025 and reaches 120.2% by 2036 on CBO’s February 2026 baseline. This is the frozen native projection, not a current actual and not a technology attribution; the article’s fiscal tests are measured against a baseline of this kind, not against zero.
Source: CBO, The Budget and Economic Outlook, February 2026, dataset 51118 (debt held by the public as a share of GDP).Actual stabilization means debt/GDP stops rising on a defined basis. It is not another name for the previous three outcomes.
Actual technology-attributable relief and cumulative public payback remain unmeasured. The conditional case stands on useful deployment and retained income, not the following calculation.
Two assumed adoption cases against the same frozen baseline
| Fiscal test, through 2032 | Broader adoption | Strong diffusion |
|---|---|---|
| Annual net primary relief in 2032, $bn | +15.558 | +112.662 |
| Cumulative primary balance, $bn | −116.726 | +24.174 |
| Cumulative financing-inclusive balance, $bn | −125.401 | +14.296 |
| Debt/GDP versus assumed baseline 115.039% | 112.942% | 110.132% |
| Final-year change in debt/GDP, pp | +1.931 | +.922 |
Annual relief in 2032 is positive in both assumed cases, but only the strong-diffusion case reaches cumulative payback by 2032, and neither stabilizes the debt ratio: both end the period with the ratio still rising. The four outcomes in the text are different tests, and a case can pass one while failing the next.
Analyst assumptions; financing-inclusive balance adds interest on the cumulative position. Debt/GDP compared with an assumed baseline of 115.039% in 2032.What is the strongest objection?
Technology may work while the fiscal promise fails.
A better service can be free or unpriced. Income can accrue abroad or to owners with offsets against taxable profit. Displaced wages can lower receipts while transition spending rises. Legislatures can consume the gain. Productive investment can raise real financing costs enough to erase a relative debt benefit. General price deflation can increase the real burden of fixed nominal debt before funding costs adjust.
There is an objection to excessive caution, too. Narrow task models may miss invention. Public procurement can seed an industry. Removing a binding power constraint can unlock valuable customer output far beyond a simple electricity-bill saving. Fiscal adjustment can improve the primary balance independently of technology. A currency issuer has choices unavailable to a household.
Those are reasons to preserve the constructive pathway and test it. They are not reasons to assign each channel an invented growth contribution. The realistic argument is stronger when the missing coefficients stay missing.
What would change the view, and when?
Keep the earlier test on its original horizon. A new monitoring window cannot erase an unsuccessful dated claim.
For the original end-2027 price-size/timing test, we use core CPI from BLS series CUUR0000SA0L1E, December 2027 over December 2026. The accepted idea is “well below 2%”; the operational rule is 1.5% or below as support and 2.25% or above as evidence the estimate was too large or early under its stated no-further-cuts premise. The interval between is inconclusive. The review date is February 15, 2028, without changing the observation deadline. Report separately whether the policy target ceiling was reduced between November 6, 2026 and December 31, 2027. A changed premise does not authorize claiming conditional success or moving the deadline. [14]
For the productivity bridge, use year-on-year nonfarm-business output per hour: 100 × [OPHNFB(q)/OPHNFB(q−4) − 1]. The criterion is growth above 2.5% in each quarter from Q1 2026 through Q4 2027, together with the labor-share index PRS85006173 higher in Q4 2027 than Q4 2025. Its review date is April 30, 2028. Revisions to one quarter do not count as extra quarters. Rising labor share is a house bridge criterion, not a mathematical requirement for real growth or receipts. [14]
For end-2027 monetary independence, retain the conjunction of appropriate policy, anchored inflation compensation and issuance lengthening. The test uses every monthly mean of T10YIE from November 2026 through December 2027 below 2.5%, December 2027 WAM above November 2026 WAM, and documentary policy review. Its review date is February 15, 2028. Broader financing-support and compelled-absorption review markers still apply; temporary market-functioning support has alternative explanations. [14]
The squeeze test is FY2028 net interest/receipts at or below 18%, without recession-driven rate relief, checked December 15, 2028. The separate end-2028 warnings and end-2029 delivery audit remain in the reference protocol, not replacements for the end-2027 deadline. They require matched output, fiscal and operating-fleet definitions; missing inputs cannot pass a test. [14]
Net interest as a share of receipts, with a visible source break
The ratio peaked near 18% in the early 1990s, fell to about 6–8% through the 2010s, and the final FY2026 Treasury statement puts it at 19.5%. The FY2026 point comes from a different source than the CBO history and is not joined to it; FY2025 differs between the two sources by about $0.4 billion of net interest. The 18% line is the FY2028 test in the watchlist.
Formula: 100 × net interest / receipts, fiscal years. Sources: CBO historical budget data (February 2026 vintage); Treasury Bureau of the Fiscal Service, final September 2026 MTS.By end-2029, require actual service delivery, payer-separated incremental demand, connected MWh and full-cost workflow evidence. Until a representative audit frame exists, this is a qualitative no-upgrade test, not a national pass from selected success stories.
For FY2031 and FY2032, the central fiscal test is positive attributed annual net primary relief in both years, with the sign surviving disclosed cost and counterfactual uncertainty, plus a measurable improvement in the debt path against the reference baseline. Separately publish the cumulative financing ledger, the ratio difference from the reference baseline and the actual annual debt-ratio change. Use December 15, 2033 as the fixed check date for data through FY2032; unavailable causal attribution stays unmeasured. No materiality claim is earned without a predefined error threshold. If financing or spending absorbs the benefit, the fiscal claim fails even if useful technology succeeds. [14]
What to watch
Follow useful delivery, domestic income and retained public savings alongside financing conditions. Keep alternative outcomes unweighted: an enacted fiscal turn under market pressure; documented financing accommodation; measured growth with annual capture; or residual grind, which includes capital-only gains, delayed diffusion, supply shocks and capex disappointments. These mechanisms can coexist; their order here is not a classification rule. Missing higher-priority evidence leaves classification unresolved, never defaulting to grind.
Abundance remains the destination worth understanding. The fiscal result has to be earned through deployment, output, taxable income and retention. A useful technology does not owe the Treasury a rescue on schedule.
Sources
1
Treasury, final September 2026 Monthly Treasury Statement, Tables 1, 6 and 9. Net interest share = 100 × 1,051.754 / 5,403.327; primary borrowing = 7,395.800 − 5,403.327 − 1,051.754. Net, not gross, interest.
2
Federal Reserve, FOMC statement of September 16, 2026. A dated counterexample, not a full current-policy review.
3
Treasury MTS plus BEA GDP via FRED, October 29, 2026 vintage. Fiscal GDP is the mean of the four quarterly annual rates Q4 2025–Q3 2026 ($32,235.466 billion; prior year $30,478.050 billion). Debt ratios use budget-financing public debt, not the separate Treasury-par numerator; the table is not spliced to CBO projections.
4
Analyst algebra and sensitivity: d = 1, r = 4.5%, primary deficit 2.7%, GDP inflation 2.4%, no stock-flow adjustment. Stabilizing g = (rd + pd)/(d − pd); real growth = (1 + g)/1.024 − 1; allowable pd = d(g − r)/(1 + g). Rebound identities carry unestimated empirical elasticities.
5–6
BEA, U.S. Space Economy Satellite Account, March 31, 2025 release, with the discontinuation notice and the fixed historical GDP vintage. Share = 100 × 142.526 / 27,720.7. Government procurement and the ultimate payer are not industry classifications.
7–8
DOE, Project UPRISE fact sheet (October 27, 2026); NRC construction permit for Kemmerer (March 9, 2026); EIA Electric Power Annual, 2025 generation table. Conditional output = 5 × 1,000 × 8,760 × 0.90 / 10⁶; share = 100 × 39.42 / 4,425.029. Assumed operation and capacity factor, not commissioned production.
9–10
NBER working paper 31161, Generative AI at Work (November 2023 revision); Penn Wharton Budget Model, The Projected Impact of Generative AI on Future Productivity Growth, September 8, 2025. The micro evidence is one deployment; PWBM is an attributed model, and the historical adoption observations keep their distinct denominators.
11
Federal Reserve, The International Role of the U.S. Dollar, 2025 edition (July 18, 2025); IFDP 1427 on the historical convenience yield (December 2025); Treasury 2021 sanctions review. Mechanisms, not a current measured funding concession.
12
Analyst calculations for the two assumed adoption cases in Exhibit 10, from the archived year-by-year illustrative model and its documented equations. Annual net relief = retained receipts − support; the financing-inclusive ledger compounds at the assumed GDP-price inflation of 2.4%. Baseline and cases use assumed parameters; no measured sector tax coefficients.
13
CBO, The Budget and Economic Outlook: 2026 to 2036 (February 2026), ten-year budget projections from CBO’s open-data repository; debt held by the public as a share of GDP.
14
Watchlist series: BLS core CPI (CUUR0000SA0L1E); nonfarm-business output per hour (FRED OPHNFB); labor share (FRED PRS85006173); federal funds target ceiling (FRED DFEDTARU); 10-year breakeven inflation (FRED T10YIE). Check dates and thresholds as stated in the text.
15
Treasury, Q4 2026 presentation to the Treasury Borrowing Advisory Committee (November 2, 2026), p. 23: two attributed maturity observations, no inferred reset schedule.