What an AI boom would do to yields

Key points

Faster expected growth raises real interest rates. People who expect to be richer need a higher return to save, and across countries higher expected growth goes with higher long-term real rates.
How much depends on belief and on the Fed. If investors fully believed Anthropic’s substantial scenario today (GDP 56.9% higher after ten years), the ten-year real yield would rise by 463 bp; with a 50% chance by 205 bp; in the Fed’s FRB/US model, where policy and prices respond, by 123 bp.
Today’s yields do not show an AI boom priced in. The rise since 2021 mostly reflects expected Fed policy and risk premia, and surveyed economists still expect modest growth.
The budget gets worse before it gets better. About 35% of Treasury debt matures or resets within a year, so higher yields raise interest costs before extra tax revenue arrives: −$288bn a year after three years in the substantial scenario, even as the debt ratio falls to 89% by 2036.

Why consumption growth can raise real rates

If people expect to be much richer in the future, an extra dollar then matters less to them than an extra dollar today. They need a higher real return to make saving attractive. The Ramsey rule captures this idea:

r = ρ + \frac{g ^{c}}{ψ}

The real rate equals impatience plus consumption growth divided by the EIS.

Here r is the real rate, ρ measures impatience and gᶜ consumption growth per person. The Elasticity of Intertemporal Substitution or EIS (ψ) measures how strongly real rates affect when people spend. Chow and coauthors cite typical EIS assumptions of 0.2–2. We use 0.735: one extra point of consumption growth raises real rates by 1.36 points, other things equal.

From GDP growth to consumption growth

AI could raise GDP in three ways: by doing tasks people do today, by making workers more productive, and by speeding up new discoveries. If that happens, incomes rise and the economy grows faster.

Today, most of AI's effect on the economy comes through spending on data centers, chips and power. That spending counts as investment, not as what households buy: the St. Louis Fed estimates that AI-related investment added 1.0 percentage point to US GDP growth in the first nine months of 2025, 39% of the total, before netting out imported equipment.

Consumption can follow later in two ways. Once the new capacity is in use, it can raise output and wages. And as AI companies' shares rise, the households that own them feel richer and spend more.

Estimates differ by horizon and rate measure

Real-rate change per 1 pp of growth

Macro-model neutral rateYield less expected inflationTIPS-based real rates

Rate and growth horizons differ; six quarters is a response lag. Bars show available 95% intervals. TIPS are inflation-protected Treasuries.

Source: Chow, Halperin and Mazlish - Transformative AI, existential risk, and real interest rates, Eggertsson, Mehrotra and Robbins - A Model of Secular Stagnation, Holston, Laubach and Williams - Measuring the Natural Rate of Interest after COVID-19, Christensen and Rudebusch - A New Normal for Interest Rates? Evidence from Inflation-Indexed Debt, SPF mean real GDP level forecasts, Federal Reserve Board - TIPS yield curve (FEDS 2008-05), Borio et al. - Why so low for so long? (BIS WP 685) · 2026-09-30

How yields move with growth-forecast revisions

Ten-year yields; bp per 1 pp year-ahead GDP revision

Full samplePre-Covid

Same survey interval; dots and 95% intervals. Full sample has 97 revisions (1999–2023); pre-Covid has 71 (1999–2016). Associations, not announcement effects.

Source: SPF mean real GDP level forecasts, Federal Reserve Board GSW nominal yield curve, Federal Reserve Board - TIPS yield curve (FEDS 2008-05) · 2026-09-30

What the empirical evidence supports

Chow, Halperin and Mazlish (2026) compare expected GDP growth and real forward interest rates five to ten years ahead across 59 countries. They estimate 1.36 points higher rates for each extra point of expected GDP growth. We use the implied EIS of 0.735; applying it to US consumption and to other maturities is our assumption.

Intuitively, better AI models should lead to higher growth expectations. However, Christensen and Rudebusch (2026) examined 26 AI model releases between November 2022 and September 2025 and found that estimates of the neutral real rate, r*, fell by a cumulative 23 bp to 35 bp across three-day windows around the releases. Their analysis did not measure whether those releases changed growth expectations.

More generally, better AI models only translate into enterprise productivity gains through useful applications and adoption, often requiring new workflows and systems. AI capabilities are often measured using benchmark scores, many of which are not aligned with economically valuable work. Many companies are still catching up with capabilities already available at the frontier, leaving a large gap between what AI can do and how widely it is used.

Nominal yields show the clearer relationship

For a shorter horizon, we compare how professional forecasters revised next year's US growth forecast with how Treasury yields moved at the same time (1999–2023). A one-point upward revision coincides with a +19 bp change in nominal ten-year yields. Of this, +4 bp is in real yields and +15 bp in inflation compensation. Pre-Covid estimates are larger; real-yield estimates are statistically indistinguishable from zero in both samples.

Chow and colleagues explicitly distinguish short-run monetary effects from the long-run growth relationship. Higher real rates can restrain near-term growth, while persistently faster expected consumption growth can raise equilibrium real rates. Revisions to next year’s growth forecast can reflect demand, an economic rebound or productivity news without implying lasting consumption gains. Their long-run finding and our noisy shorter-horizon result can therefore coexist.

The gap between ordinary Treasury yields and yields on inflation-protected Treasuries (TIPS) is not just expected inflation. It also includes extra pay for the risk that inflation surprises, and for TIPS being harder to buy and sell.

Start with Anthropic’s GDP scenarios

In a recent economic paper, Anthropic laid out three scenarios for GDP growth: modest, substantial and extreme, with GDP 1.6%, 8.3% and 32.4% above its no-AI path at the start of 2030. The paper lays out additional GDP and does not assign probabilities to the scenarios.

In this note, we illustrate how different GDP scenarios would affect the bond market. We do not predict GDP growth ourselves; we use third-party forecasts. In a future note, we will examine GDP growth projections and why we think current methods fail to fully recognise AI's contribution.

We extend each scenario past 2030, as our own assumption rather than Anthropic’s, by letting its adoption curves keep running: how much work AI can reach, how fast firms take it up, and how much each task gains. We count only gains still to come after our 18 September 2026 valuation date; for the substantial scenario that is 7.2% by 2030. The AI-use chart shows how much work uses AI in the substantial scenario, with a range for physical work. These paths do not tell us how much growth investors already expect.

Each scenario keeps its own path

Anthropic’s scenarios end in 2030 at very different points, so one shared rule after 2030 would blur them. The scenario chart instead runs each scenario forward on its own assumptions. How long the extra growth lasts still matters. If, in the substantial scenario, extra growth halved every three years after 2030, GDP would be 23.5% higher at ten years rather than 56.9%, and the ten-year yield would rise 250 bp rather than 463 bp.

The same model serves every maturity and the financing calculation that follows; thirty-year yields need GDP paths to 2056. The two-model chart sets our result beside the Fed’s FRB/US model, where a policy rule sets rates and prices and demand respond; its much smaller rise marks the other end of the range.

Because consumption trails GDP, short bonds rise least: until the delay has passed, consumption is no higher than without AI. A three-year delay, also our assumption, would lower the substantial ten-year rise to 381 bp.

AI use in cognitive and physical work

Substantial scenario; stacked shares of all work, weighted by labour cost

Cognitive workPhysical: lowerPhysical: lower–upper

Physical work is stacked above the same cognitive-use path. Only physical adoption varies across the band; it is not a confidence interval. Our extension starts in 2030.

Source: Korinek Jones Sacher Cotter and McCrory - Economic Scenarios for Transformative AI, Grace and others - Advanced AI according to 1580 researchers (ESPAI 2024), IFR - Humanoid Robots Vision and Reality, Morgan Stanley - Humanoids A Five Trillion Dollar Market, Bank of America Institute - Physical AI part 2 Humanoid robots, Goldman Sachs - Robotaxis Are Forecast to Become a 400 Billion Dollar Market in 2035 · 2026-09-30

Anthropic's three scenarios, extended past 2030

Additional GDP remaining after valuation (%)

ModestSubstantialExtreme

Anthropic's paths to 2030, then each scenario's own adoption curves. Robots after 2030 are our assumption and follow the same slow pace in all three.

Source: Korinek Jones Sacher Cotter and McCrory - Economic Scenarios for Transformative AI · 2026-09-30

Translate GDP into consumption and yields

Our calibrated capital model determines investment and consumption along each scenario, shown in the same colours in the yield-effects chart. Extra consumption then trails by 2 years: the gains go mostly to owners of capital, who spend little of new wealth at first, and the build-out comes before the output it supports, so what households do not spend is invested. For a consumption path treated as certain and unchanged time preference, the Ramsey relationship gives:

Δ y_{T}^{R} = \frac{1}{ψ T} lo g (\frac{C _{T}^{A I} / C _{0}^{A I}}{C _{T}^{r e f} / C _{0}^{r e f}})

The real zero-coupon yield change equals additional annualised consumption growth divided by the EIS.

C is real consumption, T the maturity and ψ the EIS; with expected inflation and premia unchanged, nominal and real yield changes are equal.

After ten years, the substantial scenario adds 56.9% to GDP and 40.5% to consumption. The calculated yield increase is 463 bp, against 60 bp for modest and 1058 bp for extreme. Without the consumption delay, substantial would give 596 bp; other EIS values are in the appendix. Each figure is the rise in today’s yield if investors fully believed the scenario on 18 September 2026; the ten-year figure reflects consumption growth to 2036. Repriced at once, a ten-year zero-coupon Treasury would change in price by -37%, a five-year by -7% and a thirty-year by -67%, so holders of long bonds carry most of the risk; coupons and a gradual repricing would soften the loss. With a 50% chance of the scenario the ten-year rise is 205 bp; the Fed’s FRB/US model, where policy and prices respond, gives 123 bp (see the appendix).

Adoption, investment and uncertainty

Reach measures work AI can perform or assist with, weighted by pre-AI labour cost. Use measures deployment within reach; their product is the share of work using AI. Anthropic uses Claude activity as a proxy for starting reach. In the substantial case, 14% reach multiplied by an assumed 10% use gives 1.4% deployment. In 2030, 30% reach and 40% use give 12%.

Beyond 2030, physical work adds to the 62.4% cognitive ceiling. Anthropic leaves it out, so these assumptions are ours. In all three scenarios, robots gain capability five years after office AI; use starts at 1% within reach in 2030 and spreads at a quarter of office-work speed. In our faster case, robot use follows office use five years later. Morgan Stanley, Bank of America and Goldman Sachs support using a range, but do not determine its endpoints. Task efficiency rises 2.9% a year in our extension; applied to the substantial scenario, that faster case raises its ten-year GDP gain to 81.2%.

The investment calculation uses 2024 expenditure shares and estimates capital adjustment from 1990–2024 data. The 2010–2024 sample gives no stable estimate. We assign no probabilities to the three scenarios.

Our calculation assumes the extra output reaches household spending, whether owners spend it or policy redistributes it. With wages up only 0.5% in Anthropic’s extreme case, that is a strong assumption. We think such a shift would likely bring large tax and transfer changes, which we do not model. If owners instead saved most of their gains, the yield rise would be much smaller and could even reverse.

Yield effects of the three scenarios

Rise in today's zero-coupon yields if fully priced now (bp)

ModestSubstantialExtreme

Same scenarios and colours; EIS 0.735; inflation and premia fixed, so nominal equals real. Extreme is beyond what this calculation handles well. 30-year uses GDP to 2056.

How much would yields rise? Two models

Substantial scenario: rise in today's real yields (bp)

Full belief (our model)50% chance (our model)FRB/US (Fed model)

Ours: full foresight, inflation and premia fixed. FRB/US: Fed rule, prices and demand respond; slightly larger GDP path; real yield is nominal less expected inflation.

Source: Korinek Jones Sacher Cotter and McCrory - Economic Scenarios for Transformative AI, Chow, Halperin and Mazlish - Transformative AI, existential risk, and real interest rates, BEA via FRED - US capital, depreciation, investment and consumption, Federal Reserve - FRB/US model in Python, Federal Reserve - FRB/US illustrative baseline data · 2026-09-30

Higher yields need an explanation

Long-term US yields have risen sharply since the pandemic, while published long-run GDP growth forecasts remain modest. Yields reflect expectations for growth and monetary policy, as well as the compensation investors require to hold bonds. A higher yield alone cannot tell us which changed.

The ten-year real yield rose 322 bp from its 2021 average of -0.9% to 2.3% in August 2026. The Fed's D'Amico–Kim–Wei model attributes 160 bp to higher expected real short rates; term and liquidity premia account for the rest. The real-yield chart shows both yield and expected-rate histories. Expected rates include policy expectations and cannot be read as a direct measure of expected productivity growth.

Government debt is another possible influence, but its level alone cannot explain the timing of a repricing. Even an anticipated debt path can affect yields if demand or investors' willingness to bear risk changes. We need evidence of those changes before assigning the rise to fiscal pressures or an AI boom.

What the decomposition can tell us

Over the shorter window from the 2024–25 average to August 2026, the nominal ten-year yield rose 50 bp. The D'Amico–Kim–Wei model estimates a real-yield increase of 39 bp. It attributes 5 bp to expected real rates, 26 bp to the real term premium and 8 bp to TIPS liquidity. The model estimates the components of the yield rise without identifying their economic causes. Nominal-yield decompositions also differ. The San Francisco Fed's model assigns 15 bp more to expected nominal rates than Kim–Wright does.

The yield-driver chart compares separate yield sensitivities: 13 bp for energy, 8 bp for foreign official demand, 4 bp for government debt and 2 bp for Fed runoff. These combine an oil-price regression, published elasticities and scenario assumptions. Their effects may overlap, so adding them would not explain the rise. Their ranges are not comparable confidence intervals. The methods specify the changes and measurement windows behind each estimate.

We set the central effect of AI borrowing to zero; this does not establish that it has no effect. In our earlier note, Crowding out, we found that AI borrowing affects hyperscaler credit pricing but has added little to Treasury yields so far. Neither these sensitivities nor the yield-component models identify how much AI-driven growth investors have priced.

Real yields rose alongside expected real rates

Ten-year horizon; monthly D'Amico–Kim–Wei estimates, %

Expected real short rateReal yield (TIPS)

The gap contains real term and TIPS liquidity premia. Expected real short rates also reflect policy expectations; neither line isolates AI growth. Data through August 2026.

Source: Federal Reserve Board, Tips from TIPS: Update and Discussions · 2026-09-30

How other factors could affect yields

Separate calculations for nominal ten-year yields (bp)

Oil-price regression, published elasticities and assumptions. Effects may overlap; ranges are not comparable confidence intervals. These do not decompose the observed rise.

Source: EIA via FRED - Daily Brent crude spot price, Federal Reserve Board, Tips from TIPS: Update and Discussions, Bhatt, Diercks, Eyal and Skaperdas - Federal Debt and Interest Rates (FEDS 2026-031), CBO - The Budget and Economic Outlook: 2026 to 2036, PIMCO, The Credit Market Lens: AI Capex and the Limits of Crowding Out (Lotfi Karoui), Bank for International Settlements, Quarterly Review, September 2026 · 2026-09-30

Forecasters still expect modest growth

In a recent Forecasting Research Institute paper, economists’ median forecast is 2.5% annual GDP growth over 2025–29, rising to 3.3% conditional on rapid AI progress. Eight of nine official and private forecasts compared by FRI range from 1.9% to 2.1%; OMB is 2.8%. Our reconstruction of Anthropic’s substantial scenario averages 3.5%. The growth-forecast chart compares these averages, not our model projections. The sources cover different periods and use either real or potential GDP.

Philadelphia Fed forecasters lowered ten-year GDP growth from 2.3% in 2021 to 2.1% in 2026. Their productivity forecast rose to 1.82% from 1.56% in 2024–25 (productivity chart).

More recent realised data show nonfarm business productivity rising 2.2% over the year to the second quarter of 2026, although this does not establish an AI-driven acceleration.

Our view is that surveyed economists' central forecasts understate AI's likely contribution to GDP growth. FRI's review finds repeated underestimation of progress on AI benchmarks, although the evidence on adoption and economic impacts is less conclusive. We expect productivity gains to strengthen as firms integrate rapidly improving AI capabilities into their workflows. How quickly and how far adoption catches up remains uncertain.

Policy expectations complicate the AI reading

Beaudry, Cavallino and Willems (2026) find that windows around payroll releases and speeches by senior Fed officials cover 23.9% of trading days over August 2020–3 September 2026, but contain 90.5% of the nominal ten-year yield increase. Even the five-to-ten-year forward rate rose mostly in these windows (91.3%). The ten-year yield has recently moved with the news that usually sets short-term rates, more than with slow structural forces such as AI-driven growth. Timing alone cannot separate the two: payroll news is also growth news, and policymakers’ view of the neutral rate may itself be shifting.

Their model describes one way policy beliefs can persist. Higher expected returns can reduce the saving needed for retirement. That can partly offset the usual dampening effect of higher rates on spending. A central bank may then mistake resilient demand for a higher neutral rate and maintain a higher rate path. This is a possible mechanism, not an estimate.

A further hypothesis is that competition for AI leadership makes investment in compute and power less sensitive to interest rates. Higher rates could then raise financing costs without substantially slowing that investment. Where those costs are passed through, price pressures could rise. Tightening might instead fall more heavily on housing and other rate-sensitive sectors. That could slow disinflation without showing that higher rates raise inflation overall.

GDP growth forecasts and scenarios

Average annual growth, 2025–29 (%)

Published forecastsOur reconstruction

Dots show FRI medians, OMB and our Anthropic reconstructions. Line spans eight other forecasts. Real/potential GDP and source periods differ.

Source: Karger Kuusela and others - Forecasting the Economic Effects of AI, Korinek Jones Sacher Cotter and McCrory - Economic Scenarios for Transformative AI · 2026-09-30

Forecasters raised productivity expectations

SPF mean ten-year forecasts, first-quarter surveys, percent a year

ProductivityGDP

Philadelphia Fed Survey of Professional Forecasters. Productivity is nonfarm labour productivity; GDP is real GDP.

Source: SPF ten-year productivity growth forecast, Philadelphia Fed SPF ten-year real GDP growth forecasts · 2026-09-30

Compare revenue with financing costs

The chart shows extra tax receipts minus additional spending and interest in the three scenarios; positive values mean a smaller annual deficit than CBO’s reference, not cumulative savings.

Financing costs use the same consumption-based discount factors as the yields, and the share of positive extra receipts spent rises linearly from 0% to 25% by 2036, on top of CBO’s spending path.

In the substantial scenario, the annual effect is −$288bn after three years and −$616bn in 2036, at the 2026 Q2 GDP level. The early cost comes from timing: yields rise as soon as investors expect the gains, while extra tax receipts arrive only as output grows. The modest scenario costs −$43bn after three years and −$242bn in 2036. Only in the extreme scenario does growth outrun the higher interest bill, turning −$533bn in year three into +$2.5tn in 2036.

A lower debt ratio need not mean less debt

The debt figures cover federal debt held by the public, including CBO’s growing debt stock, refinancing and additional borrowing. CBO already counts an average 0.1-point annual AI contribution, which we subtract, with its consumption effect, for the comparison.

In 2036, debt reaches 117% of GDP in the modest scenario, 89% in the substantial scenario and 53% in the extreme scenario, against CBO’s 120%. The substantial scenario changes dollar debt by +$7.0tn relative to CBO. A larger GDP denominator can lower the ratio even when borrowing is higher.

Receipts are assumed to rise by 17.2% of additional nominal GDP. Tax policy and income distribution could change that response. Unlike annual flows shown at the reference GDP scale, debt stocks are future nominal dollars.

Annual budget gain or cost from AI

Annual effect at the 2026 Q2 GDP scale ($bn)

ModestSubstantialExtreme

Same three GDP and consumption paths as above. Annual dollar amounts are rescaled to 2026 Q2 GDP; positive values mean a smaller annual deficit.

Source: Korinek Jones Sacher Cotter and McCrory - Economic Scenarios for Transformative AI, Chow, Halperin and Mazlish - Transformative AI, existential risk, and real interest rates, BEA via FRED - US capital, depreciation, investment and consumption, CBO February 2026 annual budget data, CBO February 2026 fiscal-year economic projections, US Treasury, Monthly Statement of the Public Debt, Table 3: Detail of Marketable Treasury Securities Outstanding, 31 August 2026, US Treasury, Monthly Treasury Statement, Table 1: Summary of Receipts and Outlays, September 2025, Federal Reserve Board GSW nominal yield curve, Federal Reserve Board - TIPS yield curve (FEDS 2008-05), U.S. Treasury - TIPS indexation and principal protection · 2026-09-30

Debt ratios under the same three growth paths

Federal debt held by the public / fiscal-year nominal GDP (%)

ModestSubstantialExtremeCBO baseline

Same three growth paths. Debt includes additional financing and TIPS indexation, divided by each case’s fiscal-year nominal GDP.

Source: Korinek Jones Sacher Cotter and McCrory - Economic Scenarios for Transformative AI, Chow, Halperin and Mazlish - Transformative AI, existential risk, and real interest rates, BEA via FRED - US capital, depreciation, investment and consumption, CBO February 2026 annual budget data, CBO February 2026 fiscal-year economic projections, US Treasury, Monthly Statement of the Public Debt, Table 3: Detail of Marketable Treasury Securities Outstanding, 31 August 2026, US Treasury, Monthly Treasury Statement, Table 1: Summary of Receipts and Outlays, September 2025, Federal Reserve Board GSW nominal yield curve, Federal Reserve Board - TIPS yield curve (FEDS 2008-05), U.S. Treasury - TIPS indexation and principal protection · 2026-09-30

How Treasury funds the borrowing matters

Marketable debt has an average maturity of 70 months, against 61 months on average since 1980, and a duration of about 4.6 years. About 35% matures or resets within a year, and Treasury plans to keep longer auction sizes unchanged, with bills filling any gap. Higher yields therefore reach the interest bill within a few years. Our calculation rolls each security over at its own maturity: existing fixed coupons stay unchanged, while new borrowing and floating-rate resets use the scenario’s financing curve, and temporary surpluses are held as cash until debt matures.

AI companies, by contrast, borrow long at fixed rates. The Dallas Fed puts their 2026 bond issuance at 12.5% of the duration Treasury supplies. If AI-driven growth raises rates, the government feels it sooner than the companies borrowing to build AI.

We vary the mix of securities issued, additional spending and the tax collected on extra GDP. With 25% less extra revenue, the substantial scenario’s annual 2036 budget effect is −$1.6tn at the reference GDP scale. A separate sensitivity lets lower debt ratios reduce term premia, using historical elasticities rather than AI-specific estimates. None of these fiscal assumptions feed back into GDP or consumption, and the central scenarios assume no extra inflation.

Evidence to watch

Growth and jobs: lasting upward revisions to productivity, consumption and GDP forecasts, and whether AI shows up as slower hiring in exposed jobs or as higher wages.
Energy: whether yields fall back if the Iran conflict ends and oil prices ease; if they stay high, growth or fiscal pressure becomes the likelier cause.
Policy versus the neutral rate: whether the ten-year keeps tracking news about the expected Fed path, which would point to near-term policy rather than a higher long-run neutral rate.
Where AI revenue lands: whether capital spending turns into revenue for AI firms’ customers, and whether that income goes to wages or to lightly taxed profits.
Politics: whether a shift of income toward capital brings tax and transfer changes that decide how much of the gain reaches household spending.

The central cases assume the full consumption path is priced; the model neither validates the GDP scenarios nor forecasts net yields. A future note will examine the equity trade-off between higher discount rates and faster earnings growth.

Sources

Korinek Jones Sacher Cotter and McCrory - Economic Scenarios for Transformative AI — Model-defined AI growth paths and factor-income outcomes. Anthropic Institute working paper 2026-02
BEA via FRED - US capital, depreciation, investment and consumption — Historical investment/consumption calibration for the Diffusion Capital Ramsey scenario model. Official national accounts and fixed-assets data
Beaudry, Cavallino and Willems - Anatomy of a rise: Monetary policy and the post-Covid surge in long-term interest rates — Payroll and Fed-speech windows in the post-2020 nominal yield rise. Authors' VoxEU column
Beaudry, Cavallino and Willems - Monetary policy along the yield curve: why can central banks affect long-term real rates? — Life-cycle saving and central-bank learning as a mechanism for persistent real-rate deviations. Bank of England Staff Working Paper 1117; August 2026 revision
Beechey and Wright - The High-Frequency Impact of News on Long-Term Yields and Forward Rates: Is It Real? — Different yield-component responses to news and the limits of broad time windows. Official Federal Reserve working paper FEDS 2008-39
Bhatt, Diercks, Eyal and Skaperdas - Federal Debt and Interest Rates (FEDS 2026-031) — Natural-experiment estimate splitting the debt effect between the term premium and r*. public working paper
Bank for International Settlements, Quarterly Review, September 2026 — Official commentary naming AI and fiscal factors as joint drivers of the US term premium. official publication
BLS - Productivity and Costs Second Quarter 2026 Revised — Realised nonfarm business labour productivity alongside long-run expectations. official statistical release
Bank of America Institute - Physical AI part 2 Humanoid robots — Rapid humanoid shipment outlook motivating physical-deployment sensitivity. primary research PDF
Borio et al. - Why so low for so long? (BIS WP 685) — Historical real-rate and realised-growth comparison. Official working paper PDF
CBO - How Budgetary and Economic Outcomes Might Differ From CBO's February 2026 Projections — AI contribution already included in the fiscal baseline. official CBO analysis
CBO - The Budget and Economic Outlook: 2026 to 2036 — the only official US government estimate of AI's productivity contribution. public report
Barlevy, Fisher, Pennington and Villa - The AI Revolution and Monetary Policy — Model calibration of the rate response to anticipated productivity growth. central bank publication (Chicago Fed Letter 528, September 2026)
Christensen and Rudebusch - A New Normal for Interest Rates? Evidence from Inflation-Indexed Debt — Market-based r* from TIPS and its response to growth-forecast revisions. published journal article (Review of Economics and Statistics 101(5), 2019); author PDF verified
Christensen and Rudebusch - Can Fiscal, AI, or Monetary News Explain the Rise in r*? — AI, fiscal and FOMC event-window changes in four market-based r* estimates. working paper (Hutchins Center WP 112, August 2026); full PDF verified
Federal Reserve Bank of Dallas, How AI debt financing impacts duration supply and interest rates — AI-related bond and swap duration supply compared with Treasury issuance. official Federal Reserve Bank research article
Albertazzi et al. - The cost channel of monetary policy: evidence from euro area firm-level survey data — Financing-cost channel supporting a qualified monetary-transmission hypothesis. ECB working paper
Eggertsson, Mehrotra and Robbins - A Model of Secular Stagnation — Structural OLG elasticity of r* to productivity growth. published journal article (AEJ: Macroeconomics 11(1), 2019)
Federal Reserve Board, FEDS Note: Substitutability between balance sheet reductions and policy rate hikes (Crawley, Gagnon, Hebden, Trevino) — the Board's working elasticity of the term premium to duration supply. official research note
Federal Reserve Board, Tips from TIPS: Update and Discussions — methodology for separating nominal Treasury yields into real-rate and inflation components. official Federal Reserve research note
Federal Reserve - FRB/US illustrative baseline data — Frozen joint-macro reference path. official model baseline and documentation
Federal Reserve - FRB/US model in Python — Published joint model of output, spending, prices and monetary policy. official model code and documentation
Federal Reserve Board GSW nominal yield curve — Continuously compounded nominal zero yield paired with TIPS zero at the same maturity. Staff research dataset
Federal Reserve Board - TIPS yield curve (FEDS 2008-05) — US market real interest rates from inflation-linked bonds, 1999 onwards. public dataset
Federal Reserve Bank of San Francisco, Treasury yield premiums (Christensen-Rudebusch), data to 2 September 2026 — decomposition of the 10-year Treasury yield into expected short rates and term premium. official model output, chart data file
EIA via FRED - Daily Brent crude spot price — Observed 2026 energy-price context for the AI bond note. Official daily series export
Forecasting Research Institute - How Accurate Have AI Progress Forecasts Been So Far? — Benchmark forecast errors and limits of extrapolating them to adoption and GDP. primary research review
Karger Kuusela and others - Forecasting the Economic Effects of AI — Conditional and unconditional economic forecasts and capability probabilities. FRI working paper revised 20 May 2026 PDF v1.1
Goldman Sachs - Robotaxis Are Forecast to Become a 400 Billion Dollar Market in 2035 — Physical AI outside humanoid robotics. primary research summary
Grace and others - Advanced AI according to 1580 researchers (ESPAI 2024) — Expert capability timelines for cognitive and physical tasks. AI Impacts survey paper PDF
Chow, Halperin and Mazlish - Transformative AI, existential risk, and real interest rates — Consumption-smoothing theory and survey-based evidence on long-run growth expectations and real rates. academic working paper updated June 2026
Holston, Laubach and Williams - Measuring the Natural Rate of Interest after COVID-19 — Estimated link between trend growth and r* in the NY Fed model. central bank staff report (FRBNY SR 1063, June 2023)
IFR - Humanoid Robots Vision and Reality — Uncertainty about broad humanoid deployment. industry association public summary
IMF - A Guide and Tool for Projecting Public Debt and Fiscal Adjustment Paths — Debt-ratio accounting and effective interest costs. Official IMF technical note
Randall Kroszner - What Drives Productivity Growth? Implications for the Economy and Prospects for the Future — Productivity growth real rates inflation and recognition as distinct channels. Official Federal Reserve speech
Kung Lustig and Paron - US Treasury Investors Are Long in AI — Fiscal backing channel and conditional Treasury revaluation. Author-hosted working paper
Morgan Stanley - Humanoids A Five Trillion Dollar Market — Slower humanoid adoption outlook motivating a physical-deployment sensitivity. primary research summary
SPF ten-year productivity growth forecast — Productivity recognition study evidence. Primary source; scope of inspection in notes
Philadelphia Fed SPF ten-year real GDP growth forecasts — Public survey expectations for US GDP growth over the current and next nine calendar years. Official survey workbook and documentation
PIMCO, The Credit Market Lens: AI Capex and the Limits of Crowding Out (Lotfi Karoui) — Event study of surprise AI debt deals on Treasury yields and term premia. Public article
SPF mean real GDP level forecasts — Forecast-revision response study. Primary source; inspection scope in notes
St. Louis Fed, On the Economy: Tracking AI's Contribution to GDP Growth (Rubinton and Patro) — gross demand-side contribution of AI-related investment to US real GDP growth, 2025. public webpage
US Treasury, Monthly Statement of the Public Debt, Table 3: Detail of Marketable Treasury Securities Outstanding, 31 August 2026 — Every outstanding marketable Treasury security with issue date, maturity and amount. Fiscal Data API (mspd_table_3_market)
US Treasury, Monthly Treasury Statement, Table 1: Summary of Receipts and Outlays, September 2025 — Federal receipts for fiscal year 2025. Fiscal Data API (mts_table_1)
US Treasury, TBAC charge: Bill Purchases and the Consolidated Balance Sheet (February 2026) — weighted average duration of total outstanding marketable debt and the Fed's portfolio. official Treasury presentation
US Treasury, presentation to the Treasury Borrowing Advisory Committee, August 2026 (fiscal 2026 Q3) — net coupon issuance by quarter and the constant-size projection. official Treasury presentation
US Treasury International Capital - Major Foreign Holders, official and total — the country ranking of foreign Treasury holdings and the split between official and private holders. official monthly table
U.S. Treasury - TIPS indexation and principal protection — Inflation-indexed principal and coupon mechanics. official Treasury publication