What Might AI Adoption Mean for the Fiscal and Economic Outlook?

The most-consequential variable, for understanding fiscal and macroeconomic impacts of AI, is the path of productivity growth. Over the long run, productivity growth is the dominant determinant of living standards and a key contributor to fiscal outcomes. In the context of AI-related discussions, it is important to have a realistic range of forecasts grounded in sound macroeconomic thinking as well as knowledge of LLM capabilities.

In the Karger et al. (2026) survey’s “Moderate Adoption” scenario, the median economist expected annual labor productivity growth of 2.5% from 2025-30.2 This is a large number by recent standards: average labor productivity growth from 2015 to 2025 was only 1.8%. Figure 2 shows that this 2.5% assumption is historically large but not unprecedented: the ten-year average of labor productivity growth has exceeded it twice in the postwar record, first in the 1960s and again in the late 1990s and early 2000s. However, this scenario entails lower growth than in the most aggressive forecasts by some AI experts, and lower growth than in the Rapid Adoption scenario in Karger et al. (2026), shown as the top horizontal line in Figure 2. In the Appendix, we show results for that scenario as well as the Slow Adoption scenario.

Next to productivity, the other macroeconomic effect that likely matters most is labor force participation. We therefore consider what would happen if economists’ expectations for labor force participation were to be realized in the moderate adoption scenario. Economists, like the AI experts surveyed, tended to anticipate lower participation in that scenario: 60.7% in 2030 as opposed to CBO’s assumed rate of 62.1% in the same year.3

The third and final factor we consider is the possible implication of employment loss for federal spending. This is not something asked of survey respondents in Karger et al. (2026), but it is a fiscal variable worth considering given the large participation rate reductions that survey respondents find plausible. There is considerable uncertainty as to how robust the fiscal response will be to any job loss, so we consider two assumptions that could bracket the range of possibilities. The less-generous assumption is that the federal government spends (with no offsetting receipts) as much on workers exiting the labor force as it does on the unemployed. This amounts to about $5,500 per lost labor force participant. (Here we scale by the number of unemployed rather than the number of unemployment insurance recipients.)4 The more-generous assumption is that it spends as much as it does on retirees, amounting to about $42,400 per lost participant.5 Using the participation rate reduction expected by survey respondents, as described above, we then separately apply these two assumptions.

Putting all of these factors together, we now show a variety of macroeconomic and fiscal outcomes over the budget window. The dashed black line indicates the Budget Lab’s baseline path for each outcome. Solid lines show a “stacked” outcome that progressively incorporates deviations from that baseline: first, incorporating the higher labor productivity increase (blue line); second, incorporating the labor force participation reduction, in addition to the productivity change (orange line); third, incorporating the previous changes as well as a less-generous outlay increase to support those no longer in the labor force (green line); and fourth, the same as the previous but with a more-generous outlay increase (pink line).

A few takeaways are immediately apparent. First, the budget balance and debt trajectory are generally more sustainable in this scenario than in the baseline. Even with diminished labor force participation, and with the higher interest rates that occur in equilibrium with faster productivity growth, a productivity boom of this magnitude leads to more rapid GDP growth and a more-manageable deficit and debt burden. Only the large outlay increase assumed in the “generous” scenario variant offsets the fiscal benefits from higher productivity. But this is neither a forecast nor a comprehensive assessment of how a large AI shock would transform the economy; rather, it is an illustration of how specific assumptions affect key outcomes in The Budget Lab Small Macro Model (BLSMM), which we describe in the next section.

In the future, the Budget Lab intends to follow this work with a more detailed investigation of how AI could affect fiscal outcomes. In particular, an AI-induced productivity shock (as opposed to other kinds of productivity shocks) could interact with the existing tax system in a way that reduces revenues (e.g., because capital is more lightly taxed than labor).

The modeled trajectories in Figures 3-5 are all outcomes of the brand-new Budget Lab Small Macro Model (BLSMM). It is freely available here as an interactive tool. Documentation on how it works is available here. This is a preliminary version, which Budget Lab expects to modify in response to feedback. As such, figures like those above should be regarded as illustrative demonstrations of stylized scenarios, rather than authoritative forecasts.

Intuitively, what BLSMM does is take fiscal and monetary policy as inputs, along with other user-specified values like the paths of productivity growth and labor force participation, and delivers model-consistent paths for a variety of key outcomes. Those outcomes include real GDP; unemployment; inflation; 10-year Treasury rates; outlays, receipts, and debt; the primary and total budget deficits; the debt-to-GDP ratio; the average interest rate on federal debt; and the federal funds rate.

In so doing, it generates an internally consistent economic and fiscal narrative that embodies key macroeconomic relationships. For example, BLSMM includes an expectations-augmented Phillips Curve and an Okun’s Law relationship between unemployment and GDP. This framework means that inflation is anchored to the Fed’s target but evolves in the short run as shocks hit the economy and unemployment fluctuates. In turn, unemployment is closely connected to the output gap; above-potential GDP is associated with low unemployment, and vice versa. Monetary policy is implemented with a Taylor-type policy rule: the Fed tightens when inflation is high (or unemployment is low) and loosens when inflation is low (or unemployment is high). Some of the details of this rule can be modified by users.

Where possible, the BLSMM implementation follows CBO and standard practice, as in the case of the relationship between federal debt and the real neutral interest rate (often referred to as r*), where we assume that a one percentage point increase in the debt (as a share of potential GDP) causes r* to rise by 2 basis points. Some of BLSMM’s other relationships are implemented with so-called rules of thumb from CBO. Those rules of thumb imply that higher productivity and labor force growth lead to substantial reductions in outlays as a share of output; by contrast, receipts as a share of output are minimally affected.6 For all of these assumptions, alternative choices could also be reasonable.

BLSMM was designed to be parsimonious, especially relative to the large macro models commonly used by Budget Lab and others. This has important virtues, chiefly simplicity and transparency. It is much easier to understand what is driving a given equilibrium response in BLSMM than in a larger model that functions to some extent as a black box. However, it also comes with downsides: some important macroeconomic dynamics are not captured by the current version of the model. For example, the financial sector in BLSMM is very limited, and does not feature asset values that could be used to generate wealth effects of macro or fiscal shocks. The model has no international dimension, and so does not include trade or capital flows. In some cases, these limitations may mean that BLSMM generates outcomes that conflict with users’ macroeconomic intuition. The Budget Lab expects to reevaluate these limitations in the future, balancing them against the overall complexity of the model.

The Budget Lab is grateful to Basil Halperin for insightful feedback on an earlier draft.

On 2026-05-19 The Budget Lab switched from using the as-published Karger et al. (2026) productivity forecasts to using modified versions of those forecasts that better align the Karger et al. (2026) nonfarm business output per hour concept to the BLSMM total economy output per employee concept. This refinement reduced the magnitude of productivity shocks assumed in the BLSMM forecasts.