Macro Markets and Machines_The Economic and Market Transformation Driven by AI_GWM report - Flipbook - Page 28
Macro, Markets, and Machines
November 2025
Section 4.2: AI Productivity Estimates Are Few and Far Between
Our review of existing literature on the possible impacts of AI shows that there is no clear consensus
regarding the magnitude of the positive productivity benefits from this technology, with important
implications for its effect on growth, inflation, and policy rates (Exhibit 20). With the economic relevance of AI
still in its early stages, researchers are forced to make and impose key assumptions on technological adoption
rates, speed of investment, and complementarity benefits, among others, that can drastically affect their
conclusions. Restrictive or simplified frameworks where labour markets are constrained to remain in equilibrium
(i.e., do not allow economy-wide automation-related job losses) may also be too optimistic, as they overlook
possible negatives for household consumption and thus for GDP growth, inflation, and monetary policy. At
present, there is only a small sample of research that has produced productivity estimates for AI (particularly
from unbiased academic or multinational organization sources).
Exhibit 20 – Wide Variance in Estimates for United States Productivity Gains from AI
Notes: LHS chart is an average across all studies in the RHS chart.BBK = Baily, Brynjolfsson, and Korinek (2023). OECD 2025 = Filippucci, Gal, Laengle,
and Schief (2025). OECD 2024 = Filippucci, Gal, and Schief (2024). AB = Aghion and Bunel (2023). IMF = IMF (2025). Ace. = Acemoglu (2024).
Sources: OECD; Scotia Wealth Management.
In his representative economy example cited in the previous section, Acemoglu estimates gains from AI in
the order of only 0.7% to 1.0% cumulatively over a 10-year period, equivalent to an only ~0.1% boost to
annual total factor productivity (TFP) growth. In the United States, TFP growth averaged 0.8% per annum in
the 2010-2019 period prior to the pandemic, so Acemoglu’s estimate might be relatively small and not the
massive shift or revival of productivity growth that other experts (or even markets) would be expecting. At the
opposite end of AI estimates, an oft-cited 2023 private-sector research report estimates that AI would lift
productivity growth by an average of 1.5 p.p. per year over a 10-year period. However, the authors assume that
AI gains and its adoption materialize fully over this decade, frontloading their economic impact. Brookings
Institution researchers (BBK in Exhibit 20) consider how AI-related innovation and productivity benefits result in
additional innovation and productivity growth in other sectors that further (indirectly) boost the productivity
gains from AI, for a total of up to 1.7 p.p. per annum over baseline.
GDP estimates on the possible impact of AI are less common than the already-narrow set of available
research on potential TFP gains, with even fewer authors exploring or detailing the effects of AI adoption on
inflation and monetary policy. For the United States, Acemoglu estimates an upper-bound GDP boost of
1.6 p.p. relative to baseline over a 10-year period, compared with the IMF’s estimates of around 2 p.p. (low) to
5.5 p.p. (high) relative gains over 10 years. Following Acemoglu’s framework for translating TFP gains into GDP
gains would peg the Organisation for Economic Co-operation and Development’s (OECD’s) TFP-to-GDP boosts
at ~5-15 p.p. over baseline GDP in 10 years (OECD, 2025). As we discuss later in this report, assumptions and
starting points for the magnitude of AI’s TFP benefits can have large implications for the economy’s inflationary
response. In the IMF’s high-TFP scenario (where GDP is 5.5 p.p. higher 10 years ahead), the impact on U.S.
inflation peaks at less than 0.1 p.p. higher relative to baseline alongside modest central bank tightening that
reins in price growth. In comparison, BIS’s research – among the very few that produce inflation estimates – sees
inflation around 1 p.p. higher than baseline over the first decade (average of the anticipated and unanticipated
Scotia Wealth Management
27
