Will US nonfarm labor productivity grow by more than 10% in any year through 2035?

TL;DR

I estimate a 2.4% chance that U.S. nonfarm business labor productivity exceeds 10.0% in any calendar year from 2026 through 2035. The official BLS annual series has 0 qualifying years in 78 observations from 1948 through 2025, with a maximum of 6.7% in 1950 and a latest completed-year value of 2.1% in 2025 (BLS PRS85006092, BLS Table 2). AI creates a real tail, but official forecasts and quantified AI studies point to low-single-digit productivity growth, not a 10% annual BLS print (Philadelphia Fed SPF Q1 2026, CBO 2026 outlook, Federal Reserve AI buildout note).

Context

The resolution metric is BLS nonfarm business sector labor productivity, meaning real output per hour worked by all workers, including employees, proprietors, and unpaid family workers (BLS Productivity and Costs, June 4 2026). The annual value is the calendar-year average percent change, not a quarterly annualized rate; BLS states that its quarterly percent changes in the news release are seasonally adjusted annualized rates (BLS Productivity and Costs, June 4 2026).

As of the client timestamp on July 21, 2026, the latest BLS release was the revised first-quarter 2026 Productivity and Costs release, published June 4, 2026. It reported nonfarm business productivity up 0.3% at a seasonally adjusted annual rate in 2026 Q1, up 2.8% from 2025 Q1, and up 2.1% for the completed 2025 calendar year (BLS Table 2).

Evidence

The historical backbone is decisive. The full BLS annual history I use below is PRS85006092, nonfarm business sector labor productivity, annual Q05 percent change from the previous annual average, coverage 1948-2025, N=78, current BLS vintage available with the June 4, 2026 release, rounded to one decimal place (BLS PRS85006092, BLS major-sector workbook).

My summary statistics from that BLS annual series are: mean 2.16%, median 2.15%, sample standard deviation 1.52 percentage points, minimum -1.7% in 1974, maximum 6.7% in 1950, 0 years above 10.0%, 2 years at or above 5.0%, and 8 years at or above 4.0% (BLS PRS85006092). A normal fit to those 78 annual observations gives a 10-year probability of about 0.00013%; a Student-t fit gives about 0.021%; and a generous generalized-Pareto tail fit using only the 7 observations above 4.0% gives about 0.16%, so the ordinary historical-statistical channel is well below 1% (BLS PRS85006092).

The current 2026 arithmetic is also unfavorable. BLS Table 2 reports a 2025 annual labor-productivity index of 117.8 and a 2026 Q1 index of 119.4, both index 2017=100, seasonally adjusted; for 2026 annual growth to exceed 10.0%, the 2026 annual average would need to exceed about 129.6, so the average of Q2-Q4 2026 would need to exceed about 133.0, roughly 11% above Q1 2026 (BLS Table 2). If that happened along a smooth path, Q2, Q3, and Q4 would each need to rise at roughly a 24% annualized pace after Q1's 0.3% annualized reading, which is a much harder condition than one isolated strong quarter (BLS Table 2).

The conventional forward outlook does not come close to the threshold. The Philadelphia Fed Survey of Professional Forecasters Q1 2026 median 10-year annual-average productivity growth forecast was 1.80% (Philadelphia Fed SPF Q1 2026). CBO's 2026-2036 outlook projects real GDP growth averaging 1.8% per year from 2027 through 2036, while saying generative AI should lift TFP growth only modestly over the next decade (CBO 2026 outlook). The Federal Reserve's July 2026 Monetary Policy Report says business-sector labor productivity has averaged 2.1% per year since late 2019, faster than the 1.5% pace from 2007 Q4 to 2019 Q4 but still far below 10% (Federal Reserve Monetary Policy Report, July 2026).

AI is the main upside. Goldman Sachs estimates widespread generative AI adoption could raise U.S. productivity growth by about 1.5 percentage points per year over a 10-year adoption period (Goldman Sachs). McKinsey estimates generative AI alone could add 0.1 to 0.6 percentage point to annual labor-productivity growth through 2040, with broader work automation adding more under strong redeployment assumptions (McKinsey). Anthropic's own usage-based estimate is more bullish at about 1.8 percentage points of additional annual U.S. labor-productivity growth over the next decade, while Acemoglu's NBER model is far smaller at no more than a 0.66% total TFP gain over 10 years from current generative-AI capabilities (Anthropic, NBER Acemoglu). Even the more bullish mainstream numbers usually imply something like 3% to 5% annual labor-productivity growth, not a single published BLS annual rate above 10%.

I read the July 17, 2026 Federal Reserve AI buildout note as a useful check on that tail. It says AI capabilities, investment, and adoption are moving, but it also says aggregate output and labor-market data still show limited broad-based transformation, and that high-exposure sectors have not yet shown a clean aggregate productivity break from lower-exposure sectors (Federal Reserve AI buildout note). My final estimate assigns about 0.2 percentage point to ordinary historical variation, 0.3 point to extreme non-AI shocks or revision edge cases, 0.5 point to a strong but non-transformative AI boom creating a lumpy annual print, 1.4 points to a transformative-AI or abrupt labor-displacement scenario before 2035, and 0.1 point to measurement or successor-series oddities; allowing for overlap gives 2.4%.

What's non-obvious

The trap is quarterly annualized growth. BLS quarterly productivity can clear 10% at an annualized rate; 2020 Q2 was 20.9%, but the 2020 annual BLS value was only 5.2% because the resolution metric averages the calendar year (BLS PRS85006092). A YES needs a high level of output per hour sustained across the annual average, not one spectacular quarter.

The other trap is scaling task-level AI gains to the whole nonfarm business sector. A 10% speedup in a coding, writing, or customer-support task does not become 10% aggregate productivity if the task is a small share of total cost, if complementary work remains the bottleneck, if demand is weak, or if BLS/BEA price and quality measures do not capture the gain quickly (Federal Reserve AI buildout note). That is why AI moves the forecast above the pure historical tail, but not near 10%.

Limitations

The biggest limitation is transformative-AI uncertainty. The post-1948 BLS history is a strong base rate for normal recessions, recoveries, policy shifts, and even COVID, but it is not a good sample for a world where AI systems rapidly substitute for broad categories of cognitive labor before 2035 (BLS PRS85006092, Federal Reserve AI buildout note).

The second limitation is measurement. AI could create consumer surplus, free digital output, quality changes, or internal firm time savings that do not appear as measured real nonfarm business output per hour within a one-year BLS revision window (Federal Reserve AI buildout note). The opposite edge case also exists: a severe labor-hours collapse, successor-series change, or large benchmark revision could create a high measured ratio without a true welfare boom, but normal BLS revisions are much smaller than the gap between recent 2%-3% readings and the 10.0% threshold (BLS Productivity and Costs, June 4 2026).