US productivity Q2 2026: durable or cyclical rise?

The figure released on August 6 by the Bureau of Labor Statistics (BLS) beat expectations without, by itself, establishing a new productivity regime. US nonfarm business productivity rose at a 1.4% annualized rate in the second quarter of 2026, compared with roughly 0.6% expected in the Reuters survey. Real output increased 1.7%, while hours worked rose only 0.3%. Q1 was also revised up from 0.3% to 0.8%.

The more important signal lies beyond one quarter. From Q2 2025 to Q2 2026, productivity increased 2.2%, with output up 2.5% and hours worked only 0.2% higher. Since Q4 2019, productivity has grown at a 2.1% annualized rate, compared with 1.5% during the 2007-2019 business cycle. That now matches the BLS long-run average since 1947.

Q4 2025 shows why the trend matters more than the peak

The hindsight available since our previous article is instructive. In January 2026, the Atlanta Fed’s GDPNow model reached a 5.4% annualized growth estimate for Q4 2025 before falling to 4.2% a few days later. The Bureau of Economic Analysis (BEA) eventually estimated real GDP growth at just 0.5% for the quarter. Q4 productivity was also initially reported at 2.8% before being revised down to 1.8% in March.

The Atlanta Fed explicitly describes GDPNow as a mechanical nowcast rather than an official forecast. The gaps mainly illustrate how sensitive quarterly readings are to revisions in output, hours, inventories and trade.

More output with almost the same number of hours

Labor productivity measures real output per hour worked and reflects technology, investment, organization, capital use and workforce characteristics at the same time.

The movement remains uneven. Manufacturing productivity increased at a 1.9% annualized rate in Q2, with output up 4.6% and hours up 2.6%. In durable manufacturing, productivity rose 2.7% while output jumped 7.3%. Yet year over year, total manufacturing productivity was up only 0.9%, while nondurable manufacturing productivity fell 0.3%.

Since late 2019, manufacturing productivity has increased only 0.5% per year, far below the 2.1% pace for the broader nonfarm business sector. The US improvement therefore does not yet amount to a uniform industrial productivity renaissance.

AI is becoming a credible hypothesis, not a proven cause

Digital investment is consistent with a technology effect. The BEA says Q2 investment was supported by information-processing equipment and by intellectual-property products such as software and research and development.

The IMF adds a broader structural perspective. In its 2026 assessment of the US economy, it estimates that output per hour grew by roughly 2.7% annually over the previous three years, faster than before the pandemic and stronger than in several other advanced economies. It also notes that the improvement has been particularly visible across several service industries, while manufacturing productivity gains have remained far more limited.

Research presented by economists associated with the St. Louis Fed estimates that 43% of US workers used AI on the job in 2026, compared with an average of 32% in the European countries studied. In their sample, a 10-percentage-point increase in adoption was associated with 2.9 percentage points of additional cumulative productivity growth relative to pre-pandemic trends.

The authors explicitly state that this relationship does not establish causality: industries adopting AI most rapidly may also invest more, employ more highly skilled workers or reorganize production faster. AI adoption estimates themselves are sensitive to definitions, ranging in the St. Louis Fed analyses from roughly 7% to 34% of firms depending on survey wording.

The OECD provides a forward-looking order of magnitude. Across three scenarios for G7 economies, it estimates that AI could add 0.4 to 1.3 percentage points to annual labor-productivity growth in the most exposed countries, including the United States. The estimate applies over the next decade rather than measuring AI’s actual contribution to Q2 2026.

Oil disrupted the quarter in both directions

Q2 was also shaped by a major energy shock. According to the Energy Information Administration (EIA), Brent crude reached $118 a barrel on April 29 before falling to $72 on June 26. During April and May, the average daily price swing was about $4 a barrel, compared with $1 during the same months of 2025. The EIA links this volatility to disruptions in oil flows around the Strait of Hormuz.

Higher oil and fuel prices raised transport, production and household costs. But disruptions in the Middle East also redirected part of global demand toward US refiners. Distillate exports averaged 1.56 million barrels per day in Q2, 30% above their five-year average, while jet-fuel exports reached 356,000 barrels per day, more than double their five-year average.

The BEA confirms that petroleum and related products led the increase in goods exports during Q2. But a higher nominal oil price does not directly inflate the BLS productivity measure, which compares price-adjusted real output with hours worked. No official publication currently quantifies a specific oil contribution to the 1.4% Q2 productivity gain. Any precise attribution would therefore be speculative.

Labor costs are contained, but disinflation is not automatic

Higher productivity helped restrain labor cost per unit of output. Unit labor costs rose at a 1.3% annualized rate in Q2 and 1.4% year over year. Nominal hourly compensation increased 2.7% in the quarter and 3.7% from a year earlier.

The same BLS table, however, shows unit nonlabor payments rising at a 14.0% annualized rate, while the value-added output price deflator increased 7.0%. These series can be volatile, but they show that subdued unit labor costs do not guarantee broad disinflation when other components of output prices are rising quickly. These pressures coincided with the quarter’s energy shock, but the BLS data do not measure a direct link.

Real hourly compensation fell at a 3.1% annualized rate in Q2 and 0.1% year over year. The BEA simultaneously measured a 5.7% annualized rise in the gross domestic purchases price index and a 5.1% increase in headline PCE prices, compared with 3.4% excluding food and energy, as the Federal Reserve continued to identify supply pressures including energy.

52.9%: labor’s share hits its lowest level since 1947

The BLS estimates that labor’s share of nonfarm business output fell to 52.9% in Q2, the lowest level since the series began in 1947. This does not mean nominal wages are falling. It means total labor compensation represents a smaller share of the value produced.

Reuters devoted a separate report to this record, placing more emphasis on distribution than in its general productivity story. The available data still do not allow the decline to be attributed specifically to AI. Profit margins, relative prices, sector composition, automation, bargaining conditions and capital income can all affect labor’s share.

A softer labor market makes the productivity gain harder to interpret

Data released on August 7 showed nonfarm payrolls falling by 23,000 jobs in July, while May and June payroll gains were revised down by a combined 103,000. The unemployment rate edged down to 4.1% as labor-force participation also declined. July lies outside Q2 and cannot directly explain the quarter’s productivity result, but the revisions to May and June show that labor demand during the quarter was weaker than initially estimated.

Weak hiring can itself temporarily lift output per hour; current data cannot yet distinguish that cyclical effect from a genuine efficiency gain.

GDP was modest, but private domestic demand was stronger

US real GDP grew at a 1.5% annualized rate in Q2, down from 2.1% in Q1. However, real final sales to private domestic purchasers — consumer spending plus private fixed investment — rose 3.9%, up from 1.7% in Q1. The GDP slowdown reflected, among other factors, lower government spending, slower investment and export growth, and a larger increase in imports.

Unlike in Q4 2025, GDPNow converged to 1.5% on July 28, exactly matching the BEA figure released two days later. The contrast mainly shows how the nowcast’s reliability depends on how much monthly data are already available.

Different international readings of the same signal

Reuters, a global agency headquartered in the United Kingdom, emphasized the stronger-than-expected productivity figure, low unit labor-cost growth and the possibility that AI investment could extend productivity gains.

In Saudi Arabia, Arab News published more cautious opinion pieces on the AI-productivity link. Carl Benedikt Frey argues that task-level efficiency gains do not automatically become economy-wide productivity gains when verification, organizational change and bottlenecks absorb part of the time saved. Robin Rivaton focuses more on the difficulty of spreading AI gains across an entire firm’s workflow.

In Singapore, The Straits Times placed greater emphasis on AI’s potential employment effects in US finance and technology. The cited story is based on Bloomberg reporting and therefore should not be treated as independent Asian evidence on US economic data.

The IMF and OECD take a more structural approach: the former emphasizes the recent US productivity gap with several advanced economies, while the latter focuses on future AI diffusion. The differences therefore concern framing — inflation, jobs, distribution or technology — more than the underlying facts.

Biases and conflicts of interest identified

The main quantitative evidence comes from public statistical agencies — the BLS, BEA and EIA — whose methods are published and estimates regularly revised. No specific financial conflict of interest was identified in the releases used here.

The Arab News articles cited are opinion columns rather than statistical newsroom investigations. Robin Rivaton leads a technology company and advises the French employers’ federation MEDEF on AI; his analysis is therefore used only to compare interpretive frameworks, not as quantitative evidence.

The OECD figures are scenario-based projections dependent on assumptions about adoption and micro-level productivity gains; they should not be treated as certain forecasts.

For the Fed, productivity does not automatically cancel inflation

Sustained productivity growth can allow wages to rise without increasing unit labor costs by the same amount and can raise potential growth. But the mechanism is not automatically disinflationary when price pressures come from energy or other nonlabor costs.

On August 6, St. Louis Fed President Alberto Musalem said the available evidence did not justify setting monetary policy on the assumption that AI will permanently accelerate productivity. The Fed is therefore observing strong productivity and investment while inflation remains above its 2% target.

A structural improvement is becoming plausible, but remains unconfirmed

The case for a durable improvement in US productivity is strengthening. The 2.1% pace since late 2019 exceeds the previous business cycle, while the IMF also finds a recent acceleration concentrated particularly in services.

It remains impossible to divide the gain precisely among AI, older forms of automation, conventional investment, reorganization, capacity utilization, sector composition and cyclical effects. Oil clearly affected exports and inflation during the quarter without a measurable contribution to productivity. The record-low labor share is also an important distributional signal without identifying a single cause.

The BLS is scheduled to publish its revised Q2 estimate on September 3, 2026. Q4 2025 showed that a one-percentage-point revision is possible. The real test of a regime change is therefore not whether the 1.4% quarterly figure survives unchanged, but whether gains persist, spread across more sectors and eventually coexist with stronger real compensation and controlled unit costs.

FAQ

Did US productivity really accelerate in Q2 2026?

It rose at a 1.4% annualized rate, above expectations. The stronger signal is the 2.2% year-over-year increase and the 2.1% annualized pace since late 2019.

Is AI already responsible for US productivity gains?

That has not been demonstrated. Research shows correlations and the OECD estimates significant future potential, but no official measure isolates AI’s contribution to Q2 productivity.

Did expensive oil artificially boost productivity?

Not directly. The oil shock supported some US exports and raised prices, but productivity is based on real output per hour. No official estimate quantifies a specific oil contribution to the 1.4% Q2 gain.