AI Productivity Boom or Jobless Growth? What BLS Q2 2026 Data Really Says

Q2 2026 data show a widening gap between output per worker and employment in several important U.S. business sectors.

A major investment wave in computers, industrial equipment, robotics, software and artificial intelligence is coinciding with a visible change in the U.S. economy. Productivity and output per worker are rising, while employment is barely growing—or contracting—in several important sectors. There is not yet a clear causal link among technology, productivity, employment and earnings. That developing relationship is now one of the most important economic and market signals to follow.

ATN analysis of preliminary Bureau of Labor Statistics data, updated August 17, 2026.

The productivity story in five points

The productivity expansion is broadening. Nonfarm business productivity increased 2.2% from Q2 2025 to Q2 2026, while ATN’s rolling calculations show stronger year-to-date momentum.
Durable manufacturing is the clearest goods-sector signal. Its productivity increased 2.2% year over year and its output increased 3.1%, according to the headline BLS release.
More output is being produced with almost no employment growth in the selected sectors. ATN’s output-per-worker measure increased 3.108% year over year in durable manufacturing while employment was essentially unchanged at -0.025%. The figures move together, but do not yet prove that productivity caused employment to stall.
The pattern is consistent with capital deepening and technology adoption. BEA reports that Q2 business investment was led in part by industrial equipment, information-processing equipment, software and research and development.
It is not yet scientific proof that AI is eliminating jobs. The macro data do not identify which machine, software program or management decision caused the change. They provide a strong signal that must be tested against investment, firm-level adoption and labor-market evidence.

The investment wave behind the data

The technology cycle is larger than generative AI. It includes physical and digital capital working together:

Hardware: computers, servers, semiconductors, networking, industrial equipment, sensors, automated warehouses and robotics;
Software: enterprise systems, cloud platforms, workflow automation, data analytics and cybersecurity;
Artificial intelligence: predictive systems, computer vision, machine learning, generative AI and autonomous software agents;
Complementary capital: power, cooling, data centers, communications infrastructure, training and redesigned business processes.

The Bureau of Economic Analysis reported that U.S. business investment increased in Q2 2026. Equipment growth was led by industrial, transportation and information-processing equipment. Intellectual-property investment also increased, led by software and research and development.

This matters because a technology does not raise economy-wide productivity when it is announced or purchased. It must be installed, connected to data, integrated into a workflow and used by trained employees. The implementation lag can last for several quarters or years. The Q2 figures may therefore reflect the combined effect of earlier investments in conventional automation, cloud software, computer equipment and newer AI systems.

What the official Q2 2026 BLS release says

The preliminary BLS report contains two different comparisons that should not be mixed. Quarter-to-quarter figures are seasonally adjusted annualized rates. The year-over-year figures compare Q2 2026 directly with Q2 2025.

Sector	Q2 productivity, annualized	Q2 output, annualized	Q2 hours, annualized	Productivity, year over year	Real hourly compensation, year over year
Nonfarm business	+1.4%	+1.7%	+0.3%	+2.2%	-0.1%
Manufacturing	+1.9%	+4.6%	+2.6%	+0.9%	+0.6%
Durable manufacturing	+2.7%	+7.3%	+4.5%	+2.2%	+1.7%

Two additional figures deserve attention. Nonfarm business unit labor costs increased only 1.4% over the year, while the labor share of nonfarm business output fell to 52.9%—the lowest reading in a series beginning in 1947.

That combination is favorable to corporate margins in the short term: output per hour is rising faster than the labor cost attached to each unit of output. It also suggests that the first financial gains from higher productivity are not being distributed evenly to labor.

ATN sector calculations: output rises, employment stalls

The official BLS release emphasizes output per hour. ATN extended the analysis by comparing four measures across three time windows: labor productivity, inflation-adjusted hourly compensation, output per employed worker and employment.

Sector	Comparison	Labor productivity	Real hourly compensation	Output per worker	Employment
Manufacturing	Q2 year over year	+0.903%	+0.626%	+1.675%	-0.365%
	YTD through Q2	+1.053%	+1.447%	+1.689%	-0.573%
	Four-quarter year ending Q2	+1.673%	+1.529%	+2.101%	-1.106%
Durable manufacturing	Q2 year over year	+2.191%	+1.653%	+3.108%	-0.025%
	YTD through Q2	+2.383%	+2.634%	+3.028%	-0.573%
	Four-quarter year ending Q2	+2.981%	+2.601%	+3.507%	-0.642%
Nonfarm business	Q2 year over year	+2.242%	-0.092%	+2.358%	+0.121%
	YTD through Q2	+2.585%	+0.248%	+2.828%	+0.081%
	Four-quarter year ending Q2	+2.072%	+0.465%	+2.196%	+0.110%
Nonfinancial corporations	Latest quarter, Q1 year over year	+3.743%	+0.793%	+4.263%	+0.036%
Nonfinancial corporations	Four-quarter year ending Q1	+3.232%	+1.432%	+3.461%	+0.001%

Methodology: ATN calculations use the unrounded, seasonally adjusted BLS major-sector index series. “Output per worker” is an ATN-derived ratio of the real-output index to the employment index; it is not a separately published BLS headline measure. YTD compares the average of Q1–Q2 2026 with Q1–Q2 2025. The year-ending measure compares the latest four-quarter average with the preceding four-quarter average. Nonfinancial corporate data are published with a lag, so the latest available observation in the Q2 release is Q1 2026—not Q2.

The output-per-worker calculation is deliberately separate from labor productivity. Labor productivity measures output per hour, while output per worker also reflects changes in average hours worked per employee. The two measures answer different questions.

Where the productivity gains are appearing

Manufacturing: more output from a smaller workforce

The manufacturing pattern is difficult to ignore. On a year-ending basis, output per worker increased 2.101% while employment fell 1.106%. Productivity increased 1.673%, and real hourly compensation increased 1.529%.

This is the textbook outline of capital deepening: each worker has access to more—or better—equipment, software and process automation. It may also reflect the survival and expansion of more efficient firms, changes in product mix, overtime, capacity utilization and supply-chain normalization. Technology is a strong candidate, but it is not the only possible explanation.

Durable goods: the strongest technology-sensitive signal

Durable manufacturing includes many industries in which automation has a direct physical application: machinery, computers and electronic products, electrical equipment, transportation equipment, aerospace, fabricated metals and related supply chains.

Here the gap is widest. Year-ending output per worker increased 3.507%, productivity increased 2.981% and real compensation increased 2.601%, even as employment declined 0.642%. In the latest year-over-year comparison, output per worker increased 3.108% with employment almost perfectly flat.

This is the most persuasive sector-level evidence that firms can expand production without proportionately expanding headcount.

Nonfarm business: the effect extends into services

Nonfarm business is far broader than manufacturing. It captures much of the private goods-and-services economy. Its YTD productivity increased 2.585% and output per worker increased 2.828%, while employment grew only 0.081%.

That combination is important because software and AI are more likely to change services than industrial robots alone. Customer support, finance, marketing, software development, administration, logistics, research and professional services can all raise throughput without producing a physical unit on a factory line.

Nonfinancial corporations: high output with almost no employment growth

The latest nonfinancial corporate data are even more striking, although they lag by one quarter. Productivity increased 3.743% and output per worker increased 4.263% year over year, while employment increased only 0.036%. On a four-quarter basis, employment was effectively unchanged.

Large corporations tend to have the capital, data, management systems and technical staff required to deploy automation at scale. That makes this sector a plausible early transmission channel from technology spending to measured economic output.

Does the data prove that AI and automation caused the gains?

No single BLS productivity table can prove that. Labor productivity is an outcome measure. It tells us how much real output is produced per hour, but it does not identify whether the improvement came from a robot, a new AI model, better software, workforce composition, economies of scale, management changes or cyclical recovery.

The causal case becomes more credible when three separate lines of evidence point in the same direction:

Investment: BEA reports continued growth in information-processing and industrial equipment, software and R&D.
Measured outcomes: BLS reports stronger productivity, while ATN’s calculations show output per worker rising much faster than employment.
Adoption and use: Census Bureau data show that AI is being used in business operations and is saving time on real work tasks.

The BLS total-factor-productivity accounts offer additional historical support. In 2024, private nonfarm business labor-productivity growth of 3.0% included a 1.1-percentage-point contribution from capital intensity and a 1.5-point contribution from total factor productivity. Intellectual-property products accounted for more than half of the 2.7% growth in capital input. For 2025, capital intensity still contributed 0.9 point and total factor productivity contributed 0.8 point to labor-productivity growth.

Those annual growth-accounting estimates strengthen the technology thesis, but detailed capital data arrive with a lag. They do not yet provide a clean AI-only decomposition for Q2 2026.

The responsible conclusion: Q2 2026 provides credible, data-based evidence of technology-compatible productivity growth and job-light output expansion. It does not yet provide a laboratory-style causal estimate of how many jobs were displaced specifically by AI.

We now have scientific measurements of the pattern—not scientific proof of its cause. That distinction makes the analysis more credible, not less important.

The employment effect may appear first in hiring—not mass layoffs

Technology does not need to produce a wave of dismissals to reduce employment growth. A company can allow natural attrition, leave vacancies unfilled, reduce entry-level recruitment, consolidate teams or expand output without adding the workers it would previously have hired.

This distinction helps reconcile apparently conflicting evidence. A 2026 Census Bureau study found that most adopting firms used AI to augment tasks and that AI-related employment decreases were reported by only 2% of firms. Yet another Census working paper found a sizable and persistent reduction in hiring for 22-to-24-year-old workers in the industries and states most exposed to AI. Regression-adjusted early-career employment in the most exposed group declined 12% over the ten quarters following the introduction of ChatGPT, although the authors also identify pre-existing pandemic-era trends that complicate a simple causal interpretation.

The emerging sequence may therefore be:

AI and automation assist existing employees;
output per employee rises;
the next planned hire is postponed or cancelled;
entry-level and routine roles weaken first;
headcount declines later only where productivity gains exceed demand growth.

This is a more plausible early-stage mechanism than an immediate economy-wide replacement of workers.

Why real compensation is rising unevenly

Real hourly compensation increased most strongly in the ATN durable-manufacturing and nonfinancial-corporate calculations. One possible explanation is skill-biased technological change: firms pay more for the engineers, technicians, software specialists, data professionals and experienced operators who can install, supervise and improve the new capital.

That explanation remains a hypothesis. Aggregate sector data cannot identify which occupations received the increase. Higher compensation can also reflect labor scarcity, overtime, bargaining, bonuses, benefits, a changing employee mix or the departure of lower-paid workers from the measured workforce.

The central issue for the economic system and markets would emerge if future data confirm a persistent combination of higher productivity, stopped or tightly limited employment growth, and compensation gains concentrated in the technical jobs that build and operate the new capital. That would support earnings for technology-intensive companies while narrowing access to employment income across the wider economy.

The education signal is nevertheless visible in adoption data. Census reported that AI use at work was more frequent among employees with a bachelor’s degree or higher. This supports a future labor market oriented toward technical capability and adaptable medium-to-high skills—but it does not mean that every successful worker will need a traditional four-year degree. Industrial technicians, skilled trades, equipment maintenance, cybersecurity and applied software roles may depend as much on specific training and experience as formal academic credentials.

How technology changes goods and services differently

Economic area	Primary technology channel	Likely productivity effect	Likely labor effect
Durable goods	Robotics, computer-controlled machinery, vision systems, digital twins and supply-chain software	More units, higher precision and less downtime per labor hour	Fewer routine production hires; greater demand for technicians and engineers
Nondurable goods	Process controls, packaging automation, forecasting and quality analytics	Lower waste, faster throughput and more consistent output	Gradual role redesign; less dramatic robotics effect in some industries
Business services	Generative AI, workflow agents, analytics, cloud software and automated customer service	More cases, documents, code, analysis or customer interactions per employee	Hiring restraint in routine cognitive and entry-level work
Logistics and distribution	Warehouse robotics, route optimization, autonomous systems and inventory software	Faster movement with lower error and inventory requirements	Fewer repetitive roles; more maintenance, control and exception management

The emerging economic and financial scenario

Productivity is the central bridge between technology, earnings, inflation and living standards. When a business can produce more output from each hour, several outcomes become possible:

revenue and profit can grow without an equivalent increase in payroll;
unit labor costs can slow, reducing pressure on selling prices;
cash flow can finance another round of equipment, software and R&D;
successful firms can pay scarce high-skill employees more;
weaker firms and routine occupations face a higher competitive hurdle.

This creates a self-reinforcing investment cycle:

Technology investment → higher output per hour → stronger margins and cash flow → more R&D and capital investment → further productivity gains.

The virtuous cycle for output and earnings can simultaneously become a difficult cycle for employment. If demand grows more slowly than productive capacity, firms do not need to replace every departing employee. The economy can continue expanding while job creation becomes narrower, slower and more skill-selective.

Scenario 1: benign productivity diffusion

Demand expands fast enough to absorb higher output. Productivity supports wages, margins and lower inflation, while displaced tasks are replaced by new activities. Employment growth slows temporarily but broadens again as new products and services emerge.

Scenario 2: job-light growth and profit concentration

Output rises while headcount stays flat. Large, capital-rich companies capture the greatest benefits. Earnings grow faster than employment, labor share remains weak and entry-level hiring becomes more difficult. This is the direction most consistent with the current ATN sector calculations.

Scenario 3: capital-spending disappointment

Companies invest heavily but fail to generate sufficient revenue or utilization. Depreciation, power, software, financing and continuing R&D costs remain after the productivity burst fades. Margins weaken, capital spending is cut and layoffs follow. Technology can be economically transformative even when early owners of the capital earn poor returns.

What the productivity shift means for markets

Corporate earnings

Rising output per worker is initially supportive for profit margins, particularly when unit labor costs remain contained. Investors should distinguish genuine operating leverage from temporary cost cutting. Sustainable productivity requires revenue growth, customer value and repeatable process improvements—not merely a smaller workforce.

Technology and industrial capital

The direct beneficiaries may extend beyond AI model providers. Semiconductors, networking, industrial automation, robotics, sensors, cloud platforms, cybersecurity, power, cooling and specialist software all participate in the productivity infrastructure. The market risk is paying a high valuation before the customer’s return on investment is proven.

Inflation and interest rates

Higher productivity can allow wages and profits to rise with less inflationary pressure per unit of output. Nonfarm business unit labor costs increased only 1.4% over the latest year. Manufacturing unit labor costs, however, increased 3.5%. The productivity story is therefore disinflationary at the margin, not a guarantee of low inflation across every sector.

Employment and consumption

A prolonged hiring slowdown would eventually affect household income, consumption, credit and housing. The first risk is not necessarily a sharp increase in unemployment; it is a reduction in the number and quality of routes into stable careers. That makes early-career hiring, job openings, hours worked and labor-force transitions important leading indicators.

What investors should monitor next

productivity compared with output, hours and employment—not productivity alone;
revenue and free-cash-flow growth compared with technology capital expenditure;
unit labor costs and labor share;
job openings, new hires and entry-level employment in AI-exposed industries;
real compensation by sector and occupation;
utilization of data centers, robots, software licenses and automated capacity;
depreciation, impairments and replacement spending;
the breadth of productivity gains across goods and service industries.

The relationship ATN will follow

The present data do not establish a clear productivity-employment-earnings link. Future releases can either strengthen or break the emerging pattern. ATN will follow the relationship through four questions:

Does productivity remain elevated? A single strong period may reflect normal cyclical movement rather than a technology-led structural change.
Does real output continue to grow faster than employment? Persistence would be stronger evidence of job-light expansion.
Where do compensation gains occur? Occupational and industry detail will be needed to determine whether real pay is concentrating among technical, engineering and supervisory workers.
Does weaker hiring spread into demand? The systemic risk appears only if limited employment growth begins to restrain household income, consumption, housing or credit while corporate output and earnings continue to rise.

This is the issue to watch: not productivity by itself, and not one weak employment figure by itself, but whether the three variables begin to form a durable economic chain.

Why this remains an early-stage signal

The BLS labels the Q2 2026 estimates preliminary. The revised release is scheduled for September 3, 2026. BLS notes that quarterly productivity estimates can be revised materially as more complete information becomes available.

The comparison also spans sectors with different output concepts. BLS explicitly warns that manufacturing output measures are constructed differently from business and nonfarm business output, so levels should not be compared directly across those sectors.

Finally, correlation is not causation. Interest rates, trade, demand, inventories, workforce demographics, sector composition, post-pandemic normalization and management decisions can all affect output and employment. The value of this analysis is not that it closes the case. It identifies a coherent pattern early enough for investors, businesses and workers to monitor its development.

Frequently asked questions

Is AI increasing U.S. productivity in 2026?

The data are consistent with an AI- and technology-supported productivity improvement, but BLS cannot isolate AI as the cause. Investment, adoption, time-saving surveys and productivity outcomes together make the case stronger than any one series alone.

Does higher productivity mean fewer jobs?

Not automatically. If demand grows as fast as productive capacity, output and employment can rise together. If productivity grows faster than demand, firms may expand with fewer new hires. The current data show job-light growth in several sectors, not an economy-wide employment collapse.

Which sector has the strongest productivity signal?

Durable manufacturing stands out in the ATN analysis. Output per worker rose by more than 3% across each selected comparison window while employment was flat or declining.

Why can output per worker rise faster than output per hour?

Output per worker is affected by the number of workers and the average hours each employee works. Labor productivity uses total hours worked. Changes in overtime, part-time work and weekly schedules can therefore create a gap between the two measures.

Will future jobs require more education?

The likely shift is toward higher technical capability, judgment and adaptability. Some roles will require university education; others will depend on vocational training, certifications, apprenticeships and direct experience with advanced equipment and software.

ATN bottom line: a critical link is forming, but is not yet proven

The Q2 2026 evidence marks an important change in the economic discussion. The question is no longer only whether companies are spending heavily on computers, automation, software and AI. The question is whether that spending is producing measurable output.

Across nonfarm business, manufacturing, durable goods and nonfinancial corporations, the answer is increasingly yes. Productivity is improving. Output per worker is rising sharply. In the selected comparisons, employment is not keeping pace.

For markets, this can mean higher earnings, stronger cash flow and more capital available for research and development. For workers, it could mean fewer routine openings, greater pressure on entry-level employment and a premium on the skills required to build, operate and complement the technology. The compensation data do not yet prove that this occupational concentration is occurring.

These are not yet final scientific conclusions about AI and employment. They are the first pieces of a measurable economic pattern that ATN will continue to follow. If the same relationship persists through the next revisions and across more industries, the United States may be entering a new phase: stronger productivity, better corporate economics and substantially more complicated job creation.