Is the Productivity Revival Sustainable?

Given that several forces are driving it and the AI lift can only help, most likely yes.

Productivity is getting a lot of attention these days. Investors and employers are anxiously awaiting an AI-driven spike in productivity that could bolster corporate profits and real wages. It could also raise the potential real growth rate of the U.S. and help reduce the soaring federal budget deficits and debt accumulation as a share of GDP, while allowing the Fed to run the economy a little hotter than in the past without sparking a renewed bout of inflation. In short, AI is being marketed as the antidote to nearly everything that ails the United States economically and financially.

While it’s still a little early to see a meaningful improvement in productivity growth from AI investment alone, we are already seeing an encouraging upswing over the past year. In the four quarters through 2026Q1, nonfarm productivity (i.e., real output per hour of all persons) increased a robust 2.8%, even as gains slowed over the last two quarters (Chart 1). This is a sizable improvement compared to the annual average over the last 10 years (2.0%), and a bigger 1.2 ppt improvement compared to the 2010-to-2020 average of just 1.6% per year (Chart 2).

Macroeconomists are debating whether this resurgence is a temporary post-pandemic rebound or the beginning of a sustained AI-driven productivity cycle. The implications could not be bigger for Fed Chair Kevin Warsh, who has argued recently that stronger productivity growth could allow the economy to grow faster without generating inflation, opening a window for the Fed to cut interest rates despite resilient growth.

Indeed, as labor productivity lifts, nonfarm unit labor cost growth has fallen. This reduces the cost burden for producers, which could be passed along as smaller price increases for consumers or as higher profit margins. Over the past year, unit labor cost growth has averaged a modest 0.5% a.r. per quarter compared to a quarterly average over the past ten years of 2.3%. This points to the potential deflationary impacts of sustained increases in labor productivity growth (Chart 3).

Drivers Behind Recent Productivity Revival

Economists have pointed to a number of factors that may be driving the recent productivity growth resurgence, with AI and digital technology often at the top of the list. Businesses are increasingly deploying generative AI, automation, cloud computing, and advanced analytics to help workers complete tasks faster in areas such as software development, customer services, marketing, research, and administration. Many have compared today’s AI investment to the early stages of the internet boom in the 1990s, implying that the biggest productivity gains may still be ahead.

A 2025 Fed study on generative AI argued that it may be a general-purpose technology (GPT) like electricity or the computer and an “invention of methods of invention” (IMI) technology—meaning generative AI could accelerate research, innovation, and discovery itself [1]. If so, AI could potentially produce sustained productivity gains, though it still may take years for that support to diffuse throughout the economy. The Fed is increasingly asking not whether AI raises productivity, but how large and how fast the gains will be.

A March 2026 Atlanta Fed survey of nearly 750 corporate executives found widespread but uneven AI adoption with more than half of firms already invested in AI [2]. Productivity gains were generally positive though varying across sectors and expected to strengthen in 2026, with the largest increases in finance, professional services, and other high-skill sectors. Encouragingly, most of the gains come from improved efficiency and innovation (i.e. revenue-based total factor productivity) rather than simply adding more capital equipment (i.e. capital deepening). The study noted a productivity paradox, where perceived productivity gains are larger than measured ones, likely reflecting a delay in revenue realizations. For the labor market, the Atlanta Fed study found little evidence of near-term aggregate employment declines due to AI, though larger companies anticipate reductions. They also found evidence of compositional reallocation of labor both within and across firms, with routine clerical roles declining and demand for skilled technical roles increasing.

Research highlighted by the St. Louis Fed found users of generative AI in the workplace are completing certain tasks substantially faster [3]. It also found that the productivity benefits appear strongest in knowledge-work occupations such as software development, marketing, research, and administrative tasks. As more companies formally integrate AI into workflows, we may see these gains materialize more clearly in the aggregate productivity measures. A BLS report noted that experimental studies often find workers complete some tasks 20-50% faster, especially in coding, customer support, research, and writing. Translating those task-level gains into economy-wide productivity growth is slower and much harder.

In short, the Fed’s research appears to show that worker-level productivity gains from AI are already real. Firm-level gains are emerging, especially in services and finance sectors. Economy-wide impacts are gradually emerging but are still likely very modest.

Estimates of the size of the productivity gain vary, but a widely-cited study from the Congressional Budget Office assumes generative AI will raise total factor productivity growth by about 0.1 ppt per year on average over the next decade, lifting output 1% higher by 2036. That’s a meaningful, but not earth-shattering, impact.

In contrast, researchers at Anthropic (makers of the Claude AI models) believe generative AI could nearly double labor productivity growth as adoption spreads through the economy [4]. Its November 2025 report found current generative AI models could raise labor productivity by 1.8 ppts and total factor productivity by 1.1 ppts per year. Of course, Anthropic has a vested interest in touting the most optimistic scenario. In early 2026, the company lowered its estimate of potential U.S. labor productivity benefits to around 1.0-to-1.2 ppts per year, returning annual growth to rates last seen during the late 1990s tech boom [5]. Most mainstream estimates from private sector economists place the labor productivity impact of generative AI at a much smaller 0.1-to-0.3 ppts per year over the next decade.

Another productivity driver often cited is the business investment boom itself. Since the pandemic, firms have been investing heavily in software, data centers, semiconductors, automation equipment, and supply-chain technology. Strong business investment alone increases the amount of capital available per worker, which tends to raise output per hour worked. This is what economists call capital deepening. Reshoring and investments in semiconductors, clean energy, and infrastructure have already helped boost capital spending and will eventually improve productive capacity. The six major U.S. hyper-scalers plan to spend more than $800 billion on capex this year. We forecast growth in real business equipment spending will accelerate to 9.4% in 2026 from 8.3% in 2025, marking the best back-to-back years since 2011/2012 when the economy was recovering from the Great Recession.

Another factor is labor reallocation since the pandemic. The pandemic accelerated the movement of workers from lower- to higher-productivity sectors. During and after that period, many workers left industries such as leisure and hospitality, certain retail segments, and some personal services. At the same time, employment grew rapidly in technology, professional and business services, and finance. The pandemic also created intense competitive pressures, causing less-efficient businesses to close, while stronger firms expanded. Labor shortages also forced companies to streamline operations and adopt new workflows, allowing them to produce more with fewer work hours. They automated tasks, invested in software, reduced unnecessary meetings and paperwork, and improved scheduling and inventory management. Remote and hybrid work allowed companies to source skilled talent from a larger geographic area. The post-pandemic job switching surge in 2021 and 2022 also matched more workers to jobs that used their skills more efficiently.

A fourth factor is higher business formations since 2020. Business applications spiked following the pandemic and have remained high ever since. From 2009 to 2019, total applications averaged around 237,500 a month, according to the Census Bureau. Since June 2020, the average has nearly doubled to a whopping 450,200 a month (Chart 4). New firms often adopt the latest technologies and business models, which can raise the economy’s aggregate productivity over time.

What’s the Takeaway for the Fed? The productivity revival is significant and almost entirely coming from the services side of the economy. More importantly for monetary policy is that it may be accelerating. While generative AI and the AI investment boom are generating all the headlines, there are a number of other drivers helping boost productivity growth. Macroeconomists are still divided on the sustainability of recent productivity growth, and—critically for the Fed and its new Chair—the role of generative AI in those gains. We tend to fall into the camp that believes the impacts are relatively small, closer to 0.1% per year than 1.0%. Robert Solow famously stated in 1987: “You can see the computer age everywhere but in the productivity statistics”. It took a long time for firms to integrate computer and networking investments into workflows and processes; but once that happened, aggregate productivity growth soared by the late ’90s. Alan Greenspan, the Fed Chair at the time, used it as a reason to keep interest rates lower and economic growth higher without fear of sparking inflation. He was convinced technology and productivity improvements were lifting the economy’s potential growth rate. Kevin Warsh may be counting on a repeat performance in the late 2020s, though he may have to wait a bit longer before the true AI productivity miracle materializes.

[1] Generative AI at the Crossroads: Light Bulb, Dynamo, or Microscope? (July 2025) Martin Neil Baily, David M. Byrne, Aidan T. Kane, and Paul E. Soto. Board of Governors of the Federal Reserve System. https://doi.org/10.17016/FEDS.2025.053 [^]

[2] Artificial Intelligence, Productivity, and the Workforce: Evidence from Corporate Executives (March 25, 2026) Federal Reserve Bank of Atlanta. https://doi.org/10.29338/wp2026-04 [^]

[3] The Impact of Generative AI on Work Productivity (February 27, 2025) Alexander Bick, Adam Blandin, David Deming. Federal Reserve Bank of St. Louis. https://www.stlouisfed.org/on-the-economy/2025/feb/impact-generative-ai-work-productivity [^]

[4] Estimating AI Productivity Gains from Claude Conversations (Nov. 2025). https://www.anthropic.com/research/estimating-productivity-gains [^]

[5] Anthropic Economic Index Report: Economic Primitives (January 2026). https://www.anthropic.com/research/anthropic-economic-index-january-2026-report [^]

Fed Policy: Same Rate, Different Delivery

For Kevin Warsh’s inaugural FOMC meeting as Chair (June 17), the Fed left policy rates unchanged, as expected, with the target range for fed funds at 3.50%-to-3.75%. This marked the fourth consecutive pause, and was supported unanimously (the first dissent-less confab in a year). However, the policy statement was shockingly terse. Gone were the parts the Fed tends to repeat rotely (good riddance) and the roll call (we only need to know who dissented or was absent).

The language that had previously been construed as an easing bias and employed since rate cuts commenced in September 2024 was removed (also as expected). However, it was not replaced by any other forward guidance phrasing. Chair Warsh is not a fan of forward guidance, and it seems that his arguments won out. It also helped that policymakers are very divided about what to do next. In the dot plot, among the 18 participants penciling in their fed funds forecasts, for 2026, nine were in the hike clique and eight were in the pause gang (one had a cut). For 2027, there were eight in each of the hike and cut camps (two were still holding). The median projection portrayed no net change in policy rates over the next 18 months.

Warsh is also not a fan of the dot plot and did not provide a projection. We have argued that the market and media do not fully appreciate the fleeting nature of the dots. They are ‘point-in-time’ projections of the appropriate path for monetary policy given specific economic outlooks that are sensitive to new data and events. The dot plot becomes stale very quickly. And as an indicator of voting intentions, it is biased. Seven of the 19 potential dots (37%) are from regional Fed presidents who do not vote. And, historically, the presidents have tended to be more hawkish than their colleagues on the Board of Governors (7 of the 12 voters). There are others in the FOMC who are also skeptical of the dot plot. We reckon its days are numbered.

Importantly, the dot plot’s current survival, despite Warsh’s misgivings, emphasizes the fact that it takes a majority in the Committee (or on the Board in some cases) to alter the conduct of monetary policy. Warsh has one vote, but in his role as Chair of the Board and the FOMC, he has a platform on which to build moral authority. And he is going to need lots of it. Even more meaningful changes are coming, pending the results of the five task forces he established. They are looking into: (1) Fed communications; (2) the Fed’s balance sheet policy; (3) reliance on existing data sources; (4) productivity and jobs; and (5) the Fed’s inflation frameworks. In the press conference, Warsh said: “My expectation is the task forces will begin work in the next couple of weeks and we’ll start to get some more information from them, some more framing of how they see things, starting in the fall, and hopefully most, if not all of them, concluding by yearend.”

In the press conference, Chair Warsh sounded terse as well. He mentioned “price stability” 13 times but mentioned “maximum employment” only once, hinting at where his priorities might lie. Elsewhere, when asked whether policy was currently restrictive Warsh said: “If I look at the housing markets… Fed policy appears to be somewhat restrictive. I would have a hard time managing to say those words if I were to see what’s happening in financial markets, so I’d say it’s uneven.” This does not sound like a policymaker with a predilection to pare rates. Indeed, Warsh was considered a hawk during his first stint as a Fed governor. Perhaps he never shed those feathers.

Consumer: Do as I Do, Not as I Say

By one survey (the University of Michigan’s), American consumers have never felt worse about their financial situation. They’ve been irked by higher gas prices, elevated mortgage rates, falling real incomes, and AI-related job anxiety. But a funny thing happened on the way to the forum: spending accelerated this year. Retail sales jumped 0.9% in May, following an upwardly revised gain in April. The control measure of sales, which strips out gas station receipts and some other items and is used by the BEA to estimate quarterly consumer goods spending, leaped 0.7% and is up 8.6% annualized this year, the fastest pace since the early post-pandemic period. With almost no increase in core goods prices since December, most of the advance in retail sales is “real”. As a result, we upgraded our Q2 growth estimates for consumers to 2.2% a.r. and GDP to 2.5%, both marking the fastest pace since before the government shutdown.

Much attention has focused on the role of AI-led business investment in explaining the U.S. economy’s resilience in the face of new tariffs and higher fuel costs. And rightly so, as investment in computer hardware and software accounted for nearly a third (31%) of real GDP growth (2.6%) in the past year to Q1. However, the consumers’ contribution (61%) is almost double. While they are not punching above their weight like businesses, their sheer size (68% of GDP) and solid pace of spending (2.3%) matter greatly.

Surging equity values and personal income tax cuts have helped. We estimate that the former added about one percentage point to spending growth in the past year to Q1, while the latter chipped in a quarter point. As well, even as borrowing costs remain somewhat elevated, debt payments absorb less than their normal share of disposable income. Moreover, the assumed K-shaped nature of spending, where surging equities have juiced demand among higher-income households but rising fuel costs have squeezed lower-income earners, may be overstated. The New York Fed’s Household Spending Survey shows that year-over-year nominal spending growth of lower-income households (earning less than $50,000 per year) was slightly faster (4.9%) than that of higher-income groups in April. Still-low layoffs (apart from the technology sector, which tends to pay above-average wages) are likely supporting lower-income earners. Their resilience can only add to the expansion’s longevity.

That said, consumers have spent more by saving less, which has limits. As well, inflation remains sticky even outside of energy, which could erode purchasing power for a while longer. The boost from larger tax refunds will also fade. Still, if oil prices remain contained and equity markets buoyant, consumer spending could run around 2% in the year ahead and sustain similar growth in the broader economy, even if AI-led business spending cools its jets.