US business-sector output per hour grew at an average annualised rate of 2.52% from the start of 2024 to the second quarter of 2026, while utilisation-adjusted total factor productivity grew 0.12%, by our arithmetic on the San Francisco Fed’s quarterly series by economist John Fernald.
Economist Lars Christensen put the same comparison at “around 0.1% a year” against “around 2.5% a year” in a 1 November post on X, arguing that “That gap is the economy running hot, not AI”. Our numbers match his to one decimal place.
Business Pill · The Jevons Paradox
A short explainer on why cheaper, more efficient inputs can raise total use. It is a general explainer and is not about this story.
Animated: the gap between the San Francisco Fed output-per-hour index and the utilisation-adjusted TFP index, both chained from John Fernald’s quarterly series with 2023 Q4 = 100 (our arithmetic).
The key insight: On the San Francisco Fed’s series, US output per hour grew about 2.5% a year from the start of 2024 to mid-2026 while utilisation-adjusted TFP grew about 0.1%, by our arithmetic, a gap of about 2.4 percentage points a year on the same series.
Fernald’s quarterly sheet reports every variable as a percent change at an annual rate. Averaging the ten quarters from the first quarter of 2024 through the second quarter of 2026, labour productivity grew 2.52% a year, unadjusted TFP 1.10% and utilisation-adjusted TFP 0.12%, by our arithmetic. The sheet is dated 1 October 2026.
Chained into indexes with the fourth quarter of 2023 set to 100, output per hour reached 106.5 by the second quarter of 2026 and utilisation-adjusted TFP 100.3, a gap of about 6.2 index points, by our arithmetic. Utilisation-adjusted TFP fell in five of the ten quarters, including the latest two.
US business-sector output per hour (San Francisco Fed and BLS series) and utilisation-adjusted total factor productivity, indexed to the fourth quarter of 2023. The indexes are chained by FourWeekMBA from John Fernald’s quarterly log changes and the BLS index; the averages are our arithmetic.
The Official Count
The Bureau of Labor Statistics’ nonfarm business output-per-hour index rose from 113.881 in the fourth quarter of 2023 to 120.017 in the second quarter of 2026. That is about 5.4% in total, or about 2.1% a year compounded, by our arithmetic, a little below the San Francisco Fed series.
The Chart Behind the Debate
Christensen was responding to an FT chart by John Burn-Murdoch titled “Productivity is surging in AI-adopting sectors in the US, but not elsewhere”. It plots real output per worker in tech, finance and professional services in eight economies against a 2012-19 trend.
Christensen wrote that US productivity in these sectors “took off in 2020, well before ChatGPT” and called it “an overheating economy”. He later added: “I thought @jburnmurdoch made the claim this was AI driven. He did not. I was wrong.”
The Business Lever: Measured Gains Versus Efficiency
As we read it, the two series answer different questions. Output per hour counts how much is produced per hour worked; utilisation-adjusted TFP strips out how hard capital and labour are being run, which is where Christensen locates the recent gains. Our earlier coverage of EPB Research’s sector productivity trend found faster growth in AI-heavy sectors on a different measure.
As we read the two series, they separate how much is produced per hour from how efficiently inputs are combined once their intensity of use is taken out.
As we read the BLS index, the official count also shows strong output-per-hour growth, about 2.1% a year compounded by our arithmetic, a little below the San Francisco Fed’s 2.5%.
As we read the debate, Christensen’s overheating reading and an AI reading make different predictions for the adjusted series; so far it has barely moved, by our arithmetic on the 1 October vintage.
Lars Christensen, on X, 1 November 2026
“That gap is the economy running hot, not AI”
Three Implications
FOR THE AI PRODUCTIVITY CASE Neither series measures AI use directly; the FT chart, by John Burn-Murdoch, plots output per worker in tech, finance and professional services against a 2012-19 trend.
FOR FORECASTERS Fernald’s sheet is dated 1 October 2026 and its utilisation adjustment is model-based, so the 0.12% average is our arithmetic on that vintage.
FOR THE OFFICIAL COUNT The BLS output-per-hour index rose from 113.881 to 120.017 between the fourth quarter of 2023 and the second quarter of 2026.
The framework puts it this way: “Both are true. The mistake is treating them as competing claims.”
As we read it, the productivity debate maps onto that split: measured output per hour has risen strongly, while the adjusted efficiency series that would show a broad AI gain has barely moved so far. The framework does not discuss these series.
Neither series measures AI use directly, and the overheating explanation is Christensen’s interpretation, not a finding of the San Francisco Fed or the BLS. Utilisation adjustments rely on Fernald’s model and are revised between vintages. The averages are our arithmetic on the 1 October 2026 vintage and the BLS index as published.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Output per hour grew about 2.5% a year since the end of 2023 on the San Francisco Fed series, and about 2.1% on the BLS index, while utilisation-adjusted TFP grew about 0.1%, by our arithmetic.
A note on sourcing. Growth rates come from John Fernald’s quarterly utilisation-adjusted TFP series published by the Federal Reserve Bank of San Francisco (1 October 2026 vintage) and the BLS nonfarm business output-per-hour index; averages, chained indexes and the gap are our arithmetic. Quotations come from Lars Christensen’s 1 November post on X.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
