US productivity is surging. The AI case is still thin

Key Points

1. America is in its strongest productivity surge in decades, but AI's share of it remains unclear
2. San Francisco Fed data credit capital investment and more intensive use of resources for most gains
3. Companies are spending hundreds of billions on AI without being able to isolate its return

The latest:

America’s strongest productivity surge in decades is proving hard to pin on artificial intelligence. Federal Reserve Bank of San Francisco data show investment and more intensive use of labor and equipment added 1.3 percentage points to annual productivity growth since early 2023 — nearly twice their contribution over the prior two decades. Executives say they cannot separate AI’s effect from everything else they changed.

Details:

• The headline number: Increased investment and more intensive use of labor and equipment contributed 1.3 percentage points to annual productivity growth since early 2023, according to the San Francisco Fed. That is close to double what the same factors delivered across the preceding two decades, making them the largest identifiable driver of the current surge.
• The cleaner measure: A broader efficiency measure that strips out extra investment and more intensive use of resources has grown 0.7% annually, against 0.5% between 2005 and 2022. The improvement is real but modest, far smaller than the capital-driven component, and not attributed specifically to AI.
• The overlap: Some of that investment may itself be AI-related, including software that lets employees do more in less time. That overlap is why the data cannot be cleanly split: spending on AI tools registers as capital deepening, not as a separate line labeled artificial intelligence.
• What executives say: Keith Waddell, CEO of staffing firm Robert Half, told investors last week that moving from individual productivity to organizational productivity involves “a lot of slip between cup and lip.” He added the firm is not ignoring that its people are getting some gains.
• The accounting problem: S&P Global targets roughly 2% in annual productivity gains but cannot isolate AI’s contribution from other operational changes. CFO Eric Aboaf said Tuesday on the company’s earnings call that some of it is directly AI, some is non-AI, and most is mixed and integrated.
• The Beige Book: Businesses in the Fed’s San Francisco district credited AI and traditional improvements — including staff training and streamlined processes — roughly equally for their productivity gains, according to the central bank’s Beige Book survey of regional business conditions.
• Where it is visible: The gains are far clearer at companies building AI. Anthropic disclosed in its IPO filing this week that Claude wrote more than 90% of contributions to its codebase last month, and that average daily code volume per engineer was more than 45 times its 2024 level.
• The market split: Revenue per employee and profit margins have climbed among the Magnificent Seven but stayed broadly flat across the other 493 companies in the S&P 500, Apollo chief economist Torsten Slok wrote over the weekend. Slok argued the AI story will be settled in the S&P 493, not the Mag 7.
• The other explanation: Part of the surge may trace to a post-pandemic wave of entrepreneurship now paying dividends, a far less-discussed driver that competes directly with the technology narrative for credit over the same period.

Between the lines:

The investment case rests on a measurement gap. Firms are spending hundreds of billions on a technology whose return they concede they cannot isolate, as S&P Global’s CFO effectively acknowledged. The strongest evidence so far — Anthropic’s codebase figures — comes from a company whose business is selling the technology, the narrowest possible test of whether it works everywhere else.

What’s next

Watch whether revenue per employee and margins start rising across the S&P 493 rather than only the Magnificent Seven, and whether upcoming Beige Book releases show firms outside the San Francisco district crediting AI more directly.