Productivity & the AI Revolution — Implications for the Economy and Markets

Across the economic landscape, no variable looms as large as productivity growth — it’s the closest thing there is to a “free lunch.”  Perhaps a few times a century, the path of productivity is pushed upward by a transformative new technology — the steam engine, the railroad, electrification, the automobile, and the internet are notable examples.  We believe artificial intelligence (AI) is likely to soon be added to the list.  A new Must C report from Citi Research examines the linkages between U.S. productivity and the emerging AI revolution, addressing questions including whether AI will bring an appreciable uplift to U.S. productivity growth and, if so, when; how large the gains from AI are likely to be; and what the implications are for investors and financial markets.      

The following are some conclusions that emerge from our work:  

• U.S. productivity growth has bounced back in recent years relative to the lows posted after the global financial crisis.  This acceleration is particularly notable in the services sectors.  However, the drivers of this rebound look to be more cyclical than structural.  As such, we see little evidence to date of a sustained acceleration in productivity as a result of emerging AI technologies.
• The academic literature suggests that roughly 20 to 40% of production tasks could potentially be automated with AI, and the resulting labor cost savings are likely to be around 30 to 40%.  This implies a total gain in productivity of 6 to 16%, or a boost to productivity growth averaging ½ to 1½ ppts a year if these gains accrue over a decade.
• We judge that the diffusion of AI is proceeding at a historically rapid pace.  With other transformative technologies, meaningful gains often were not reaped until decades after their first introduction.  We hypothesize that AI’s comparatively rapid diffusion reflects the highly competitive and fluid nature of the modern economy, as well as the power and accessibility of AI technologies.
• We devise a new measure of U.S. AI investment which draws on widely available macroeconomic data. This measure is centered at $60 billion in 2023:Q4, $150 billion in 2024:Q4, and $255 billion during Q2 of this year.  AI investment thus looks to be supporting US economic growth through its demand-side effects.  We judge that this upward impulse has been equal to roughly 20 to 40bp in recent years.
• Even so, these are still “early days” for AI adoption.  Recent surveys find that only 5% of GenAI projects are fully scaled and creating meaningful value.  And there is debate as to whether the recent growth of AI investment will be sustained.  Other “speedbumps” that will determine the pace (and extent) of AI diffusion include worker acceptance, the construction of robust physical infrastructure (including sufficient power supplies), steps to ensure the reliability of information, and the creation of supportive legal and regulatory frameworks.
• By our reckoning, AI investment is rapidly approaching levels that characterized the 1990s productivity boom.  A straight read-through of our estimates indicates that the United States could see a similar productivity boom within the next few years.  However, these estimates are not sufficiently precise — and the lessons of history not sufficiently extrapolatable—to make more specific predictions.
• The further diffusion of AI would also have first-order implications for financial markets.  We find that an acceleration in productivity driven by AI advances would bring higher real bond yields and lower gold prices than currently prevail.  In principle, it would also mean higher equity prices and a stronger dollar.  For these markets, however, such effects look, at least in part, to already be priced in.
  • More specifically, the “AI trade” for equities initially focused on companies producing AI technologies, the AI “enablers.” But the focus is now broadening to include AI “adopters.”  As more companies find ways to employ AI technologies, the resulting productivity gains should further propel U.S. equity prices.  Even so, this dynamic could leave markets vulnerable if the benefits from AI are less pronounced, or come more slowly, than investors currently expect. 

Bottom line, AI technologies are clearly still in an adoption phase in which the necessary capital—data centers, electrical grid, and hardware—are being developed and built out.  In tandem, AI experts are being trained, and workers are adapting to the new technology at varying paces across firms and industries.  But, at the same time, the potential productivity benefits from this technology look significant.  And the realization of such gains, while still uncertain, strikes us as increasingly likely and proximate. 

The remainder of the report considers the recent performance of U.S. productivity, develops an estimate of U.S. AI investment, and assesses where we currently stand in the adoption cycle.  As part of this discussion, we consider some lessons from history, including the experience of the internet productivity boom and, even earlier, the adoption of electricity.  We conclude with reflections on the implications of these developments for financial markets and investor portfolios.