AI is redefining the economy. This is a not a question or a prediction, but a fact. Companies such as Meta, Amazon, Alphabet, and Microsoft are investing hundreds of billions of dollars in AI data centers in 2025—the equivalent of the GDP of small nations. From November 2022 – when ChatGPT launched – to the end of 2024, 65% of the growth in market capitalisation of the S&P 500 came from companies that either deploy AI or integrate AI into their core operations. Labor is being redefined by the emergence of AI and agents acting as general purpose technologies (GPTs) as defined in GPTs Are GPTs (2023). At this point, we have enough evidence to suggest AI is redefining the economy. Yet despite this unprecedented investment and market response, we still lack granular empirical evidence about how AI exposure translates into actual economic transformation at the firm level.
A more interesting question, then, is how is AI impacting the economy. If we focus our question narrowly enough, such as, "how are the dynamics of AI reflected in labor demand among firms?" we can start to see some interesting patterns. At least, those are the claims my co-authors and I make in our recently published paper Extending "GPTs Are GPTs" to Firms (2025) in the American Economic Associations (AEA) Papers and Proceedings.
Using workforce composition data from 7,894 publicly traded firms, we find unmistakable variations in AI exposure across companies. The average firm has approximately 17% of its workers' tasks exposed to large language models alone—but this figure jumps to 47% when accounting for partial integration with complementary software tools such as GitHub Copilot. These aren't hypothetical projections; they represent measurable exposure based on the actual occupational mix within each firm's workforce.
Perhaps most intriguingly, as my coauthor Sam Manning highlights in his analysis, we document substantial gaps between firms' measured exposure to AI and their reported adoption rates. This suggests that AI integration is happening organically at the worker level, potentially flying under the radar of executive surveys and formal adoption metrics. Tech firms and those with higher concentrations of AI-skilled workers show the highest exposure, while larger firms consistently outpace smaller ones—patterns that align with broader adoption trends but reveal much more granular economic dynamics.
This bears repeating: the impacts of AI already seem to be dependent upon productivity exposure and are highly-varied between firms in the economy—with potentially the biggest benefits going to "superstar firms" with the greatest digital capital (see Digital Capital and Superstar Firms (2020). Do these results imply that AI will inflame labor inequalities, or attenuate them? And, will these inequalities be most visible intra-firm or inter-firm? This is an area of active research among my co-authors and me.
What we do know now is that the redefinition of the world economy by AI may end up being uneven—with superstar firms reaping greater benefits. That being said, such an outcome wouldn't preclude more economy-wide gains. Smaller firms may well become more productive, relying on complementary technologies such as Copilot to operate with leaner labor forces, producing more than they'd do on their own. Moving forward, the question isn't whether AI will reshape competitive advantage—it's whether and how firms will recognize the transformation while it's happening.