AI tools are everywhere. Profits may not be.
Listen to this article:
Generative AI looks like a startup’s dream. It’s widely available, relatively cheap to access, and capable of automating tasks that once required long hours and large teams. In theory, that should give newer, smaller firms an edge for challenging incumbents. But as reported in the new working paper “Generative AI and the Superstar Firm Effect,” co-written by former PI of the HBS AI Institute’s Digital Value Lab Suraj Srinivasan, early evidence from public markets points in a different direction: investors appear to believe GenAI’s biggest gains will accrue to larger, established firms, not their leaner rivals. Using the public release of ChatGPT in November 2022 as a natural experiment, Srinivasan and his co-authors examine how investors repriced the economic potential of generative AI across thousands of public companies. What they found should interest anyone thinking carefully about where AI value actually lands.
Key Insight: AI Superstars
“In this paper, we examine how the sudden and widespread emergence of the latest general-purpose technology, GenAI, impacts the superstar firm effect.” [1]
The research begins with a practical question: when a powerful general-purpose technology becomes universally accessible, who captures the value? One possibility is that the new tech benefits disruptors. GenAI tools are accessible through APIs, open-source models, and rented computing infrastructure, which could lower the cost of experimentation for smaller firms. Another possibility is entrenchment. In this case, large incumbents may have an advantage because they already own the surrounding assets, like data, expertise, and integration capacity, that make AI tools productive.
To find an answer, the authors looked at “superstar” firms, those in the top decile of intangible-adjusted Tobin’s q, a valuation metric that captures how much investors value a firm relative to the replacement cost of its tangible and intangible assets. They also tested alternative definitions based on market capitalization, return on invested capital, and firm-level markups. To measure GenAI exposure, the authors scored occupations in the O*NET database (a dataset that contains comprehensive information on the occupations of US workers in 923 distinct occupations) based on how well LLMs could perform their component tasks, and then mapped those scores to each firm using detailed workforce composition data. A firm whose employees spend most of their time on tasks that AI models can replicate scores high on GenAI exposure. They then compared stock-market reactions around ChatGPT’s release and over longer horizons, including 12-month and 24-month adjusted returns.
Key Insight: The Premium Lives in the Interaction
“[I]t is the combination of superstar status and GenAI exposure that drives abnormal returns, not either characteristic in isolation.” [2]
In the first ten trading days after ChatGPT’s launch, the market priced GenAI exposure and superstar status as separate positives, but didn’t see a special bonus from combining them. Twelve months in, that picture changed. Superstars with high GenAI exposure earned a 2.5% monthly characteristic-adjusted return, more than four times the 0.6% that non-superstars with similar exposure managed, while everyone else sat near zero. Cross-sectional regressions estimate the interaction at roughly 19 percentage points at one year and 39 at two years, suggesting the market’s confidence in these firms keeps building.
The authors provide two frameworks to explain why. The first is complementary assets. GenAI generates value where it has rich material to work with: proprietary data, organizational knowledge, and domain expertise. Large incumbent firms have the thick intangible capital bases and dedicated data infrastructure needed to deploy GenAI effectively at scale. The superstar premium is large among firms with high intangible capital and high data capital, and absent among firms with low levels of these assets. The second framework is coordination cost reduction. Large, complex organizations, those with many layers of middle management and employees spread across multiple geographies, face significant internal friction in communication, synthesizing information, and aligning decisions. GenAI’s language processing capabilities directly attack those frictions. Consistent with that idea, the superstar premium is strongest among firms with thick middle-management layers and geographically dispersed workforces.
Key Insight: A Widening Moat
“[I]nvestors rewarded superstars positioned to capitalize on GenAI while penalizing those that were not.” [3]
The authors strengthen their superstar premium finding with placebo and robustness tests. Absent before ChatGPT, the GenAI superstar premium grows over the two years of the study and persists even when excluding tech firms and the Magnificent Seven. Replacing GenAI exposure with general software exposure eliminates the premium, confirming this isn’t just a generic technology effect.
Together, these findings point toward a meaningful structural shift: the firms best positioned to benefit from GenAI are precisely those that already dominate their markets. They hold the data and organizational complexity that makes coordination tools valuable, and the financial capacity to invest in integration. The result is a potential amplification of the superstar firm effect—a concentration of productivity and profit that competition policy has struggled to address even before AI entered the picture. The authors also find early signs that the market response was connected to real behavior: high-GenAI superstar firms discussed GenAI more frequently in earnings calls, and by fiscal year 2023 showed stronger profitability than comparable firms.
Why This Matters
If the market is correct, the lesson for newer firms is sobering but useful. GenAI may not level the playing field, but neither does it lock in incumbents forever. Younger firms might still have advantages to compete, especially in narrow domains where their data is uniquely deep, in workflows where superstar organizational complexity becomes a tax, or by partnering to access the kind of intangible capital that takes years to build alone. The market is already picking winners. Leaders must urgently build the complementary data and organizational assets that make GenAI uniquely productive to ensure that their firms will be part of them.
Bonus
Even if incumbents have advantages thanks to GenAI, not every company knows where to use it. For a closer look at why AI value depends on finding the right workflows, not just adopting the right tools, read Everyone Has AI. Which Firms Are Going to Win?
References
[1] Chen, Wilbur, Bullipe R. Chintha, and Suraj Srinivasan, “Generative AI and the Superstar Firm Effect,” Harvard Business School Working Paper, No. 26-083 (May 2026): 2.
[2] Chen et al., “Generative AI and the Superstar Firm Effect,” 23.
[3] Chen et al., “Generative AI and the Superstar Firm Effect,” 23-24.
Meet the Authors
Wilbur Chen is an assistant professor in accounting at the Hong Kong University of Science and Technology.
Bullipe R. Chintha is an assistant professor at the Singapore Institute of Technology.
Suraj Srinivasan is Philip J. Stomberg Professor of Business Administration and Chair of the MBA Elective Curriculum at Harvard Business School, and former PI of the HBS AI Institute’s Digital Value Lab.
Watch a video version of the Insight Article here.