One of generative AI’s biggest promises is the idea that it can help smaller companies and startups take on established firms by providing access to advanced capabilities.
Yet since the introduction of ChatGPT in 2022, investors have placed their biggest bets on the more dominant companies—those they saw as best positioned to benefit from the new technology, according to research by Harvard Business School Professor Suraj Srinivasan. And they were rewarded: these companies’ returns outperformed their peers’ by 19 percentage points in the year following the platform’s launch.
ChatGPT sparked the imaginations of leaders at companies of every size and industry when it emerged four years ago, with investors watching closely. So far, large organizations have made more progress on adopting AI agents than smaller firms, according to a McKinsey survey.
Srinivasan’s findings—detailed in the working paper “Generative AI and Market Expectations of Superstar Firms”—shed light on why.
While “lean organizations might adopt the new technologies faster; complex ones have more friction to remove, and thus greater benefits from adopting these technologies,” says Srinivasan, the Philip J. Stomberg Professor of Business Administration.
Srinivasan coauthored the paper, updated in September, with Wilbur Chen, a professor at the Hong Kong University of Science and Technology, and Bullipe R. Chintha, a professor at the Singapore Institute of Technology.
Does innovation topple leaders—or strengthen them?
The researchers set out to test a question raised by French economist Philippe Aghion in his 2025 Nobel Prize lecture: does a major technological shift topple dominant firms—or lock them in?
The answer they arrived at—that the market favors complex organizations already doing work AI is best at—surprised Srinivasan and his colleagues.
“In hindsight, the economics make sense,” he says. “A large superstar firm is, among other things, an expensive machine for moving information.”
But the premium wasn't just about organizational complexity. The study found that these companies also gained from their research and operational know-how, and deep benches of data talent. The gen AI premium ran 28 and 29 points, among firms rich in these two categories of intangible assets, respectively, and was absent among firms without them.
What sets superstar AI companies apart
To answer Aghion’s question, Srinivasan and his team analyzed 2,596 US companies, leveraging data from Compustat and the Center for Research in Security Prices.
The researchers were particularly interested in “superstars,” which they defined as the top 10% of companies by market value relative to their invested tangible and intangible capital. They calculated that relationship using an adjusted version of Tobin's Q, a widely used valuation measure.
To identify firms that are more likely to be affected by generative AI, the team used occupational data from Revelio Labs to measure their workforce exposure to the technology. Firms scoring above the sample’s median were classified as “highly exposed” to the technology.
After mapping the companies across those dimensions, “we then tracked their returns in the months following ChatGPT’s release as a measure of investors’ revised expectations about their future performance,” Srinivasan says.
The researchers were surprised to see “how sharply the market separated the GenAI-aligned superstars from others,” Srinivasan says.
At the two-year mark, the study found that the market’s conviction in its judgment only got stronger: the higher return of superstar firms highly engaged with AI’s best side had doubled to 39 points from the first-year premium.
Do superstars always win?
Srinivasan and his team considered other potential explanations for why investors may have favored highly AI-exposed superstar companies over their peers.
First, they ran the identical analysis on companies as they existed before ChatGPT’s introduction, finding that the combination of superstar status with high exposure in AI produced essentially no benefit in that time frame.
“The superstars-always-win logic fails,” Srinivasan says. “If this were generic momentum, all superstars should have kept rising; instead, the premium appears concentrated where dominance meets a gen AI-exposed workforce.” But, he says, “both classifications were fixed before ChatGPT existed.”
Similarly, the researchers wanted to know whether they were seeing effects of general software-driven automation, rather than a surge from generative AI. So they ran the analysis again, this time using a broader measure of exposure to general purpose software technology, rather than one specific to gen AI.
That test flipped the script: companies that combined dominance with general software technology exposure underperformed what those two traits would have gained on their own, by a statistically significant 9.8 percentage points.
“The gap that widened is not between big companies and everyone else,” Srinivasan says. “It’s between superstar companies that are more likely to benefit from generative AI and everyone else.”
What it takes to succeed with AI
Srinivasan says their study’s findings offer several lessons for companies still defining their relationship to AI:
Benchmark your company first
Determine where you register on the scale of size and performance (i.e., superstar companies) and gen AI exposure before developing a strategy.
"Think about what share of your workforce is involved in knowledge work that gen AI can actually do well," Srinivasan says. "Your position should [set] the budget—for genuinely low-exposure firms, gen AI doesn't need to be at the center of the capital plan."
He cites Chipotle as an example of a low-exposure superstar: the restaurant company focused on mechanical automation initiatives, such as an avocado-processing prototype, rather than a gen AI-centric strategy—in Srinivasan's view, "the right read of its position."
Bulk up on data knowledge and talent
Superstar companies such as Visa and ServiceNow carried roughly one data-management employee for every 11 workers into the ChatGPT era, compared to one in 40 at the typical firm, Srinivasan says.
"Any firm navigating an AI investment plan should work that lever through attrition and hiring and should think of this as a long-term investment," he says. "Tilting your workforce toward data- and AI-heavy roles compounds over time and prepares you for the evolving generations of this technology."
Target AI for functions that thrive on information
"One managerial instinct could be to deploy generative AI on visible individual work like coding or customer service," Srinivasan says. But gen AI's demonstrated strengths in translation, summarization, and synthesis "mean the gains come easier at the layers of the organization, in status updates, project summaries, and communication between units," he says.
Image: Ariana Cohen-Halberstam
Have feedback for us?
Generative AI and Market Expectations of Superstar Firms
Chen, Wilbur, Bullipe R. Chintha, and Suraj Srinivasan. "Generative AI and Market Expectations of Superstar Firms." Harvard Business School Working Paper, No. 26-083, May 2026. (Revised September 2026.)

