Top 10% of Firms Take 73% of AI Hires, but the Leaders Keep Changing

In 2010 the top tenth of hiring firms took in 55.8% of the AI workers who changed employers. By 2022 that share was 73.0%. A new arXiv preprint by researchers at MIT, Peking University, the Singapore-MIT Alliance for Research and Technology and the University of Florida follows those moves through about 535 million employment records, and finds that the firms at the top keep being replaced.

AI hiring concentrates while the top ranks rotate

[Fact] The authors used Revelio Labs profile data to rebuild, for each year from 2010 to 2022, which firms hired AI workers from which other firms. The panel covers 23,596 listed firms headquartered in the United States, Europe and Southeast Asia. A position counted as AI-related when its job title or description matched at least one of 213 keywords.

[Fact] Concentration rose. The Gini coefficient of AI talent inflows climbed from 0.602 in 2010 to 0.785 in 2022, while the figure for non-AI moves stayed near 0.885. In all three markets, firms that had already hired many AI workers attracted a more-than-proportional share of later hires. For non-AI workers the estimate fell below proportional.

[Fact] The core changed members quickly. The paper defines the core as the top 10% of firms by PageRank, a score that rises when a firm hires from other well-connected firms. In the United States 72.6% of AI-core firms were still in the core a year later, against 90.5% in the non-AI network. In Europe the figures were 66.9% and 91.0%, and in Southeast Asia 32.0% and 79.8%. Each year 3.6% of US firms outside the AI core entered it, compared with 1.1% in the non-AI network.

[Fact] Network position tracked market value. A one-standard-deviation rise in a firm's AI PageRank was associated with about 2.3% higher enterprise value the following year, after controlling for AI headcount, other headcount, assets and cash (20,413 firm-years from 3,329 firms). Across 136 cases where a firm jumped sharply up the ranking, enterprise value afterwards ran 0.100 log points above matched firms. [Estimate] That converts to roughly 10.5%. [Fact] Revenue rose after those jumps, evidence on profit was weaker, and cost-to-revenue ratios did not fall.

[Fact] The central employers differ by market. Seven of the ten most central US firms in 2022 were technology companies, led by Amazon, Alphabet, Meta, Microsoft and Apple. In Europe, Accenture and Capgemini ranked first and second, followed by Airbus, Siemens and Volkswagen. In Southeast Asia, Sea, Grab and GoTo led a top ten split evenly between technology firms and banks such as OCBC, UOB and DBS.

What data scientists and developers can prepare

The paper sorts positions by keyword and reports no occupation codes. [Estimate] The job titles that match its machine-learning terms most directly belong to data scientists, computer and information research scientists and software developers who build AI systems.

• Check this year's hubs. [Estimate] From the retention rates, about 27% of US AI-core firms and 33% of European ones leave the core each year. A target-employer list drawn up two years ago is partly out of date.
• Look beyond technology companies outside the US. [Fact] In Europe, moves by AI workers from manufacturing and from finance into technology firms ran well above the level that non-AI job moves would predict, and the most central employers included consultancies and manufacturers. [Estimate] Sector experience combined with AI skills is an established route in that market.
• Write the AI work into your profile. [Fact] A position whose title and description contained none of the 213 terms was classified as non-AI. [Claim] The authors suggest that central firms gain "implementation know-how" carried by incoming workers, something headcount does not show. Naming the models you built, deployed or integrated makes that experience visible to anyone reading profiles the way this study did.
• Ask where a prospective employer hires from. [Claim] The authors read a high PageRank as a sign that a firm attracts specialists from other well-connected employers, and suggest it may signal career opportunities.

Limits

The study period of 2010 to 2022 ends before generative AI tools spread, and it covers only moves between listed firms within the same market, drawn from online profiles that represent countries and sectors unevenly. The valuation results are associations: the authors note that improving business prospects may attract both talent and investors, and the study measures no wages or outcomes for the workers themselves. The paper is a preprint that has not been peer reviewed, and its Supplementary Information was not read for this summary.

Sources

• Donghang Li, Yunhan Zheng, Alok Prakash, Shenhao Wang and Jinhua Zhao, "Understanding Interfirm AI Talent Flow Networks through Online Professional Profiles", arXiv:2610.08264, October 6, 2026: https://arxiv.org/abs/2610.08264

About this analysis

This article was written with AI assistance, from the full main text of the preprint. Figures marked [Estimate] (the 10.5% conversion, the 27% and 33% exit rates, the occupation mapping, the sector-experience reading) are our own arithmetic or reading of the paper's reported numbers. No figures or tables from the paper are reproduced.