---
id: "lib-2026-10-06-understanding-interfirm-ai-talent-flow-networks"
type: "library"
title: "Understanding Interfirm AI Talent Flow Networks through Online Professional Profiles"
lang: "EN"
org: "MIT / SMART"
source_type: "Paper"
date: "2026-10-06"
week: "2026-W41"
url: "https://nosov.ai/library/lib-2026-10-06-understanding-interfirm-ai-talent-flow-networks/"
original_url: "https://arxiv.org/abs/2610.08264"
tier: "A"
tier_name: "Strong evidence"
tier_reason: "MIT, Peking University and Singapore-MIT Alliance working paper built on about 535 million Revelio Labs employment records with panel and matched event-study estimates; a preprint, not yet peer reviewed."
summary_basis: "full-text"
original_minutes: "45"
tags: ["future of work", "talent strategy", "ai talent"]
---

# Understanding Interfirm AI Talent Flow Networks through Online Professional Profiles

Tier A (Strong evidence): MIT, Peking University and Singapore-MIT Alliance working paper built on about 535 million Revelio Labs employment records with panel and matched event-study estimates; a preprint, not yet peer reviewed.

Summary based on the full text.

## In short

An arXiv preprint (6 October 2026) by Donghang Li, Yunhan Zheng, Alok Prakash, Shenhao Wang and Jinhua Zhao (MIT, Peking University, SMART, University of Florida). It reconstructs how AI workers move between firms in the United States, Europe and Southeast Asia from 2010 to 2022. AI hiring is concentrating in fewer firms, yet the set of central firms changes far more often than in ordinary labour mobility. A firm's position in these flows predicts its later market value beyond the size of its AI workforce.

## Key numbers

- 73.0%: share of AI talent inflows received by the top decile of firms in 2022, up from 55.8% in 2010
- 72.6% vs 90.5%: year-to-year retention of the US AI network core versus the non-AI core
- 2.3%: higher next-year enterprise value per one-standard-deviation rise in AI network centrality, net of AI headcount
- 0.100: log-point enterprise value gain after a sharp jump in network position, versus matched firms

## What it reports

- Data: about 535 million employment records across 58 countries; the US AI network in 2022 covers 2,255 firms, with the most connected firm linked to 711 partners.
- Top AI hubs by PageRank in 2022: in the US, Amazon, Alphabet, Meta, Microsoft and Apple lead, with JPMorgan Chase, Walmart and General Motors in the top ten; in Europe, Accenture and Capgemini lead, followed by Airbus, Siemens and Volkswagen; in Southeast Asia, Sea, Grab and GoTo.
- The pooled Gini coefficient of AI talent inflows rose from 0.602 in 2010 to 0.785 in 2022 (non-AI stayed near 0.885). The top decile took 73.0% of AI inflows in 2022, up from 55.8%.
- Preferential attachment is stronger for AI hiring: exponents of 1.046 (US), 1.035 (Europe) and 1.239 (Southeast Asia), against sub-linear 0.965, 0.960 and 0.972 for non-AI flows.
- Core retention is lower for AI: 72.6% vs 90.5% (US), 66.9% vs 91.0% (Europe), 32.0% vs 79.8% (Southeast Asia). Annual entry into the AI core: 3.6%, 4.1% and 7.4%, versus 1.1%, 1.2% and 2.8%.
- A one-standard-deviation rise in log AI PageRank is associated with 0.023 log points higher next-year enterprise value (s.e. 0.007; 20,413 firm-years, 3,329 firms); the EBITDA link is smaller and weaker (0.013, P = 0.078).
- After sharp improvements in network position, enterprise value rose 0.100 log points relative to matched firms (s.e. 0.048, P = 0.036), with no pre-event divergence.

## How they got it

AI positions were flagged by matching job titles and descriptions against a keyword dictionary, and firm-to-firm moves were built from individual career histories, keeping only moves within each headquarters market. Annual directed networks were analysed with centrality measures and preferential-attachment estimates. Valuation links use firm and year fixed effects with controls for AI and non-AI headcount, assets and cash, plus an event study with matched controls for 2013-2019 rank jumps.

## Limits

- Online professional profiles over-represent some roles, countries and firm types; keyword classification of AI roles will misclassify some positions.
- The valuation results are associations and predictive patterns, not proof that hiring position causes value.
- Data stop in 2022, before the generative AI hiring wave.

## Why it matters for companies

For a traditional company, the useful finding is that the AI talent core is contestable: firms outside Big Tech do enter it, and those that move up see value gains relative to peers. Where a firm's AI hires come from, and where its alumni go, is measurable and appears to matter beyond headcount. The European pattern, with consultancies and industrial firms such as Siemens and Volkswagen at the centre, suggests non-tech employers can be talent hubs. HR and strategy teams can map their own position in these flows using the same kind of profile data.

Original: https://arxiv.org/abs/2610.08264