13 Aug 2026 · LinkedIn · 6 min read
The cost of AI talent
In 2013, $15 million for three researchers reset the market. It is now roughly the going rate for one.
What does exceptional AI talent cost per head? At the frontier today, $50–100 million per person, and occasionally more than $100 million. Over the past decade, the market for talent appears to have moved through three distinct eras.
Classic
2010 – 2017/~$15M / head
- In 2013, Google acquired DNNResearch — the University of Toronto spinout founded by Geoffrey Hinton, Alex Krizhevsky and Ilya Sutskever — for $40–44 million. Hinton effectively ran an auction, with Google, Microsoft, Baidu and DeepMind among the bidders.
- The auction had a second-order effect: it established a market-clearing precedent for valuing AI lab talent. Google later acquired DeepMind for $500–600 million (Demis Hassabis, Shane Legg, Mustafa Suleyman, et al.).
- Other deals followed, including Magic Pony, Nervana, Lattice, but the top end of the market remained broadly anchored around $10–15 million per person.
Middle
2018 – 2023/~$30M / head
- The period looked more like conventional strategic M&A, with a larger number of FAANG stock tuck-ins. Apple, Microsoft and Facebook acquired 300+ startups in this window.
- Focus shifted to technical capability-building plus talent plus codebase.
- Most notable: MosaicML, acquired by Databricks for $1.3 billion with ~60 employees, which pushed the implied value of talent above $20 million per person.
Frontier
2024 – present/+$100M / head
- Transaction structure changed significantly. The explosion of foundation models led to talent scarcity of elite researchers and engineers with demonstrable ability to train, scale and productize frontier systems.
- Deals grew larger and structures evolved into reverse acquihires — hiring key people and licensing tech without formally buying the company.
- One lab reportedly offering ~$100–$300 million pay packages over several years. In response, other labs compressed vesting schedules and issued large off-cycle retention grants.
- Side-effects: arguably weaker incentives for new startups, and impacts on remaining employees and investors. But also significantly higher talent mobility and a premium on human capital. IP shifting to team plus tacit knowledge, versus the codebase.
In 2013, $15 million for three researchers reset expectations. It is now roughly the going rate for one, and it reads more like the new floor for exceptional talent than the ceiling. A small group of exceptional researchers can plausibly create enormous enterprise value before the codebase. Going on first-time neolab fundraises, +$200 million a head looks plausible in the near term. The curve is steepening and the barbell widening.
Prices are a mix of official figures, PitchBook deal records, and press reporting from the WSJ, The Information, and the Financial Times. Several are best-available estimates — DeepMind is reported anywhere from $400–650 million, and DNNResearch’s ~$44 million surfaced later via the auction account but was never officially confirmed. Headcounts are either the people who moved that were reported, or else total staff on the deal announcement date.