Tunguz models AI spend per engineer crossing a full salary by 2028-2029
Tom Tunguz's 'When AI Costs More Than the Engineer' models three paths for AI spend per engineer, with the base case crossing 100% of a fully-loaded $224k salary around 2028-2029 — anchored by Anthropic's 2.3x-payroll compute, roughly $2m per employee. The real signal is the spread: 2.3x at the frontier, 0.4x at the top of the market, near zero at the median, and whether it closes turns on token prices falling 10x a year against Goldman's projected 24-fold rise in token use by 2030.
Source: tomtunguz.com ↗
Anthropic spends 2.3x its payroll on compute. With ~5,000 employees & roughly $10b in inference & training spend in 2026, that works out to about $2m of compute per employee per year against a likely all-in comp of $500k+.
Tom Tunguz
Why this matters
- → AI infrastructure costs may soon exceed engineer salaries at frontier firms, reshaping unit economics.
- → 680x spending gap between top 1% and median suggests AI adoption remains highly concentrated.
- → Token price deflation vs. demand surge will determine whether AI stays optional or becomes mandatory capex.
Compute outpaces payroll