When AI Costs More Than the Engineer
6 hours ago
- #Market Scenarios
- #AI Spend Trends
- #Compute vs Payroll
- Anthropic's compute spend is 2.3x its payroll, with about $2 million per employee annually, while the top 1% of companies spend $89k per engineer (40% of salary), and the median spends only $137.
- Three scenarios (Bear, Base, Bull) project annual AI spend per engineer from 2026 to 2029, with Bull case reaching $596k (230% of salary) by 2029, matching median SaaS employee revenue.
- Bull drivers include stable frontier model prices, agentic workflows increasing token consumption 24-fold by 2030, and competitive pressure; Bear counterweights involve token price deflation and cost-effective open-weight models.
- Anthropic and OpenAI lead in revenue per employee at $14 million and $6.5 million respectively, reflecting a cost structure where infrastructure dominates payroll.