Anthropic Compute Spend Reaches $2m per Employee in 2026

Anthropic operates at a 2.3x compute-to-payroll ratio with 5000 employees. The top 1% of software firms spend $89k per engineer on AI, 40% of a $224k senior salary, while the median firm spends $137. Three scenarios track convergence: bear shows token deflation capping spend at $106k by 2029, base reaches $363k, and bull hits $596k when agentic workflows scale token demand 24x per Goldman Sachs projections.

Token prices have declined 10x annually since 2023 while open-weight models narrow capability gaps. Frontier revenue per employee already hits $14m at Anthropic and $6.5m at OpenAI. When AI spend alone matches or exceeds median SaaS revenue contribution per employee, hiring freezes and workload rationing become structural responses rather than temporary cost controls.

Bull case assumptions require sustained frontier pricing, training cost plateaus, and demand exceeding supply through 2029. Companies enforcing role-based limits slow the curve. The economic threshold arrives when incremental model output no longer justifies marginal token cost versus an additional engineer, reshaping startup capital efficiency and enterprise headcount planning.