An AI consultant told Axios that one of their enterprise clients burned through roughly $500 million on Anthropic's Claude in a single month after handing thousands of employees licenses with no spending caps, no usage limits and no dashboards to track consumption in real time.
The costs spiraled because modern AI usage is billed per token and amplified by autonomous agents, large-context memory windows and background workflows that run around the clock. Engineers using agentic coding or parallel large-context sessions can each generate hundreds to thousands of dollars per month — numbers that turned catastrophic once multiplied across an entire workforce with no guardrails.
Anthropic's enterprise console already offers role-based access, automated alerts and hard spending caps, but they must be switched on deliberately. The episode echoes a broader squeeze: Microsoft reportedly pulled back internal Claude Code licenses after per-engineer costs hit $500 to $2,000 a month across its teams.