Breaking

Meta and Microsoft Reduce Claude Spending as Enterprise AI Costs Spiral

October 5, 2026 at 15:45 EDT

Foundation Models
Sales & Business
Infra & Chips

Meta and Microsoft, two of Anthropic's biggest corporate customers, are steering employees toward their own AI tools as costs from Anthropic's Claude platform exceed the salaries of the staff using it. The trend underscores a central tension emerging across enterprise AI adoption: token-based pricing models that work for light usage become economically unsustainable when companies deploy models at scale.

October 2026 · The Information

Enterprise AI Budgets Collide With Token-Based Reality

Meta and Microsoft abandoned large-scale internal Claude spending as token costs exceeded employee salaries. The question now: can any enterprise sustainably pay per-token for AI at scale?

73.7T

Tokens consumed at Meta in 30 days

$3.4B

Microsoft spent in 4 months on Claude Code

50%

Drop in Meta's Claude Code users (60K → 30K)

$2,000

Per-engineer monthly Claude cost at Microsoft

Employee Shift: Meta's Internal Tool Migration

30K

Active Claude Code Users

→

Code Llama 4

Internal Alternative

The Core Problem: Unit Economics at Scale

Token-based pricing breaks down when enterprise teams use AI daily. Microsoft's math: 5,000 engineers × $500–$2,000/month = $2.5M–$10M monthly on Claude alone, or $30M–$120M annually. At that scale, building internal infrastructure and shifting to in-house models (Llama, Grok) becomes cheaper than paying per token, even accounting for development cost.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.