Microsoft reports AI is more expensive than paying human employees
3 hours ago
- #AI costs
- #Productivity paradox
- #Enterprise adoption
- Firms are pressuring employees to maximize AI use for productivity gains, but this is causing unsustainable costs and operational issues.
- Microsoft canceled most Claude Code licenses after six months due to high internal usage, shifting engineers to GitHub Copilot CLI, though its partnership with Anthropic remains unaffected.
- Uber exceeded its 2026 AI coding tools budget in just four months after incentivizing adoption via internal leaderboards, highlighting rapid cost overruns.
- The cost of AI adoption, particularly compute expenses, is becoming a bottleneck, challenging early forecasts about replacing or augmenting human labor with AI.
- Companies like Meta and Amazon are encouraging high AI token usage (e.g., 'Claudeonomics' leaderboard, 'toxenmaxx'), but token-based pricing raises costs as efficiency improves.
- Goldman Sachs predicts a 24-fold increase in token consumption by 2030 due to AI agents, potentially reaching 120 quadrillion tokens monthly, driving up aggregate costs.
- While individual AI token costs are expected to drop sharply by 2030 (e.g., 90% less for inference on advanced models), increased consumption and agentic models may keep enterprise AI expenses high.
- Gartner warns that cheaper tokens don't equate to cheaper AI overall, as factors like higher token usage per task and incomplete cost pass-through to consumers will likely push inference costs higher.
- Nvidia's CEO envisions an agentic future with AI workers alongside employees, but if token consumption outpaces falling unit costs, it could lead to unexpectedly heavy financial burdens for firms.