Token bills to push most enterprise AI workloads onto open-weight models
Summary
McKinsey said at Tech Week Singapore that most enterprise AI workloads will shift to open-weight models rather than frontier models from OpenAI and Anthropic, as agentic systems drive up token consumption. Per-token prices have fallen about 90% since 2023, but agentic models need five to 30 times more tokens per task, and 93% of enterprises overspent their AI budgets in the past six months.
Bubble analysis
This touches one of the core mechanisms of the bubble debate: whether AI spending can be covered by the returns it generates. Widespread enterprise AI budget overruns and some firms cutting back AI use because of operating costs show real cost pressure, but the article gives no hard capex or revenue figures, so it is only indirect evidence.
Related signals
linked by shared tags
- AI Adoption Is Everywhere, But Transformation Is Not#enterprise-ai
- Generative AI Market Forecasts Growth from $185.45B (2026) to $1.65T by 2033, Profiling AWS, NVIDIA, OpenAI, Anthropic & Microsoft#enterprise-ai
- Zscaler promotes Sanjay Yadave to Asia Pacific & Japan#enterprise-ai
- Meta, MongoDB shares fall after Meta launches enterprise AI unit#enterprise-ai
- Meta launches enterprise AI platform, hires MongoDB CEO to lead new initiative#enterprise-ai