OpenAI's Sam Altman Admits AI Token Costs Are Now a 'Huge Issue' as Companies Blow Q1 Budgets
Key Takeaways
- ▸Token costs went from a non-issue to a 'huge issue' in just one year, with companies now complaining they've spent entire 2026 budgets by Q1
- ▸Token usage is growing exponentially—from 100K tokens/month (top user 6.5 years ago) to 100B+ tokens/month today, with per capita usage now at 100K monthly
- ▸The 'tokenmaxxing' strategy is backfiring: companies face diminishing returns and some find hiring humans more cost-effective than AI inference
Summary
OpenAI CEO Sam Altman disclosed at the Intelligence at Work event that enterprise clients are now expressing serious concerns about AI token costs—marking a significant shift from early 2026 when cost was never mentioned. Companies are reporting they've exhausted their entire annual AI budgets in just the first quarter, prompting OpenAI to accelerate work on model efficiency improvements.
The token cost crisis is being driven by companies engaging in "tokenmaxxing," betting that increased AI usage would boost productivity and revenue. Real-world examples include Amazon employees using AI agents unnecessarily to stay on internal leaderboards, Microsoft cutting Claude Code licenses due to expense, and individual developers racking up $1.3 million monthly bills. Even Uber's CEO admitted there's currently no proven link between AI spending and actual product success.
The underlying issue reflects exponential growth in token consumption outpacing efficiency gains. Token usage has grown dramatically—six and a half years ago, OpenAI's top customer used 100,000 tokens monthly; that figure is now the global per capita average, with the company's leader consuming 100 billion tokens monthly. However, costs are rising faster than efficiency improvements, and some companies are finding that hiring human workers is now cheaper than running AI models. This dynamic mirrors the Jevons paradox: as resources become cheaper, consumption rises exponentially, but with agentic AI becoming increasingly sophisticated, token usage is growing faster than price reductions can offset.
- OpenAI and other labs are racing to improve model efficiency, but productivity gains may not keep pace with exponential usage growth
- Jevons paradox in action: cheaper tokens drive more consumption, but agentic AI's token appetite is outpacing cost reductions
Editorial Opinion
The AI token crisis reveals a market correction was inevitable—companies experimented without proper cost controls, assuming productivity gains would justify any expense. While OpenAI remains profitable, the unsustainable unit economics suggest token prices must fall faster than they currently are, or enterprises will fundamentally rethink AI adoption strategies. This pressure could accelerate the push toward smaller, more efficient models and on-device inference, potentially reshaping the entire AI infrastructure market.
