TokenBudgeting: Our Conversations with Enterprises on Token Spend
Crystal Huang
· SemiAnalysis
· June 30, 2026 at 18:32
· ⏱ 12 min read
| Read on Substack ↗
Summary
Enterprise AI token spending is not collapsing but normalizing into budgets, with continued growth driven by coding use cases and new verticals like cybersecurity and knowledge work. The article argues that headline tokenmaxxing stories from Meta and Uber are overblown and that AI lab API revenue continues to grow robustly, supported by strong demand from top percentile customers and expanding TaaS providers.
•Meta employees consumed over 60 trillion tokens in a 30-day period, with the top user burning ~280 billion tokens, before the company shut down the tracking dashboard.
•Uber imposed a $1,500/month/employee limit on AI tokens after burning through its Claude Code and Codex annual budget in four months.
•Median Ramp customer spends only $136 per employee per year on AI, while the 99th percentile spends ~$90,000 per employee, highlighting extreme skew in AI spend.
•70%+ of ARR at OpenAI and Anthropic is attributed to coding use cases, with Anthropic's B2B mix (90%+) higher than OpenAI's (60%).
•Enterprise budgets for AI tools range from $250/month (aerospace & defense) to $2,000/month (Workday, Stripe), with most employees not approaching their limits.
The article states 'Our estimates for AWS Bedrock this quarter drive our total AWS growth rate number well above street' — directly implies stronger-than-expected AWS revenue from enterprise AI token
The article states 'Our estimates for AWS Bedrock this quarter drive our total AWS growth rate number well above street' — directly implies stronger-than-expected AWS revenue from enterprise AI token consumption, which is a bullish indicator for Amazon's cloud segment.
Risk: Enterprise budget caps could eventually slow token growth if coders hit limits, but author sees continued momentum.