Tokenmaxxing Headlines Overstate Reality: Enterprise AI Spending Remains Modest, Says SemiAnalysis Report
Key Takeaways
- ▸Tokenmaxxing was concentrated among a handful of high-profile tech companies; median enterprise spending per employee is just $136-$2,000/year, far below headlines suggest
- ▸Enterprise token budgets now range from $250 to tens of thousands monthly, with companies implementing per-employee limits and requiring approval for overages
- ▸The 99th percentile of enterprise customers spend ~$90,000/year per employee while the 90th percentile spends ~$7,300, revealing a highly skewed distribution driven by coding and VC-backed AI companies
Summary
SemiAnalysis conducted on-the-ground research with over 50 enterprise customers to understand token spending trends following the widely-reported "tokenmaxxing" craze of early 2026. While Meta, Uber, and Salesforce made headlines with aggressive token consumption and subsequent budget cuts, the actual distribution of spending reveals a much different story: most enterprises are nowhere near the high spend levels reported in the press, and median enterprise spending on AI models per employee remains under $150. The research shows that token budgets—ranging from $250 to tens of thousands monthly—have become the new norm, with companies downgrading default models and implementing usage limits. Most significantly, the data suggests that despite recent tokenmaxxing episodes, the enterprise AI adoption curve still has substantial runway ahead, with median Fortune 500 spending well under $100 per employee annually.
- Despite token budgeting shifts, the enterprise AI adoption s-curve remains early with significant growth potential, particularly as non-tech-forward enterprises begin AI deployment
Editorial Opinion
The media's focus on token-burning spectacles at Meta and Uber has obscured a more important trend: most enterprises haven't even begun their AI spending journey. SemiAnalysis's data reveals that even high-spending Fortune 500 companies remain well below $2,000 per employee annually in AI costs, suggesting the industry is in the first inning of adoption. Token budgeting is simply enterprises' natural response to managing a new cost center—not a sign of market saturation.
