The median AI spend per employee per month is $11.38.
That's from Ramp's real spending data across 70,000+ US businesses in June 2026. One ChatGPT Teams seat. That's where most companies are sitting right now.
The top 1%? They're spending $7,449 per employee per month. A 680x gap — and it grew 14.1% last month alone.
- Median spend is $11.38/employee/month — essentially one SaaS subscription
- Top 1% spend $7,449/month — a 680x gap growing 14.1% MoM
- The divide is chatbot subscription vs. agentic infrastructure
- Agentic AI costs 30x more per workflow than chatbots ($1.20 vs $0.04)
- At this scale, AI spend becomes a competitive moat, not just a line item
Most companies are spending $11 a month on AI
Ramp tracks actual card and invoice spend across 70,000+ US businesses. The June 2026 AI Index shows where spending really lands — not where companies think they should be.
| Tier | Monthly spend per employee | What it typically covers |
|---|---|---|
| Median (50th percentile) | $11.38 | One ChatGPT Teams or Copilot seat |
| High (75th–90th percentile) | $50–$500 | Multiple SaaS AI tools + some API usage |
| Top 1% | $7,449 | Agentic workflows, internal AI infra, token budgets |
Two signals worth noting from the June data: Anthropic overtook OpenAI in enterprise spending share for the first time. And DeepSeek topped the trending tools list — meaning the highest spenders are actively mixing frontier and open-source models, not just paying for one premium subscription.
What is the top 1% actually doing with AI?
Three companies that hit the ceiling — and what triggered the cap.
Meta — "Claudeonomics" and the Token Legend
Meta built an internal AI usage leaderboard called "Claudeonomics," complete with a "Token Legend" badge for the highest consumers. The program was designed to drive adoption. It worked too well. In June, Meta was forced to impose hard usage limits after costs exceeded projections.
Uber — Annual budget gone in 4 months
Uber's engineering team hit 84% AI adoption. The annual AI budget was exhausted by April. The culprit was agentic coding assistants running autonomously on large codebases — hundreds of runs per day, each at $1.20 per workflow.
Mercor — More on AI than on people
Mercor, an AI recruiting platform, now spends more on AI tokens than on employee compensation. Their entire hiring pipeline runs on autonomous agents.
Why is the gap widening faster?
The structural reason is not adoption rate — it's cost architecture.
| Dimension | Chatbot (2023) | Agentic (2026) |
|---|---|---|
| Cost per workflow | $0.04 | $1.20 (30x) |
| Who triggers the action | Human | Agent (autonomous) |
| Daily run volume | 10–50 | Hundreds to thousands |
| Competitive visibility | Low (commodity tool) | High (proprietary infra) |
When an agentic system runs hundreds of times per day at $1.20 per run, the monthly token bill compounds fast. EY's framework puts it plainly: companies building agentic infrastructure now are building a cost moat. The gap is not about budget size — it's about which companies have already built the pipelines that justify the spend.
Which tier is your company in right now?
Five steps to locate where your AI spend actually lands — and what to do next.
- Calculate your current per-employee AI spend
Add all AI subscriptions and API costs for the month. Divide by headcount. If you do not know the number off the top of your head, that is already a signal. - Identify your highest-cost single workflow
Which one workflow or tool is driving the most spend? If it is a SaaS subscription, you are likely in the median tier. If it is API calls from an internal tool running autonomously, you are moving up. - Check whether you have any agentic workflows running autonomously
Agentic means the AI takes actions without a human approving each step. If you have zero agentic workflows, your $11 spend is probably appropriate for what you are doing. - Decide: SaaS tenant or infrastructure owner?
SaaS AI tools are fast to deploy and easy to manage. Infrastructure ownership — custom agents, model routing, token budgets — is expensive but builds proprietary capability. The top 1% has made this choice explicitly. - If scaling: build cost controls first
Meta and Uber both hit hard limits because they scaled without budget visibility. Before expanding agentic workflows, implement per-team token budgets and model routing so cheaper models handle simple tasks.
Want to go deeper?
Ramp AI Index — June 2026 The primary source: 70,000+ US businesses, actual card and invoice spending by AI tool and category. ramp.com
How Much Companies Spend on AI Tokens in 2026 Grey Journal breaks down the Ramp data with additional context on tier distribution and per-tool spend benchmarks. greyjournal.net
EY — Agentic AI Enterprise Token Cost Analysis of the cost structure shift from chatbot to agentic workflows, including the $0.04 to $1.20 cost-per-workflow framework. ey.com
Meta AI Token Spending The Claudeonomics program and what happened when an internal AI leaderboard drove adoption faster than budget controls could handle. cryptobriefing.com
Three Fault Lines Reshaping Enterprise AI in 2026 The Uber budget story and what agentic adoption looks like at scale — with analysis of the structural shifts driving enterprise AI spend. marketscale.com
