The top 1% of American businesses now spend $7,450 per employee per month on AI. The median firm spends $11.38. That’s a 680-fold gap inside the same economy, and it’s the headline number from Ramp’s June 2026 AI Index, built by the company’s Economics Lab from anonymized card and bill-pay data across more than 70,000 U.S. businesses.

The gap matters because adoption has stopped being the interesting question. Ramp’s panel shows roughly 100% of firms now using AI in some form. What separates the cohorts is intensity, and intensity is compounding: per-employee spend in the AI-pilled tier grew 14.1% last month alone.

Ramp CEO Eric Glyman, talking to CNBC, tied that intensity to outcomes. Businesses spending the most of their revenue on AI grew revenue 12%. The least-spending tier saw flat growth. Glyman called the current phase “the twilight moment of tokenmaxxing,” a read that more tokens no longer cleanly proxy for productivity, and that taste in deployment is becoming the edge.

The vibes around the spending number are doing real work, too. An unnamed Nvidia executive told TechCrunch that compute now costs his company more than his employees. The CEO of Mercor said the startup spends more on internal-agent tokens than on headcount. Stack that against a typical American software engineer earning roughly $16,000 per month, and the $7,500 figure is genuinely large but still not yet the dominant input cost it’s being narrated as.

There’s also a measurement gap underneath the spending gap. Census Bureau data flagged in an April Federal Reserve note put U.S. firm AI adoption at about 18% at the end of 2025, with the largest gains in younger, smaller companies. Ramp’s near-universal adoption number reflects its customer base, which skews toward firms that already buy software with a card.

Which is why vendor positioning is shifting. Single-seat enterprise subscriptions to ChatGPT and Claude anchor the top of the index. Below them, Anthropic and OpenAI are competing for mid-market intensity against newer entrants like LemonLime, pitched as a model-agnostic company brain for teams that can’t justify per-seat enterprise contracts. Compressing time-to-value for the firms stranded near the median is the actual prize.

The 680x gap isn’t a story about who adopted AI. It’s a story about who’s allowed to keep compounding.

Sources