Cracks in the AI Thesis Part 3: Ramp AI Index October 2026 Update
B · Useful, limited Monthly update from Ramp Economics Lab based on first-party card and bill-pay spend data; real transaction data, but limited to Ramp's US customer base, with revised and volatile top-tier estimates.
Summary based on the full text·Original: about 5 minutes to read
In short
Ara Kharazian's October 2026 update of the Ramp AI Index tracks which US businesses on Ramp pay for AI. Adoption has slowed to well under a point a month, which he reads as the top of the S-curve: more than half of businesses already pay. OpenAI narrowed Anthropic's lead. Among the heaviest spenders, AI spend per staff member is still below its July peak while token volumes rose, which he attributes to falling prices per token.
Key numbers
- 55.9%Ramp businesses paying for AI in September, up 0.62 points
- 43.8% vs 40.6%share of Ramp businesses paying Anthropic vs OpenAI
- $9,568monthly AI spend per staff member at the top 1% of firms in September, down 3.1% from July's $9,873
- 80%share of spend on OpenAI and Anthropic driven by 1% of firms
As reported by the source.
English·Web page·opens econlab.substack.com
What it reports
- 55.9% of Ramp businesses paid for AI in September (+0.62 points, after +0.40 in August).
- Anthropic 43.8% (+0.25 points), OpenAI 40.6% (+1.12 points). xAI rose 0.6 points; Mistral is under 1% of US firms.
- The top 1% of firms drive 80% of OpenAI and Anthropic spend. Their spend per staff member rose 11.7% in the month to $9,568, still 3.1% below July's $9,873.
- Two straight months of almost no growth in token spend; early October reached a new peak 1.5% above August.
- Kharazian's explanation: frontier price cuts and efficiency gains lower the blended price per million tokens.
- Enterprise contracts with the labs are described as not restrictive, with discounts from list price.
How they got it
Spend data from Ramp's own platform, aggregated across its customer businesses; adoption is the share of businesses paying a vendor in the month. The post does not restate the method in detail. Top-1% figures were revised up across all periods after new AI vendors were added.
Limits
- Ramp's customer base is not representative of all US firms and excludes non-Ramp spend.
- Top-1% estimates are volatile and subject to change, as the author notes.
- Contract observations are anecdotal ("what I've heard").
Why it matters for companies
Falling token prices mean budgets can stay flat while usage grows, so spend alone is a poor measure of AI use or value. Buyers appear to hold negotiating power with model vendors, which argues for short commitments and multi-vendor options. The concentration figure is a reminder that most firms are still light users; benchmark your own spend against peers of similar size, not the top tier.
Summary based on the full text. Read the original for the full argument and data.Read the original on econlab.substack.com