The data behind tomorrow’s business headlines. Powered by transaction data from 70k+ companies on @tryramp

Joined June 2019
Would you pay extra for a kid-free flight? JetBlue’s first ones just sold out. More in today's Leading Indicators: • The AI earnings scaries • What stood out in Anthropic's S-1 • A record number of companies switching AI providers Link in replies.
2
1
9
380
Anthropic (@AnthropicAI) made its IPO filing public today. We went back through the Ramp data we published this year. Here’s Claude’s 2026 via Ramp numbers 🧵
2
1
1
19
4,652
The race kept moving after the adoption crossover. By September, 8% of businesses were changing their top AI model provider by token spend, a record three-month average in Ramp's series.
1
2
144
For token buyers, there was another shift. Ramp's blended effective price per million Anthropic tokens fell from about $1.55 in January to about $0.80 in October. The figure combines input, output, and cached tokens across models.
2
123
SF company cards are last for bars but first for restaurants (18.5% of payments, Jan–Sep 2026)
1
6
361
The companies spending the most on AI aren't spending it on AI tools. For the top 1% of businesses, 88% of AI spend goes to compute and models. Apps, wrappers, everything else: 12%.
Interpret this how you will, but more than a third of AI spend by the top 1% of firms is already going to compute
12
494
Leading Indicators retweeted
Everyone has been asking me why Ramp is building a router We started by helping companies manage spend. As more of their money goes to buying intelligence, our job is to help 85k customers decide between models, too Excited about the progress so far, but still a lot of work ahead
How do model routers compare to individual models on real agentic tasks? Introducing the Vals Router Index. Routers promise to balance quality and cost across a fast-moving model frontier. We put that promise to the test with an independent evaluation of five routers (OpenRouter Auto, FireRouter, Ramp Router, Jev, Fugu Max) against every single model on the Vals Index, using the same agents, ~1,250 tasks and graders across finance, coding, law and tax. We found that routing didn’t consistently deliver better value. Every router was beaten on both score and cost by at least one single model.
1
4
66
3,909
More than half of enterprise project-management buyers purchased Jira in the past 12 months. No other tool reached 30%. The top 5, based on Ramp spend data 👇
5
578
Leading Indicators retweeted
Join me and @nathanbenaich who put together one of the best reports in the industry. NYC for the State of AI meetup
nyc 👋 i'm hosting @stateofai meetup on 10/27 i'll share the report's key findings across research, industry, politics, safety and my predictions then, a fireside w/the one and only @arakharazian, lead economist at @tryramp - the best person to talk ai adoption come hang!
3
2
29
3,944
The big worry for the AI industry in the short term is that customers have started aggressively bargain-hunting. They're switching to cheaper models and switching between models more often. As a result, spending on AI is plateauing even as everyone uses more AI.
Talk of an AI bust is back. AI is earning a lot more money than it did last year, but data centers are being built with a lot more debt. noahpinion.blog/p/ai-vs-the-…
16
21
4
166
30,033
The leading AI model providers by observed spend across six industries, based on the September 2026 Ramp AI Index.
4
22
929
“It’s not that businesses are using AI less — in fact, they’re using it more and more.” They’re “beginning to bargain-hunt.” @Noahpinion uses our AI Index data from @arakharazian to ask who captures the value as AI gets cheaper. Read: noahpinion.blog/p/ai-vs-the-…
1
10
551
Delaware had the highest share of Ramp businesses paying for an AI product or service in September: 72.7%.
3
1
14
850
The most-purchased AI vendor in each category, based on Ramp spend data. Search, dictation and coding have clear leaders, with Perplexity, Wispr Flow and Claude Code each bought by 87%+ of category buyers. Inference, GPU capacity and AI-first GTM are still fragmented.
9
21
17
146
87,382