Highlights from our research in AI and software spending
2026Summer
BusinessSpending Report
Introduction · p.3
A note from our economist
Uses your device's voice where available; otherwise the highlight runs on its own at reading pace. Double-click any word to jump there.
What's inside
Recently, I've been thinking about mixed messages with respect to AI's impact on our economy.
If you're a business owner trying ChatGPT or Claude for the first time: it's nice to have, but is this really the technology that's going to vault the U.S. into a new paradigm of productivity growth?
If you're a college student: learn AI or get left behind. Also, if you join a company using AI, you're going to get laid off.
If you're a policymaker: tech companies will produce great wealth from this new technology, so much wealth that it might be a temporary bubble! But tech firms also need help in the form of deregulation and subsidies, even though they seem to be doing just fine without them?
These market participants all face the same issue: AI is a nascent technology for which there is little data to inform economic decision-making. Since I joined Ramp nearly two years ago, we've sought to bring more data to these conversations. Most recently, we published our first research paper on AI's impact on jobs and were featured in The New York Times, The Wall Street Journal, The Financial Times, and more.
But we've never been fans of one-off reports. This market is changing daily, so we update our outlook as new data comes in. Our main motivation is answering your questions about AI, software, and the state of business spend.
The research that follows is informed by questions from our customers. If you have questions about business spend, we might be able to answer them using Ramp data. You can reach me directly at ara.kharazian@ramp.com.
Economist, Ramp
p.4
About this report
The Ramp Business Spending Report is a quarterly analysis of corporate spending trends based on billions of aggregated, anonymized transactions from over 70,000 businesses using and corporate cards.
Our transaction set is built using company models that extract line-item text from paid receipts and bills uploaded by the purchaser following a sale. We use internal and external data, alongside proprietary Ramp models, to categorize businesses in size, segments, and sectors.
Small businesses represent companies with one to 24 employees. Medium-sized businesses represent companies with 25 to 99 employees. Large businesses include companies with 100 or more employees, including enterprise firms with thousands of employees. These and other report definitions are subject to change.
This report analyzes card and bill pay data observable by Ramp or data available from trusted third-party sources. Any conclusions should not be taken as an indication of a company's or Ramp’s business performance. Data points are not inclusive of all Ramp businesses.
p.5
Key findings
Anthropic surpassed OpenAI in adoption by U.S. businesses for the first time. Anthropic adoption surged by 21.8 percentage points from January to July.
Chart 02 → 0210.2%Total headcount rises after AI adoption. Companies that invest heavily in AI grew headcount 10.2% over the two years following adoption. Entry-level headcount grew even faster (12% over the same period). These gains are entirely driven by high-intensity adopters.
Charts 09–10 → 03$7,400Heavy adopters have significantly higher budgets per employee. The top 1% of AI adopters on Ramp spend more than $7,400 on AI per employee per month, and the top 10% spend $650. Meanwhile, the median adopter spends $11.95.
Chart 05 → 0436%Traditional SaaS vendors are increasingly charging for AI features. While seat-based billing still dominates for these embedded AI products, usage-billed products now account for 36% of embedded AI spend, up from 24% a year earlier.
Chart 13 →Before you go on
Choose your track
Both tracks keep every chart and every interactive. Quick trims the supporting paragraphs and adds a 30-second gist to each chapter. Deep keeps the full report text and the editor's deep dives.
Chapter one of four · pp.6–14
Ramp AI Index
The share of businesses using AI is rapidly approaching 100%. The focus now shifts to intensity of adoption.
In 30 seconds
- 55.7% of U.S. businesses paid for AI in July 2026, more than double the government's survey-based 21.6%.
- Anthropic overtook OpenAI in April and now leads 43.5% to 39.7%.
- Spend per employee is the new yardstick: $11.95 for the median adopter, $650 for the top 10%, $7,400.50 for the top 1%.
Since inception, the has focused on tracking simple business adoption: are firms using AI? But the share of businesses using AI is rapidly approaching 100%. Our focus is now shifting to tracking intensity of adoption. Our updated index now includes:
- 1Monthly AI spend per employee
- 2Granular filters like state, sector, business size, and funding background
- 3Business spend on subscriptions vs. coding agents vs. tokens and APIs
Few datasets track business AI adoption, and when they do, they’re based on surveys that are unreliable and often disconnected from actual business behavior. We don’t rely on surveys.
Our AI Index uses real transaction data: aggregated and anonymized card and bill pay spend from more than 70K U.S. businesses on Ramp.
Chart 01
Business AI adoption hits 55.7%, led by Anthropic gains
The share of U.S. businesses paying for AI services hit a record 55.7% in July, up about 5 from our last quarterly report. The spike was led by a surge in spending on Anthropic. Ramp’s estimate of AI adoption is more than double the U.S. government’s survey-based estimate (read why below).
Why is Ramp’s AI adoption estimate so much higher than the government’s?
The uses a survey to estimate AI adoption, asking businesses whether they use AI for any business function. Recent surveys by unaffiliated academics have estimated business adoption of AI lies between 40% and 85%. Ramp is the only data source that uses actual purchases of AI services to produce its estimate.
Deep dive · Stated versus revealed behavior
Economists distinguish what people say they do from what their spending shows. A survey asks a manager whether the business uses AI "for any business function." A ledger records whether someone actually paid for an AI subscription, API tokens or a coding agent. The two can disagree in both directions: a survey misses the employee who expenses a chat subscription without telling anyone, and a transaction set misses free tiers. Chart 01 is a reminder that the measurement method can move the answer by more than 30 points.
Editor's note, written for this edition.
Guess the adoption rate
Ramp dataWhat share of U.S. businesses do you think paid for AI services in July 2026?
Source: Ramp AI Index, business spend data from Ramp; U.S. Census Business Trends and Outlook Survey. Lines traced from the print chart; labelled end points are exact.
Chart 02
Anthropic surpasses OpenAI as the leader in business adoption
In April, Anthropic surpassed OpenAI in adoption by U.S. businesses for the first time. Anthropic adoption spiked 21.8 percentage points from January to July, and it now leads OpenAI by 3.8 percentage points. Over the same period, OpenAI adoption fell by 1.4 percentage points.
Overall includes businesses subscribed to any AI product or service based on Ramp spend data.
The model race
Ramp dataSource: Ramp AI Index, business spend data from Ramp. Lines traced from the print chart; labelled end points are exact.
Chart 03
Middle-of-the-pack sectors are closing the AI adoption gap
Every sector Ramp tracks rose since last quarter’s report, but middle tiers gained the fastest since our last report. Manufacturing led adoption gains, up eight percentage-points, followed by construction and healthcare, both up about seven points. Finance and insurance was up over five points.
Chart 04
Nearly half of small businesses now adopt AI
The larger a business is, the more likely it is to adopt AI tools and services. Still, nearly half of small businesses (49.5%) now pay for AI.
Find your peers
Ramp dataSource: Ramp AI Index, business spend data from Ramp. Lines traced from the print charts; July 2026 values are exact.
Charts 05 – 07
AI-pilled firms have significantly higher budgets per employee
The top 1% of AI adopters on Ramp spend $7,400.50 on AI per employee per month, and the top 10% spend $650. Meanwhile, the median adopter spends $11.95.
Chart 06
VC-backed companies spend the most per head on AI
Venture capital-backed businesses spend over 10x more on AI per employee than firms. Venture-backed companies lead in AI adoption (with an 80% adoption rate) as VC firms influence portfolio companies to use the latest tools.
Chart 07
Smaller companies spend more per employee on AI
Small companies spend $22.79 per month per employee on AI, versus $9.28 for medium companies and just $3.39 for larger companies.
AI spend includes LLM subscriptions, coding agent subscriptions, API tokens, and GPU cloud and infrastructure spend.
Deep dive · Why do small firms spend more per head?
Per-employee figures divide a bill by a headcount. A five-person startup that buys a few coding-agent seats and an API plan spreads that across very few people; a 2,000-person company may buy a site license whose cost per head is tiny. The ranking in Chart 07 says as much about the denominator as about appetite. It flips the adoption story in Chart 04, where larger businesses are more likely to pay for AI at all.
Editor's note, written for this edition.
Benchmark your spend per head
Ramp dataSource: Ramp AI Index, business spend data from Ramp. Lines traced from the print charts; July 2026 values are exact. The median line in Chart 05 is too close to zero to trace and is shown at its July value.
Chart 08
50 states of AI: Delaware leads in AI-adopting businesses
The state famous for LLCs should be famous for LLMs. At 73% adoption, the corporate capital of the world has nearly a 10 percentage-point lead over AI hubs like California and New York.
Ramp AI Index: State adoption
Ramp dataTap a tile or choose a state.
Share of U.S. businesses with paid subscriptions to AI models, platforms, and tools. Latest month: July 2026.
Chapter two of four · pp.15–17
A New Look at AI’s Impact on Jobs
These firms adopted AI. These similar firms didn't. Did jobs go up or down?
In 30 seconds
- Ramp linked AI spend to Revelio Labs workforce records for 21,559 U.S. firms.
- Heavy AI adopters grew headcount 10.2% over two years; entry-level headcount grew 12%.
- Low-intensity adopters show no statistically significant change.
The most important economic question of this decade is how AI will affect jobs. And yet there wasn’t a good dataset to answer it.
There's been some work, using surveys that ask firms who they're hiring or firing. But further research has shown people aren't really good at answering those surveys, and reasons for firings tend to be mixed. Other research tracks jobs based on AI “exposure” scores, but economists disagree on the validity of those scores, too.
Ideally, we'd have a very simple dataset. These firms adopted AI. These similar firms didn't. Did jobs go up or down? So we built it.
This summer, we released our first research paper: A New Look at AI's Impact on Jobs.
We used AI spending from Ramp card and bill pay data, linked to workforce records for 21,559 firms in the United States. Here’s what we found:
- 1Firms that heavily adopt AI grow headcount 10.2% over the two years following adoption, and entry-level headcount at these firms grew even faster (12%).
- 2AI adoption and the associated gains are unevenly distributed. AI adopters are already larger, more engineering-intensive, more likely to be venture-backed, and faster-growing than non-adopters. These firms then grow faster upon adoption.
The following charts highlight our findings.
How the study works
Illustrative- Spend
- Link
- Match
- Track
- Compare
Chart 09
Total headcount rises after AI adoption
Companies that invest heavily in AI grow headcount 10.2% over the two years following adoption. These gains are entirely driven by . Low-intensity adopters see no statistically significant change.
Chart 10
Entry-level headcount rises after AI adoption
Entry-level hiring rose 12% over the two years following adoption at the companies making the largest AI investments. At the end of the 24-month analysis period, firms in the high-intensity group also increased their workforce share of entry-level workers (up 1.15 percentage points compared to the ).
Deep dive · How to read an event-study chart
The horizontal axis counts months relative to the month a firm started paying for AI (month 0). Everything is measured against similar firms that did not adopt, so a flat line at zero means "no different from the control group." The shaded band is the confidence interval: where it still covers zero, the effect could be noise. Before month 0 the lines hover around zero, which is what you want to see. It suggests adopters weren't already pulling away just before they adopted.
The vertical axis is labelled "Effect on log headcount," as in the print. The headline percentages (10.2% and 12%) are the paper's estimates; see the source for how the plotted scale maps to them.
Editor's note, written for this edition.
Predict, then explore
Paper dataTwo years after heavy AI adoption, what happens to a firm's headcount, compared with similar firms?
Source: Kharazian, A., Simon, L., & Stevens, R. (2026). A New Look at AI’s Impact on Jobs: Firm-Level AI Spending and Workforce Adjustment. Ramp Economics Lab. ramp.com/data/ai-jobs-impact. Lines traced from the print charts; bands approximate.
Chapter three of four · pp.18–21
Ramp Rate
The leading and fastest-growing vendors by software category.
In 30 seconds
- Spend on AI models is rising fast, and so is spend on category-specific vendors.
- AI-native vendors are gaining ground: Profound up 25%, Granola up 23%, PolyAI #1 on the trending list.
- Incumbents like Figma, GitHub and Salesforce still capture the lion's share, and are starting to charge for AI.
Software purchasing decisions are leading indicators of broader industry shifts.
Software stocks have struggled this year over worries that AI agents will replace many enterprise SaaS uses. But while business spending on AI models is rising fast, spending on category-specific vendors is also rising. Businesses are boosting category-specific and seat-based spend even as model spending surges.
Our Ramp Rate directory tracks market share, business adoption, and growth trends across software vendor categories, highlighting the fastest-growing and most adopted vendors, backed by transaction data across 70,000+ businesses on Ramp.
One of the biggest recent trends: AI-native and AI-adjacent vendors are gaining ground in new categories. For example, adoption of vendor Profound is up 25% year-over-year. And AI meeting notes app Granola is up 23%. Meanwhile, AI support bots are emerging as one of the clearest enterprise AI use cases. PolyAI is #1 on the trending list, and Sierra is #4. This is a rare category where AI’s effect can be directly measured (e.g. resolved CX complaints, average time to close a ticket).
Incumbents continue to capture the lion's share of total spend in their categories, suggesting that the is not now.
But despite the rise of AI-native vendors, incumbents like Figma, GitHub, and Salesforce continue to capture the lion's share of total spend in their categories, suggesting that the SaaSpocalypse is not now.
Meanwhile, several traditional SaaS vendors (including the three listed above) are beginning to charge for their AI-powered features. Many initially bundled these tools into existing subscriptions at no extra charge while they were still in beta. Now that the testing phase is ending, AI is becoming a paid add-on.
These features tend to be billed either as or . The cost difference between billing models is significant. If usage-based billing takes off, a sea change in SaaS spending could follow.
Chart 11
Traditional SaaS vendors are increasingly charging for AI features
Fourteen percent of traditional SaaS vendors billed for an AI-powered feature in the past year, up from 8% a year earlier. These features, such as chat and agent subscriptions, can be billed as seat or tier-based or as consumption-based (metered API usage).
Chart 12
Seat-based billing still dominates for embedded AI products
Of traditional SaaS vendors billing for AI, the share of vendors who exclusively bill on a seat-based model far exceeds the share of usage-only and model vendors. 9% of traditional SaaS vendors use a seat-only model, versus 3% that bill based on usage-only and 2% with dual-billing models.
Who's charging for AI
Ramp dataSource: Ramp Business spend data. Monthly values traced from the print charts.
Chart 13
The cost difference between usage and seat-based models is significant
Usage-billed products now account for 36% of embedded AI spend, up from 24% a year earlier, even as they remain a minority among traditional vendors. The median business billed based on usage spends 10.6x more annually on than one paying only for seat-based products.
Seat or usage?
Ramp dataSource: Ramp Business spend data. Bar heights traced from the print chart; first and last bars match the stated 24% and 36%.
Chapter four of four · pp.22–26
Watercooler Data
Miscellaneous data of interest from our Leading Indicators newsletter.
In 30 seconds
- Starbucks is 45.2% of trip restaurant transactions among the top 10, about 6x Dunkin.
- All top five flight pairs rose by double digits; SEA–SFO is up 41%.
- World Cup host-city hotel rates peaked at +60% in May; New Yorkers still order hot coffee most months.
Chart 14
Corporate America flies on Starbucks
Starbucks is by far the most-expensed restaurant on business trips booked on Ramp. Over the past year, the java giant had more transactions than the next three chains combined. It sees about 6x more travel transactions than coffee rival Dunkin.
Guess the most-expensed chain
Ramp dataAmong restaurants on business trips booked on Ramp, which chain gets the most transactions?
Source: Ramp. Share of cleared Ramp card restaurant transactions linked to trips, July 1, 2025–June 30, 2026. Percentages show each restaurant’s share of transactions across the top 10; excludes food delivery, catering, and supermarkets.
Chart 15
Flight prices spike for the top corporate travel routes
Catching flights to catch meetings isn’t what it used to be. Prices for the top five city pairings booked on Ramp all saw double-digit increases in Q2 compared to a year earlier. San Francisco (SFO) pairs saw the biggest surges, led by the Seattle (SEA) and SFO pairing, up 41%. The median price for the 2-hour-ish flight was $637. The LAX-SFO pairing wasn’t far behind, up 36% to $374.
Chart 16
The FIFA premium on corporate stays
Businesses weren’t booking World Cup trips. They were booking regular business travel to corporate hubs like NYC and Atlanta that happened to be hosting the tournament. Ramp Travel data shows companies paid a premium for employee stays in host cities, even if their team offsite or client visit had nothing to do with soccer (sadly, usually the case).
Trip budget check
Ramp data| Roundtrip route | YoY change | Median Q2 ’26 |
|---|
| Month | 2025 median | 2026 median | YoY lift |
|---|
Source: Ramp. Chart 15: median USD roundtrip ticket totals, including taxes and fees, for bookings created April–June 2026 versus Q2 2025. Excludes cancelled tickets and Ramp's internal business. Routes are undirected airport pairs, ranked by Q2 2026 bookings. Chart 16: median booked nightly rates calculated from Ramp Travel hotel bookings in World Cup host cities by booking month.
Chart 17
Summer is not enough to break NYC’s hot coffee habit
New Yorkers like it hot (always). Ramp transaction data shows that in NYC, hot coffee purchases dominate iced orders for most of the year. Even in the sweltering heat of summer in the city, cold brews barely have a lead.
Source: Transaction data from 50K+ businesses on Ramp’s spend platform. Monthly values traced from the print chart.
Checkpoint
Five questions on the report
Quiz
3 multiple choice · 2 true/falseBack matter
Sources and data notes
About this edition
Body copy, chart titles, chart data, source lines and About text are from the Summer 2026 Business Spending Report. Gists, editor's notes, deep dives, glossary entries, widget prompts, quiz explanations and sample discussion threads are written for this interactive edition.
Charts 01–07, 09–13 and 17 are redrawn from the print. Labelled values (end points, table figures, map values) are exact; the monthly paths between them are traced from the print charts and are approximate. Chart 08 and Charts 14–16 use the printed figures directly.
Interactive results (guesses, benchmarks, trip estimates) are calculated in your browser and stay there.
p.27
About Ramp
Ramp is how companies save time and money on every dollar they spend. It's the smart financial infrastructure behind every card swipe, invoice, and reimbursement – streamlining approvals, processing payments, and closing the books automatically. More than 70,000 organizations, from family farms and space startups to the Fortune 100, have saved over $12 billion and 27 million hours with Ramp. For the median customer, that translates to 5% savings on expenses and 16% revenue growth in their first year. Founded in 2019, Ramp powers over $200 billion in purchases annually.
Credits
- Data
- Ara Kharazian, Cara McCabe, Ian Macomber
- Editorial
- Rebecca Moretti
- Creative
- Kyla Arsadjaja
- Comms
- Lindsay McKinley, Nikki Neumann, Nathan Allebach
- Marketing
- Danielle Poublon, Grace Cummins, Alexa Jacques, Mehrisa Malik
- Legal
- Thomas Hopkins
Interactive edition of the Ramp Business Spending Report, Summer 2026. Single self-contained file; no tracking.