Ramp, the corporate expense management company, announced on November 11, 2026 that AI-related spending rose 1.65% month over month in October, setting a new record. In the department-level data published for the first time, customer support and customer success accounted for 13.3% of AI spending for which the department could be identified. That is up sharply from 0.8% in June, suggesting AI spending is spreading beyond engineering. Meanwhile, the average price per million tokens fell 21% over the past month.
To analyze corporate AI spending, it is not enough to look at the growing number of companies paying for AI services. You also need to distinguish which departments use AI and how much they pay. The paid adoption rate is the share of companies that paid for AI, while the departmental breakdown shows how classifiable spending is divided. Even if AI use spreads beyond engineering, revenue for model providers does not necessarily rise. Whether growth in usage outpaces falling unit prices will determine future growth.
80% of AI spending is concentrated in the top 1% of companies
According to Ramp's November 2026 AI Index monthly report, the top 1% of companies by spending account for 80% of spending on OpenAI and Anthropic among the companies observed. These companies are mainly venture capital (VC)-backed tech companies and firms in finance and professional and technical services. Total AI spending continues to grow, but the structure in which a small number of companies bear most of it has not changed.
There is also a large gap in monthly AI spending per employee. The median across all companies paying for AI is $16.32, but it reaches $800.84 for the top 10% of companies by spending and $10,199.20 for the top 1%. These figures compare the median per-employee spending within each group; they are not total company spending divided by headcount. Also, the top 10% and top 1% groupings are based on AI spending, not on company size by employee count or revenue.
Based on the definitions in Ramp's official dashboard, the main October metrics are summarized below.
| Metric | October value | Denominator / scope |
|---|---|---|
| Paid AI adoption rate | 56.8% | Share of Ramp's companies that paid for an AI product or service during the month |
| Spending share by department | Engineering 72%, customer support/customer success 13.3% | AI spending for which the department could be identified |
| Monthly AI spending per employee | $16.32 for all paying companies, $800.84 for the top 10% by spending, $10,199.20 for the top 1% | Median per-employee spending within each group of companies paying for AI |
Even if the paid adoption rate rises, spending is not necessarily spread across a broad range of companies. While more companies pay for AI, a small number of large customers may continue to account for most of total spending. For model providers, maintaining spending from existing large customers matters as much as winning new ones.
Customer support's share of spending rises to 13.3%
In October's department-level AI spending, engineering accounted for 72%, still the largest share. Meanwhile, the share for customer support and customer success rose from 0.8% in June to 13.3% in October. The new departmental breakdown makes it possible to see in more detail than before which internal departments are increasing or decreasing AI spending.
According to Ara Kharazian of Ramp, engineering AI spending has declined from the high levels reached in the summer, while growth in customer support spending is helping sustain overall growth. Executive-level AI spending also showed a modest increase. However, the growth in customer support has not yet matched the rapid growth engineering recorded in early 2026.
Kharazian cites expanded use of customer service chatbots as a reason for the rising AI spending in customer support. However, the data shows only which department incurred the spending. It does not reveal business outcomes, such as how many inquiries AI handled or how much response times were shortened.
In addition, Ramp's monthly report does not disclose the number of companies included in the departmental tally, the share of total AI spending for which the department could be identified, or the dollar amount for each department. The rise in customer support's share to 13.3% may reflect not only increased spending in that department but also decreased spending in engineering. The increase in share therefore cannot be treated as the growth rate of spending or of inquiries handled. Judging whether customer support will become a lasting source of AI demand requires continued monitoring of actual usage, spending, and the effect on operational efficiency.
Why spending is slow to grow even as usage increases
Ramp reports that AI token usage grew faster than spending over the past month. Tokens are the unit used to count the amount of text and other information an AI processes. Even if usage grows, a decline in the average price per token holds down the growth in what companies pay. In October, token-related spending rose 2.1% month over month for OpenAI and 0.5% for Anthropic.
Based on the 21% decline in average price that Ramp reported, we can calculate the usage needed to keep spending constant. Because spending equals average price times usage, a 21% price drop means that, at the same usage, spending falls to 79% of its previous level. To maintain spending, usage must therefore rise to about 1.266 times its previous level. The required increase in usage is calculated as follows.
\[ \left(\frac{1}{1-0.21}-1\right)\times100 \approx 26.6\% \]
In other words, if the average unit price falls 21%, usage must rise by about 26.6% to keep spending the same. However, this is a theoretical figure that assumes tabulation conditions and the mix of models used remain unchanged; it is not the observed growth rate of token usage. The average price Ramp tallied may also reflect changes in the types of models used and their mix. A cut in individual model prices must be distinguished from a decline in the average price companies actually pay.
Meanwhile, the decline in average unit prices is happening alongside rising spending on high-performance frontier models. According to Ramp's report, the spending share of the frontier model group, comprising Opus, Fable, Sol, and Astra, rose from 67.7% previously to 69.5% in October. Kharazian points out that fast, low-priced models may be better suited to customer support work, but across total spending, concentration in the frontier model group is advancing.
This trend means that even as AI use cases expand, the growth in spending varies with the types and prices of the models used. A rise in the use of relatively cheap models alone does not necessarily boost model providers' revenue significantly. Whether adoption outside engineering becomes a new revenue source depends on how far growth in usage can offset the decline in average unit prices.
After adoption, department-level usage and spending are what to watch
The paid AI adoption rate in October was 56.8%, up 1 percentage point from the previous month. By payee, Anthropic rose 1.1 points to 44.8%, OpenAI was flat at 40.6%, and SpaceXAI rose 2.2 points to 6.3%. However, because the same company may pay for multiple AI services, these figures do not represent each company's exclusive market share.
The paid adoption rate is tallied from more than 85,000 U.S. companies that use Ramp for corporate cards, invoice payments, and other services. Ramp explains that because it is hard to capture use of free services and business use through personal accounts, it may be underestimating actual AI use and spending. On the other hand, The Decoder's report notes that Ramp's customers tend to be growth-oriented and above the U.S. corporate average in technological sophistication and company size, so the paid adoption rate may be skewed high. In other words, both underestimation from missed payments and overestimation from a skewed customer base need to be considered.
Furthermore, the data analyzing token usage and average unit prices is based on a different sample from the paid adoption rate. According to the definitions on Ramp's official dashboard, the token analysis covers usage and billing data from AI providers connected by companies that use Token Spend Management. It is therefore not representative of AI spending across all Ramp customers. Note that the companies covered by the paid adoption rate, the token analysis sample, and the spending for which departments could be identified are each separate tallies.
The data shows that AI spending is concentrated among a small number of large customers, while paid use is beginning to spread to departments beyond engineering. However, whether customer support becomes a continuing revenue source cannot be judged from spending share alone. Going forward, it will be important to compare department-level usage, actual spending, and business outcomes over the same period to determine how much economic value expanded AI adoption is producing. For model providers as well, a key benchmark will be whether growth in usage outpaces falling unit prices and leads to sustained revenue growth.