← Back to Blog
AI & Automationby ELHASSANE Studio

Enterprise AI Spending Is Slowing. Is Demand Cooling—or Is the Market Maturing?

Enterprise AI Spending Is Slowing. Is Demand Cooling—or Is the Market Maturing?
ELHASSANE Studio

The paradox on every AI dashboard right now

Here is a strange pair of numbers.

In September 2026, 55.9% of American businesses sampled by Ramp were paying for AI tools — up 12.1 percentage points year over year, and still climbing month by month. Meanwhile, the growth of what those businesses spend on AI tokens slowed so sharply that TechCrunch ran the headline “AI spending is stalling.”

Both things are true at the same time: the spread of AI adoption and the growth of AI-related spending do not necessarily move at the same pace.

If that sounds contradictory, it isn’t. It is what happens when the price of something falls faster than demand for it rises. And it may be the most important chart for small businesses deciding how to invest in AI this year.

What the Ramp AI Index actually measures

The Ramp AI Index is not a survey. Ramp is a corporate spending platform — corporate cards, bill pay, and expense management — serving more than 85,000 American businesses, and instead of asking firms whether they “use AI,” it counts the actual payments that pass through its systems: subscriptions, API bills, and token charges.

This matters because surveys miscount fast-moving adoption, and Ramp’s methodology note is refreshingly honest about the reverse risk too: because free tools and personal accounts show no corporate payments, the index may understate the true spread of AI. So 55.9% is not a ceiling — it is a careful measurement of paid adoption, likely below the real usage picture.

A sign the data world takes it seriously: the St. Louis Fed’s FRED database added Ramp’s AI adoption figures in October 2026, making the series available alongside other economic datasets.

One note on terms before the table: “AI adoption” is Ramp’s term for businesses that paid for an AI product during the month — free use does not count. The September 2026 numbers by segment:

Segment AI adoption, businesses paying for AI (Sep 2026)
All businesses 55.9%
Large businesses 65.7%
Medium-sized businesses 62.5%
Small businesses 50.0%

Adoption is broad. It includes the unglamorous parts of the economy: 49.9% of retailers, 43.2% of construction firms, 33% of restaurants and accommodation businesses now pay for something AI. The real version extends well past software offices into shops, job sites, and kitchens.

Spending vs. usage: two different curves

Now to the part the headlines got right — and slightly wrong.

TechCrunch, reporting from Ramp’s index, described token spending growth stalling: roughly +1.5% from August to September, after a +42% surge in July. That is a genuine slowdown in the growth of dollars. “Stalling” is a fair word.

But look at what happened next in Ramp’s own published window (the week ending October 23 compared with the same week a month earlier), and the story flips:

  • Total token spending rose 8.9%
  • Total token volume rose 37.1%

Read those two numbers together — carefully, because they come from different measurement windows and datasets (a monthly spend snapshot versus a weekly token window over a smaller tracked sample), so they are not a single continuous spending trend. What they do show: businesses in the tracked sample processed 37.1% more tokens than the month before — and paid only 8.9% more for them. Volume is sprinting while spend walks. (One caveat: the token metrics come only from Ramp customers using its AI token cost management product, so they are directional rather than statistically representative.)

The difference between those curves is not demand collapsing. It is the price of intelligence falling.

The price war nobody is calling a price war

One important clue lies in the changing cost of tokens: blended across OpenAI and Anthropic tracked customers, the price per million tokens fell 26.8% between the two weekly windows — from $0.76 to $0.56. (That measures what Ramp’s tracked customers actually paid, not a formal list-price scale.)

Break it down and the competitive dynamic appears:

  • OpenAI’s blended price fell 42.4% (from $0.80 to $0.46 per million tokens)
  • Anthropic’s fell 15.7% (from $0.73 to $0.61)

Neither company calls this a war, and the Ramp data shows the price movement but not a single cause: competition, efficiency improvements, and pricing changes may all be contributing to lower costs per token.

Economics 101 follows naturally. When the price of a good falls sharply:

  1. Buyers use more of it (volume up 37%)
  2. Revenue per unit falls (blended price down 27%)
  3. Total buyer spending can grow slowly even while consumption grows fast (+8.9%)

This is why “AI spending stalled” and “AI usage exploded” are not contradictory claims. Falling unit prices are an important — quite possibly decisive — part of the story, though the two readings are not one series: dollar growth measured monthly across firms, versus token flow across a smaller weekly sample. The buyer’s wallet and the API endpoint can tell apparently different stories for a while.

Anthropic, OpenAI, and the IPO problem

The second layer is what this means for the providers themselves — and it cuts the other way.

For buyers, falling prices are great news. For AI providers and their investors, the same trend is pressure on revenue and margins. Anthropic is reportedly approaching a possible initial public offering, and both AI lab valuations have been revised downward in recent weeks: Bloomberg reported Anthropic’s expected valuation trimmed from around $2 trillion to the $1.2–1.3 trillion range, while the Financial Times reported OpenAI’s cut from $1.4 trillion to about $1.25 trillion. (These are reported figures from named outlets, not officially confirmed numbers.)

But — and here is the part the panic headlines missed — the token share data suggests the competition is intensifying in a way that should be uncomfortable mainly for whoever assumed the market would tip to one player:

  • Anthropic’s share of tracked token spending among Ramp-based businesses jumped from 51.3% to 60.5% in one month
  • OpenAI’s fell from 44.2% to 35.1%

One month of data on one platform’s customers is not destiny — and note that this measures share of spending, not share of usage volume. Adoption share is far more balanced — 43.8% of businesses pay for Anthropic products and 40.6% for OpenAI’s within the same window (a business can pay both, so these shares need not sum to 100). But if token volume is growing fast and prices are falling the way they are, the real contest is over unit economics at scale, not this month’s leaderboard.

What small businesses should learn from all this

If you run a small business, none of this is sector gossip — it is market intelligence about the raw material you are likely buying this year. Four lessons:

1. AI is becoming more financially accessible to small businesses. Ramp’s adoption table has small businesses at 50% and climbing, and the median AI-paying firm spends just $15.70 per employee per month. For most firms that is not a capital decision; it is closer to a lunch. For many small businesses, the question is increasingly not just what AI costs, but which task is worth automating first.

2. Don’t budget by the headlines, budget by output. Over the measured four-week window, the same dollar could buy roughly a third more tokens, based on the observed blended price decline — so any AI business case built earlier deserves a re-cost, downward.

3. Concentration risk is real, but two-sided. Anthropic’s share of tracked token spending is rising fast, and month-to-month shares can swing wildly (9 points in one window, on spending — not usage). For a service that becomes infrastructure in your operations, maintain a second provider relationship. Where practical, test model portability — switching costs can rise when a model becomes deeply integrated into your workflows.

4. Measure value per dollar, not dollars. Ramp’s own data shows how extreme the spending spread is: the median AI-paying firm spends $15.70 per employee per month, the top 10% a median of $813.43, and the top 1% a median of $9,567.53. What the data does not tell us is return on investment — there are no output figures for any of these cohorts. High spending does not automatically mean high value. The only meaningful question is whether the money spent on AI produces measurable business results: hours saved, errors reduced, revenue moved.

What’s next: cooling, or maturing?

Here is the honest answer to the question in the headline: we don’t know yet — and that is the most instructive part.

Two interpretive hypotheses compete, and Ramp’s data settles neither. The bear reading says enterprises may be reining in auto-renewing AI subscriptions and that price cuts may be masking a demand ceiling, while the provider-side valuation trims reportedly reflect investors seeing the same curves as TechCrunch. The maturity reading says this may be what infrastructure markets look like when they work: adoption broadens, unit costs collapse, spending grows slowly, and value moves to those who apply the technology rather than those who merely resell it. As a loose historical analogy — not a measured parallel — electricity and cloud computing each passed through periods where spending growth disappointed while actual usage kept accelerating.

Upcoming updates — and Anthropic’s IPO story — may help clarify the picture. December’s index should offer a hint on whether October’s +8.9% was a blip or the new normal: moderate spend growth on top of fast volume growth.

Conclusion: judge AI by what it produces, not what it costs

Six months ago the question hanging over business AI was “are companies spending enough?” Now the data has moved on to subtler questions: what are businesses getting per dollar, and where does that dollar rank beside alternatives?

The useful takeaway is unchanged by the answer. If you are a small business owner or a solo creator, the latest Ramp snapshot suggests that, within its tracked sample, businesses are processing more tokens at lower per-token prices. The opportunity is real, but the value still depends on what you do with that capacity.

The next phase of the AI market will not be defined by how much businesses spend on technology. It will be defined by how much measurable value they create with it.


Data and sources: Ramp AI Index — official page (monthly data through September 2026; token metrics through the week ending October 23, 2026), and FRED, St. Louis Fed, which began hosting the dataset in October 2026. Reporting context: TechCrunch, November 6, 2026; the valuation figures attributed to Bloomberg and the Financial Times are cited via that report, and we could not verify those outlets’ underlying articles directly. Ramp’s own caveat applies to all token figures: they are drawn from a subset of customers using its AI token spend management product, are directional rather than representative, and its methodology may underestimate true adoption because unpaid use is not captured.