Enterprise & Work

AI token prices fell 21% in six weeks as usage hits record high

Tokens got 21.2% cheaper in six weeks while volume hit a record, Ramp data shows; Stripe sees AI revenue growth slow from 140% to 80% annualized.

AI industry growth remains strong, but the pace is slowing
AI industry growth remains strong, but the pace is slowing652234 / pixabay
By James Calloway5 min read

Why it matters

  • Tokens got 21.2 percent cheaper in six weeks while token volume rose 1.5 percent to a record, per the Ramp AI Index covering 85,000+ US companies.
  • Anthropic held 58.7 percent of all token spending in the week ending November 15, 2026, versus OpenAI's 37.2 percent.
  • Stripe AI Index: AI company revenue grew 103 percent year-over-year, but 13-week annualized growth fell to about 80 percent from 140 percent.
  • Average price per million tokens for OpenAI and Anthropic combined fell to $0.45, down 11.2 percent month over month.
  • 56.8 percent of tracked US companies paid for AI in October 2026, up 13.3 percentage points year over year.

Tokens — the units of computation AI providers bill on — got 21.2 percent cheaper in six weeks, according to the latest update of the Ramp AI Index, even as token volume rose 1.5 percent to a record high. US companies are using AI models more intensively than ever, and paying less for the privilege. That gap between usage and revenue is the story the index tells, and it keeps widening.

The Ramp AI Index draws on transaction data from more than 85,000 US companies. Ramp lead economist Ara Kharazian published the numbers, and they cut in opposite directions: volumes up, prices down, spending off its peak. Token spending — what companies actually pay for AI APIs — fell 6.3 percent from its peak two weeks earlier.

What do the month-over-month numbers show?

Comparing the week ending November 15 with the week ending October 18, total token spending still rose 12 percent. Token volume, however, jumped 27.2 percent over the same span. Companies are consuming more than twice as fast as their bills grow.

The average price per million tokens for OpenAI and Anthropic combined fell to $0.45, down 11.2 percent from the prior month. OpenAI is the cheaper of the two at $0.38 per million tokens, against $0.50 for Anthropic. Takeaway: deflation in the AI API market is now measured in weeks, not quarters.

Does OpenRouter's data confirm the trend?

It does, with caveats. An analysis by Peter Walker, published on X, tracks traffic through OpenRouter, an API routing platform. Weekly token volume for closed-weight models from OpenAI, Anthropic and Google grew 1,215 percent since the start of 2026. Weekly spending rose 455 percent over the same period — far more slowly.

The spending peak came in the week of October 5, according to the OpenRouter data, which runs through the week of November 9. Spending has declined since then. Both sources — Ramp, which imports usage and billing data directly from AI providers, and OpenRouter, which measures traffic through its own routing layer — show the same shape: an October peak, falling outlays, rising volume. Neither is representative of the overall market, and OpenRouter in particular offers only a narrow window into a very specific user base. Takeaway: two independent datasets pointing the same way is meaningful, but neither covers the whole economy.

Who is winning the token market?

In the week ending November 15, 2026, Anthropic accounted for 58.7 percent of all token spending, up 1.7 percentage points from the prior month, according to the Ramp AI Index. Its share of token volume was even higher at 59.4 percent. OpenAI fell to 37.2 percent of spending and 37.7 percent of volume. Google came in at 1.4 percent, and xAI (listed as SpaceXAI in the index) at 2.4 percent.

Kharazian noted on X that Anthropic holds this share on IPO day — a pointed marker for a company heading to public markets. The growth picture is less lopsided, though: in October, Anthropic's token spending grew only 0.5 percent month over month, while OpenAI's grew 2.1 percent. A dominant share with near-flat growth is a weaker position than the headline number suggests.

Adoption keeps climbing regardless. In October 2026, 56.8 percent of all tracked US companies paid for AI products or services, up one percentage point from September and 13.3 percentage points more than a year earlier. Anthropic leads adoption at 44.8 percent; OpenAI sits at 40.6 percent. Takeaway: more businesses are paying for AI every month, but they are spreading spend unevenly across a shrinking price base.

What does Stripe's supply-side index say?

The Ramp AI Index tracks demand — what companies spend on AI. Payment processor Stripe has published its own AI Index measuring the supply side: incoming payments at roughly 20,000 AI companies that use Stripe.

Revenue at AI companies grew 103 percent over the 52 weeks ending November 8. The typical (median) AI company grew only 41 percent, which suggests a few large providers are skewing the overall picture. Aggregate AI growth flatters the sector; the median company does not.

The slowdown over shorter time frames is pronounced. Annualized over 13 weeks, growth stood at about 80 percent, down from 140 percent just a few months ago. The US leads with 104 percent year-over-year growth, while the EU sits at 67 percent. AI companies are still growing far faster than traditional software firms — 103 percent compared with 33 percent in Stripe's SaaS Index. Takeaway: fast by software standards, but decelerating by its own.

Why does the slowdown matter?

Both indexes paint the same picture from different ends of the market. At Ramp, cooling shows up through falling prices and concentrated demand. At Stripe, it shows up through slowing growth rates on the provider side. AI is still growing fast in absolute terms; the momentum behind that growth is what is fading.

The stakes are financial. The AI industry depends on extraordinarily strong growth if it wants to recoup the massive capacity buildout it has planned — the capital already committed to data centers and compute. A widening gap between token consumption and token revenue makes that payback harder, not easier.

New growth could come from more capable or more reliable models, or simply from better integration of current models into business workflows. The potential is real: AI has upended the entire coding market in a matter of months. Whether that kind of demand shock repeats elsewhere is the open question both indexes cannot answer.

Original: ramp.com

Share this article:

More from James Calloway

James Calloway

Show full bio

News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.

791 articles

Related articles

  1. OpenAI's Enterprise Report: 1 Million Business Customers, 320x API Growth
  2. US Corporate AI Spending Hits Record in October as Growth Slows to 1.65 Percent
  3. Apple Cuts Engineer AI Token Budgets From $300 to $200 Per Day
  4. 4.5% of US AI Users Pay — and the Top 1% Spend $900 a Month
  5. Microsoft ships seven in-house AI models that undercut OpenAI on price

« Previous articleNext article »