Ramp AI Index: Business AI Use Rises While Spending Falls

Ramp AI Index: Business AI Use Rises While Spending Falls

US companies are running more AI workloads than ever, and they are paying less for them. That is the main finding of the latest Ramp AI Index, compiled by Ramp economist Ara Kharazian.

The two lines are moving in opposite directions. Spending peaked in July and has fallen since. Usage has climbed about 50 percent over the same period and reached a record high at the end of September.

Why the bill is shrinking

Kharazian identifies two causes:

  • Price cuts on top models. The most capable models have become cheaper to call.
  • Better budget options. Standard and lite models have become cheaper and more efficient, so more work can move to lower-cost tiers.

He says the decline comes almost entirely from competition between two companies: OpenAI and Anthropic. As they undercut each other, business customers pay less per unit of work and use more of it.

This is not the first time spending in the index has dropped. What stands out this time is how long the decline has lasted. A brief dip could be noise. A slide that has run from July to the end of September looks more like a change in pricing.

Two vendors dominate the spend

The data also shows how concentrated the market is. In the last week of September covered by the index, Anthropic took 51 percent of token spending. OpenAI took 44.5 percent. Together, the two account for more than 95 percent of the tracked spend.

Open-source models remain a small part of that picture. According to Kharazian, they still make up less than five percent of business spending. This holds even though capable open-weight models that run on local hardware keep getting released.

What the index measures, and what it misses

The numbers come with caveats, and they matter for how far the findings should be pushed.

  1. Sample size. The Ramp AI Index is based on transaction data from more than 70,000 US companies.
  2. Token data. The token figures, including the vendor split, come from a subsample of that group, not from the full set.
  3. API only. The analysis covers API spending. Other ways companies pay for AI fall outside its scope.
  4. Skew toward big buyers. Kharazian notes the data leans toward large AI customers.

So the index is best read as a signal about how larger US companies buy AI through APIs. It does not capture the whole market. Smaller firms and spending outside the API may follow a different pattern.

What This Means

For businesses building on AI, the index points to a buyer's market at the API level. Falling prices alongside rising usage suggest companies are not simply saving money. They are spending the savings on more AI work. That fits a pattern seen across recent releases, where vendors pitch newer models such as OpenAI's GPT-6.1 Sol on lower cost as much as on raw capability.

The concentration figure deserves attention too. If price competition is driven by just two vendors, it depends on both continuing to compete hard. That pressure could change as their finances come under closer scrutiny, with Anthropic's leaked IPO filing one example. It is worth watching whether spending bottoms out in the coming months, whether the vendor split shifts, and whether open-source models ever move beyond their small share of business budgets.