peekgenticBook a call

News analysis · AI costs

AI prices are falling and the top spenders have slowed. What it means for your next contract

New Ramp data shows companies running far more AI work while their bills grow far less. Its economist calls the deals customer-friendly. Here's what that changes for a mid-size company buying AI, what it doesn't, and what to ask your vendors.

By Miguel GutierrezCo-founder, Peekgentic · Dallas
· 4 min read
Preview card: AI prices are falling. What it means for your next contract.
Companies ran far more AI work in October, while their bills grew far less.Peekgentic

In brief

You probably have more bargaining power than you think. New Ramp data shows AI usage up 37% in a month while spending rose only 9%, because the price per unit of AI work fell about 27%. That's a good reason to ask your vendors whether those savings reach you, to keep contracts short and to use cheaper models for simple steps. It's not a reason to wait, and it may not last. It's one company's data, and most of it comes from heavy tech spenders.

  • Ramp shows token use up 37.1% in a month, spending up only 8.9% and blended price per token down 26.8%.
  • Ramp's economist says customers hold the better position in enterprise deals. That's his read from conversations, not a survey.
  • The median firm paying for AI spends $15.70 per employee per month, so you aren't behind if you don't spend like the top 1%.
  • Lower token prices reach you only if your vendor passes them on. Ask, keep terms short and test cheaper models on simple steps.
  • It's one company's data, skewed toward heavy tech spenders, and prices could move back the other way.

What the numbers say

On Nov. 6, TechCrunch reported on new data from Ramp, a company that runs corporate cards and bill pay. Spending on tokens, the units AI models bill by, grew only 1.5% across August and September. In July it had grown 42% in a single month.

Usage didn't stall. Prices fell. In the Ramp AI Index, the week ending Oct. 23 compares with the same week a month earlier like this:

One month of AI tokens: more work, slower bills

Week ending Oct. 23, 2026 vs. week ending Sept. 25. Businesses using Ramp's token tracking product, a sample that Ramp says isn't representative of all its AI spend.

  1. Tokens usedup 37.1%
  2. Dollars spentup 8.9%
  3. Price per token (OpenAI and Anthropic blended)down 26.8%

Source: Ramp AI Index, data through October 2026.

A blended price also moves when customers shift to cheaper models, so that isn't the same as a 27% price cut. But real cuts are happening. TechCrunch reports OpenAI halved API prices on two GPT-6 models in September, and that Anthropic made big cuts too, which it said in a filing took about $600 million off its revenue growth.

Ramp's lead economist, Ara Kharazian, summed it up to TechCrunch: "You can use more AI, but use it for less." He wrote in October that, from what he's heard, customers hold the better position in enterprise deals, and that the labs will discount from the sticker price. That's his read from conversations, not a survey.

You aren't behind if you aren't spending like a tech company

Headlines about firms spending $10,000 per employee can make an owner feel late. The same data says otherwise.

Monthly AI spend per employee, among firms paying for AI

GroupAugust 2026September 2026
Median firm$14.74$15.70
Top 10% of firms$740.47$813.43
Top 1% of firms$8,568.17$9,567.53

Source: Ramp AI Index. Ramp says the top 1% figure is more volatile than the others.

Kharazian wrote that 1% of firms drive 80% of spending on OpenAI and Anthropic. Even that group's September number is 3.1% below its July high. And he told TechCrunch it isn't clear that a doctor's office or a car dealership would ever need that much. We'd add a machine shop or a distributor to his list. About 61% of manufacturers on Ramp paid for some AI in September, so most peers there have started. That tells you they pay for something, not how much.

Don't measure yourself by what the top 1% spends. Measure what each order or invoice costs you.

What doesn't change

Cheaper tokens don't fix a bad process. If nobody has decided who reviews exceptions, a lower bill doesn't help. Our view is that for most office projects, the model bill is only one line. Setup, review time and keeping things running often matter more, so a price drop alone rarely decides whether a project makes sense.

The payoff isn't automatic either. In its November report, the Bank of England says UK firms estimate AI has raised their productivity only about 0.5% over the past three years. The same firms expect a much bigger lift, about 0.9 percentage points a year, over the next three. That's the UK and it's firms' own estimates, but it fits what Ramp's numbers suggest: buying AI is the easy part.

Where the other side has a point

Ramp has a stake and a skew. It sells spend management, and it only sees its own customers. Kharazian himself says nearly all AI spending comes from venture-backed, high-growth tech firms. Mid-size manufacturers may see something different.

The slowdown may already be easing. Spending rose 8.9% in the latest month, and Ramp says its top 1% estimate moves around a lot. Two soft months aren't a trend you should bet a budget on.

Cheap may not last. Ride-hail fares rose after the early discounts ended. Kharazian argues AI differs because there are too many competitors, including cheap and open models. He may be right, but that's a forecast. And local votes against data centers could tighten supply over the years ahead, which would push the other way. Both can be true: cheaper this year, less certain later.

Your price may not move at all. This is our reading. Many tools charge per seat or per document, so a drop in token prices doesn't reach you unless your vendor passes it on. And metered billing is growing, so a lower price per token can still come with a higher bill if use climbs. Companies like ours that build on these models benefit from lower prices too, which is one more reason to ask how it reaches you.

What to do this month

  1. Ask each AI vendor one question. "Model prices have dropped. What does that change on my bill?" A good vendor will have a real answer.
  2. Keep terms short. Prices are moving too fast for a long prepaid deal. If you do sign for longer, ask for a clause that lets you reset the price, and ask what happens if prices fall.
  3. Ask if it can swap models. A step like "is this a new PO or a change order?" may not need the priciest model. Test a cheaper one on your own documents first, since models can lose track in long documents.
  4. Set a budget and a limit. Kharazian says experienced buyers set defaults and budgets to keep costs in check. Ask whether your tool can cap use, and turn that on before someone needs it.
  5. Track cost per item. Orders, invoices or emails handled, against what you spend. Here's how to measure it.
  6. Don't wait for prices to bottom out. If a project works at today's price, it works better later. If it doesn't, cheaper tokens probably won't save it.

One more thing to watch. TechCrunch says Bloomberg reports that Anthropic's potential valuation fell from about $2 trillion to $1.2 to $1.3 trillion, and that OpenAI lowered its proposed valuation too. These are big companies and this isn't a crisis. But it's a good time to make sure you can export your data and settings from any AI tool you rely on.

Common questions

Are AI prices actually falling?

Yes, by Ramp's measure. Its blended OpenAI and Anthropic price per token fell 26.8% between the weeks ending Sept. 25 and Oct. 23, 2026. A blended price also moves when customers shift to cheaper models. TechCrunch also reports real price cuts from both companies.

Why is AI spending slowing if companies use more AI?

Mostly because each unit of work costs less. Ramp shows token use up 37.1% while spending rose 8.9% in the latest month. Its economist also points to competition and buyers getting better at controlling use.

How much do companies spend on AI per employee?

In Ramp's September 2026 data, among firms paying for AI, the median firm spent $15.70 per employee per month. The top 10% spent $813.43, and the top 1% spent $9,567.53.

Will my AI bill go down?

Not by itself. Many tools charge per seat or per document, so lower model prices reach you only if your vendor passes them on. Ask directly. If you pay by usage, the bill can still rise if use climbs.

Should we wait for AI to get cheaper?

Probably not. If a project pays off at today's price, waiting costs you those savings. If it doesn't, the model bill is rarely the only reason. Keep contract terms short so you can benefit if prices drop.

Will AI stay this cheap?

Nobody knows. Ramp's economist argues there's too much competition for prices to climb the way ride-hail fares did. Limits on new data centers could push the other way. Plan for prices to move in either direction.

Sources

  1. TechCrunch, Ramp data shows AI spending stalling just as Anthropic nears its IPO, Nov. 6, 2026
  2. Ramp AI Index, data through September 2026, token data through Oct. 23, 2026, checked Nov. 7, 2026
  3. Ramp Economics Lab, Cracks in the AI Thesis Part 3, Oct. 14, 2026
  4. Bank of England, Monetary Policy Report, November 2026

More from Peekgentic

Published November 7, 2026 · Written by Miguel Gutierrez, Co-Founder, Peekgentic (Dallas, TX). Every statistic links its source.