OpenAI vs Anthropic Market Share: Read the Ramp Data
The OpenAI vs Anthropic market share figure everyone is quoting this week comes from Ramp, and as reported on 20 August 2026 it reads: Anthropic held nearly 44% of business AI spend in July against OpenAI's nearly 40%, and OpenAI is now growing faster.
The OpenAI vs Anthropic market share figure everyone is quoting this week comes from Ramp, and as reported on 20 August 2026 it reads: Anthropic held nearly 44% of business AI spend in July against OpenAI's nearly 40%, and OpenAI is now growing faster. In May the split was 41% to 39%. So Anthropic is still ahead and the gap is closing.
That is a genuinely interesting datapoint. It is also not market share, and the difference matters if you are about to pick a vendor on the strength of it.
What this OpenAI vs Anthropic market share number measures
Ramp is a corporate card and expense management company. Its index reads card and bill-pay transactions across more than 70,000 American businesses spending billions through the platform, and counts what share of those businesses are paying which AI vendor. TechCrunch's write-up has the reporting.
So the measurement is: among companies that use Ramp as their spend tool, how many have a paid subscription with each vendor. That is a real signal. It is not the same thing as market share, and Ramp does not claim it is.
Four biases you have to carry
Bias | Direction it pushes the number |
|---|---|
Ramp skews toward tech companies and startups | over-weights vendors that developers and technical teams pick first |
Large enterprises often pay by other means | under-counts the segment where multi-year committed contracts live |
It counts subscriptions, not dollars | a company paying for five seats counts the same as one paying for five thousand |
It counts payers, not usage | a lapsed pilot that nobody cancelled looks identical to a system of record |
Stack those together and the honest description is not market share. It is the share of a tech-leaning American SMB population that has a paid relationship with each vendor, weighted equally per company. Useful, narrow, and not the number a procurement deck should quote as if it were global.
The one number that survives all four biases
Buried under the head-to-head is a figure that the biases do not really distort: the share of Ramp customers paying for any AI service at all rose from 50% in March to nearly 56% by July.
That works because it is a within-panel change over time on a fixed population. Whatever Ramp's sample is unrepresentative of, it is unrepresentative of it in March and in July equally, so the delta is the reliable part. Six points of adoption in four months, in a population that was already half penetrated, is the real story in this release.
What it should and should not change about your choice
Vendor share tells you almost nothing about which model fits your workload. What it tells you about is the survivability of your integration, and that is a narrower question than it sounds.
Do not switch on share data. Two vendors within four points of each other, measured with this much noise, is a tie. The evaluation that matters is your own, on your own tasks, which is the argument in how to tell if a new AI model release is actually a big deal.
Do care about the trend line for lock-in reasons. A vendor losing share fast is a vendor whose pricing and roadmap may change under you. What that exposure costs and how to bound it is in AI app builder vendor lock-in.
Write prompts that survive a vendor change anyway. The cheapest insurance against any of this is not picking correctly, it is not being welded to the pick. There is a method in how to write prompts that work across AI models.
Watch your own unit costs rather than anyone's share. Vendor share is not your margin. Your token bill is, and it is more tractable than it looks: see how to reduce AI API costs.
How to read the next one of these
Spend-panel indices arrive roughly monthly now, and they get quoted as market share almost every time. Three questions strip most of the noise out:
Who is in the panel, and who structurally is not? Every panel excludes someone. Ask who.
Is the unit companies, seats, or dollars? Per-company counting flattens a hyperscaler and a four-person startup into one vote each.
Is the claim a level or a change? Levels inherit every sampling bias. Changes within a fixed panel mostly do not.
Apply those three to this release and you get: a tech-leaning American SMB panel, counted per company, reporting a narrow and closing gap. Which is worth knowing, and is not a market share figure.
Frequently asked questions
Who has more business users, OpenAI or Anthropic?
In Ramp's July 2026 panel, Anthropic, at nearly 44% against nearly 40%. That covers Ramp's more than 70,000 American business customers rather than the whole market, and Ramp reports shares rather than dollar amounts.
Is Ramp's AI index reliable?
It is reliable about what it measures, which is paid subscriptions among its own customers. It skews toward tech and startups, misses enterprises paying through other channels, and counts each company once regardless of size.
Is OpenAI overtaking Anthropic?
Not yet. OpenAI reversed earlier losses and is growing faster in Q3 2026, but Anthropic still led in the most recent reading. A four-point gap in a panel of this kind is close to a tie.
Should vendor market share decide which AI I build on?
No. Use it as a weak signal about vendor stability, and decide on evaluation results against your own tasks, your cost per request, and how portable your integration is.
Both vendors ship fast enough that this ranking will move again before the year ends, which is an argument for a reading routine rather than a one-off check: how to keep up with AI news covers one.
How did this land?
About the author
Staff Engineer, Platform
Carlo works on the platform that turns prompts into running apps. He writes the engineering deep dives and the changelog notes worth reading.
