§ Claim under review · Business
"80% of OpenAI and Anthropic's enterprise revenue reportedly comes from just 1% of customers, according to data from business expense tracking company Ramp based on spending observed through its platform"
Verdict
Source exists but framing is misleading
Confidence
MediumSummary
This number is real and correctly attributed, but it does not mean what most people will think it means. Ramp's lead economist did publish, on September 2, 2026, that 80% of OpenAI's and Anthropic's enterprise revenue comes from 1% of customers. The catch is that Ramp only sees money spent by US businesses that use Ramp's own corporate cards and bill pay, a panel of roughly 70,000 companies. So "1% of customers" really means the top 1% of that panel, not 1% of everyone who pays OpenAI or Anthropic, and the figure is a spending proxy rather than either company's actual revenue. Neither OpenAI nor Anthropic publishes audited financials, and neither has confirmed or denied the number. The Instagram post deserves some credit for saying the figure is not an official revenue disclosure, which much of the coverage dropped. What remains unverified is whether this concentration is genuinely true of the companies' real books, and I could not find a published Ramp report showing the methodology behind this specific chart.
The readings
key figures from the evidenceshare of OpenAI/Anthropic enterprise revenue from top 1% of customers
share of customers generating 80% of revenue, per Ramp
Why this verdict
Evidence
The claim is a faithful relay of a real, findable statement. Kharazian posted on September 2, 2026 that Ramp data shows "80% of OpenAI and Anthropic's enterprise revenues come from 1% of their customers, and it's not getting better," describing it as "a level of concentration risk unseen in any other software category we track," and noting that the top 1% skews heavily toward the tech sector and AI products and services. He tied the observation to the approach of IPOs for both companies.
Ramp's own methodology explainer describes its dataset as aggregated, anonymized business spend from US companies that use Ramp, covering both corporate card and invoice-based payments, with invoices making up the majority of payment volume. Recent AI Index editions describe a panel of more than 70,000 US businesses. TechCrunch, reporting on an earlier Ramp AI Index release, stated plainly that because the index only represents companies that use Ramp, it is not a perfect proxy for the wider market.
An X Community Note appended to a widely reposted version of the statistic states that it is a proxy based on Ramp's platform transactions and not verified figures from OpenAI or Anthropic.
I found no statement from OpenAI or Anthropic confirming, denying, or commenting on the figure. Neither company publishes audited revenue or customer-concentration disclosures.
Findings
✓ What's accurate 5
- Ramp is a business expense and spend management company, and it does run a research arm that publishes recurring analysis of AI spending, the Ramp AI Index.
- The 80% and 1% figures are real and accurately quoted. Kharazian's own wording is essentially identical to the claim.
- The figure is derived from spending observed on Ramp's platform, exactly as the claim states.
- The claim's hedges are appropriate and present: "reportedly," attribution to Ramp by name, and the caption's explicit statement that this is not an official revenue disclosure from either company. That is more caveating than most coverage of this statistic carried.
- Kharazian did characterize the concentration as unusually high relative to other software categories Ramp tracks, and did say it is not improving.
≈ What's misleading 4
- **Subgroup generalization:** the claim says "1% of customers," which a reasonable reader takes to mean 1% of OpenAI's and Anthropic's customers. Ramp can only see businesses that use Ramp. The measured group is the top 1% of the roughly 70,000 US companies in Ramp's panel that pay these labs. OpenAI's and Anthropic's actual largest revenue sources, including consumer subscriptions, cloud-partner-mediated API consumption, government and very large enterprise contracts, and all non-US customers, are largely or entirely outside Ramp's view. A concentration statistic computed on a panel that structurally excludes the biggest accounts is not the same statistic as one computed on the real customer list, and it can run in either direction.
- **Omitted qualifier:** the phrase "enterprise revenue" is doing heavy lifting. Ramp is measuring dollars flowing through its platform, which is a spend proxy, not either company's revenue under any accounting definition. The post's closing caveat partly discloses this, but the headline sentence still equates the two.
- **Marketing as evidence:** Ramp is not a neutral statistical agency. The AI Index is a research and visibility product for a fintech company, and this particular number was released as a framed narrative about "the latest threat to the AI trade" ahead of two anticipated IPOs. That does not make it wrong, but it means the figure is an interested party's self-published proxy with no external audit and no chart-level methodology I could retrieve.
- **Rumor as fact, in a mild form:** downstream coverage has already begun dropping the qualifier. PYMNTS headlines the figure as "80% of Revenue" rather than enterprise revenue, and financial-press items present it as a settled fact about IPO risk. The Instagram post is not the worst offender here, but it sits in that chain.
? What's uncertain 6
- Whether the figure appears in a published Ramp report with a methodology note. I could not locate a September 2026 Ramp AI Index or Econ Lab post containing this chart. As of today it traces to social posts by the analyst.
- The definitions behind the number: the time window, whether "customers" means distinct legal entities or accounts, whether subscription and API token spend were pooled, and whether the 80/1 figure is a monthly snapshot or a trailing average.
- Whether the same concentration holds in the companies' actual books. This is unknowable from public sources: neither company files audited financials, and neither has commented.
- Whether the claimed comparison, that this exceeds concentration in every other software category Ramp tracks, holds up. No comparison table was published that I could retrieve.
- The related assertion appearing in downstream coverage, that Cursor and GitHub Copilot together account for close to a quarter of Anthropic's revenue, is a separate claim from separate reporting and was not investigated here.
- The post's image shows event footage with what appears to be conference branding, not a Ramp chart. Nothing in the image supports or contradicts the number.
Sources
8 of 8 linked to recordsAra Kharazian (Ramp lead economist), X post, September 2, 2026, 5:37 PM
Ramp, "How Ramp data works: an explainer"
Ramp AI Index, August 2026 edition
X Community Note attached to Kalshi's repost of the statistic
PYMNTS, "OpenAI and Anthropic Get 80% of Revenue From 1% of Customers," September 4, 2026
Seeking Alpha news item, September 3, 2026
TechCrunch, "Anthropic now has more business customers than OpenAI, according to Ramp data," May 13, 2026