On May 13, 2026, Ramp's AI Index reported that 34.4% of US businesses on its platform paid Anthropic in April, against 32.3% paying OpenAI. It is the first month Claude has been in front in the index's history. Ramp's Lead Economist Ara Kharazian, who wrote the post, attached three reasons the lead may not survive Q2 to the same paragraph.

Anthropic's number quadrupled in twelve months. OpenAI's grew 0.3% over the same window. "Anthropic's adoption has quadrupled over the past year while OpenAI rose just 0.3%," Kharazian wrote. The 2.1-point lead is small. The trajectory under it is not.

What Ramp is actually measuring

The index counts a business as an "AI adopter" for a given month if it has a positive transaction amount on its Ramp corporate card or bill-pay product against that vendor in that month. The sample is roughly 50,000 US companies. It is share-of-companies-paying, not share-of-spend, and Ramp's own methodology page is explicit that the panel skews toward US-headquartered, tech-forward, venture-backed firms.

That is the read-the-fine-print part of the headline. Ramp is not saying how much US enterprises are paying Anthropic versus OpenAI. It is saying how many of them have at least one Claude or ChatGPT line item on a corporate card. For early-stage SaaS, that is a leading indicator. For Fortune 500 IT spend, it is a vibe.

The bias cuts in Anthropic's favour. OfficeChai's coverage of the same report noted that among VC-backed companies, Anthropic adoption is 66% and OpenAI is 59% — a much wider gap than the headline 34.4%/32.3%. Strip out the Ramp-typical customer and the lead probably narrows or inverts. That is the asterisk Ramp's economist did not put in his own headline.

The three reasons it could come back

Kharazian lists them in the report and on the index data page. They are worth quoting because the company that just won the metric is the one issuing the warning.

First, cost alignment. Anthropic's business is metered: heavier token usage on Claude generates more revenue. Kharazian's read is that this creates a quiet incentive to recommend expensive solutions when a smaller model would do. The Loop has been watching the same dynamic since the Opus 4.7 launch, which tripled the per-token cost for image-bearing prompts. A company paying Claude Opus rates for a workflow that fits in Haiku is one re-architecture away from being an OpenRouter customer.

Second, service quality. Recent weeks have brought outages, rate limits, and complaints from heavy users; Kharazian noted Anthropic addressed the immediate issues but flagged the recurrence pattern. The line he chose: "A company actively turning away revenue because it lacks the compute to serve it is an unusual problem to have." The Claude subscriber side of this story has been the SpaceX Colossus 1 lease — Anthropic renting all of Musk's xAI cluster because its own compute pipeline is months behind demand. The metric Ramp picked up is the rate-limit pain leaking into procurement decisions.

Third, the OpenRouter problem. Cheaper third-party inference platforms — OpenRouter, Together, Fireworks, Lambda — are increasingly how businesses pay for open-weight models. Those line items don't appear as "Anthropic" or "OpenAI" on the corporate card; they appear as the inference provider. As open-weight quality closes the gap with the frontier labs, the metric Anthropic just won could start measuring a smaller share of the actual AI spend. DeepSeek V4 going open-weight at a million-token context was the first hard signal here. The next quarter's index will tell whether it converts.

What changed in twelve months

The historical comparison is what makes the story unmistakable. Ramp's March 2026 index, also Kharazian's, had Anthropic at 24.4% and OpenAI declining 1.5% month-over-month — the largest single-month OpenAI drop the index had ever recorded. Year-over-year, Anthropic moved from roughly 4% to 24.4% in twelve months. Then it added another ten points in April. OfficeChai's reporting puts the full YoY gain at 26 percentage points for Anthropic.

The other half of the swing is OpenAI's. Zero-point-three percent growth over a year, in the same window in which it shipped GPT-5.5 Spud, stood up the $4 billion DeployCo subsidiary, and pushed Codex onto Cerebras silicon, is the number that should worry the OpenAI sales team. The product line is broader than it has ever been. The new-logo additions on the corporate-card metric did not move. Zero-point-three is the kind of number you put in eight-point font.

Kharazian's verdict, quoted in The Decoder's writeup of the same report, reads as close to a call as a Ramp economist gets: "the software industry is dynamic, where newcomers can disrupt market leaders in a matter of months."

What the corroboration looks like

Two adjacent signals back up the index. OpenRouter's public router usage shows Anthropic share growing on third-party routing — the segment Ramp's card data does not capture cleanly. Anthropic's own $7 billion B2B annualised run-rate disclosure at the developer day three weeks ago was the company's headline figure. The index is not the only place the curve is bending.

Worth saying what the index is not saying. It is not saying Anthropic has more revenue than OpenAI — OpenAI's consumer ChatGPT business is still multiples of Claude's. It is not saying Anthropic is winning Fortune 500 enterprise — that segment is in neither sample. It is saying that, in the slice of US tech-forward business that runs on Ramp cards, Claude is now the more common line item.

What this means

Three things follow.

1. The "OpenAI for everyone, Anthropic for engineers" framing is dead. It was always weak — Claude's adoption was higher among technical buyers from day one — but at 34.4% paid US business adoption, Anthropic is not a developer niche any more. The next time someone calls Anthropic the "AI for builders" shop, the answer is that one in three US business buyers on the Ramp platform put a Claude charge on a corporate card last month.
2. OpenAI's defensive playbook now has to land. DeployCo, the Bedrock cross-listing, the 92% price hike on GPT-5.5 — each was pitched as a B2B-share play. The Ramp number says none of them moved the corporate-card metric over twelve months. Whichever lever OpenAI pulls next is being pulled into a wind it has been losing against for a year.
3. The compute story is the lead's biggest risk. If Anthropic cannot serve the demand it just won — and Kharazian's "turning away revenue" line says it cannot — the same buyers who switched in to Claude in April will be the first to add an OpenRouter line item or a Bedrock OpenAI fallback in June. The index Ramp publishes in three months is the one that tells whether the cross-over was a milestone or a peak.

The 2.1-point lead is the headline. The sentence Ramp's own economist used elsewhere in the same post — "newcomers can disrupt market leaders in a matter of months" — runs in both directions. Anthropic took the metric from a standing start in eighteen months. There is nothing in the methodology that says OpenAI cannot take it back in the same window.