- Ramp’s AI Index put Anthropic at 34.4% of paying companies and OpenAI at 32.3% — Anthropic’s first lead on the metric.
- Anthropic quadrupled its B2B penetration over the past year; OpenAI grew only 0.3 percentage points.
- Ramp’s index tracks payment share, not actual usage, and skews toward U.S. companies.
- Ramp economist Ara Kharazian flagged three near-term risks to Anthropic’s lead: model-cost pricing, Claude outages, and Opus 4.7 image-processing costs tripling.
What Happened
Anthropic surpassed OpenAI in business-to-business AI adoption for the first time, according to spending data from corporate-card and invoicing platform Ramp, The Decoder reported on Wednesday. The Ramp AI Index measured Anthropic at 34.4% of the companies paying through Ramp, with OpenAI at 32.3%. Anthropic quadrupled its share over the past year, while OpenAI’s share grew by 0.3 percentage points.
Why It Matters
The crossover is a milestone in a competitive narrative that has tilted toward Anthropic on developer adoption metrics through 2025 and 2026. Claude Code‘s release in February 2026 and the Claude Cowork collaborative environment launched in March have widened Anthropic’s product surface in enterprise developer and knowledge-work contexts. The Ramp data captures the financial expression of that shift: more companies running Anthropic invoices through their corporate-spend platform than OpenAI invoices.
The caveats matter. Ramp tracks share of paying customers, not usage or revenue. A small Anthropic API subscription counts the same as a large one. Ramp’s customer base also skews toward U.S. small- and mid-market firms, so the index understates the global enterprise picture where OpenAI’s Azure-distributed deployments remain dominant.
Technical Details
Ramp economist Ara Kharazian flagged three risks to Anthropic’s lead in the report. First, Anthropic earns more revenue when customers run more expensive models, which could push customers toward cheaper alternatives — Uber’s CTO publicly disclosed exceeding the company’s 2026 AI budget. Second, users have been vocal about Claude outages and perceived quality regressions. Third, the recently released Opus 4.7 model triples the cost of image processing relative to its predecessor.
Agentic workflows can also distort the spending picture. Per Kharazian, an agentic task generates more tokens by calling tools, writing or running code, and generating intermediate reasoning. A workflow that looked identical a year ago can burn through far more tokens today even when the underlying user demand is flat. Kharazian wrote: “The truth is we have never seen a software industry as dynamic, where newcomers can disrupt market leaders in a matter of months, and where the pace of development overrides the typical forces of vendor stickiness.”
Who’s Affected
Anthropic’s leadership team and investors gain a clean external data point ahead of the company’s next funding round. OpenAI’s enterprise team faces a narrative challenge ahead of its anticipated 2026/2027 IPO, where B2B traction is a primary equity story. Cheaper inference platforms running open-source models — including Together AI, Fireworks AI, Anyscale, and Groq — are picking up share at the lower end of the market, per Kharazian’s note. Mistral, Cohere, and other model providers continue to compete for the long tail. Enterprise procurement teams now have a third credible data source — alongside Stack Overflow’s Developer Survey and Vellum’s State of AI — to reference in vendor evaluations.
What’s Next
Ramp publishes the AI Index monthly; whether Anthropic holds the lead in subsequent updates will depend on enterprise reception of Opus 4.7 pricing and on Claude’s perceived reliability through May and June. Anthropic’s launch of Claude for Small Business on the same day as the Ramp data release suggests management is already pushing into the next tier of customers. OpenAI is expected to address the competitive picture at its upcoming developer conference. The Ramp data set itself has limitations — share of payers, not revenue or usage — that any single-month interpretation should account for.
