For the first time, Anthropic has more verified business customers than OpenAI in Ramp's monthly AI Index โ 34.4% of participating businesses pay for Anthropic services versus 32.3% for OpenAI, drawn from corporate expense data across more than 50,000 companies on the Ramp platform. The trajectory is the part worth dwelling on: Anthropic sat at 9% of businesses in May 2025; by May 2026, that climbed to roughly 35% โ a 26-point gain over twelve months. OpenAI's share over the same period fell by one point. Overall enterprise AI adoption rose nine points, so this is a share shift, not a pie-expansion artifact. Ara Kharazian, Ramp's economist, framed Anthropic's path explicitly: "start with a very technical customer base, focus on their needs, really succeed in execution and then start broadening out through tools like Cowork."
The data has the load-bearing caveats. Ramp customers skew tech-forward โ startups, finance, professional services โ so the index over-represents the cohorts where Claude already had strong adoption (Claude Code, Claude API, developer tools). Kharazian himself noted Anthropic "has already been in the lead amongst the high adoption groups," and that OpenAI still leads "across the other firms" though the gap is shrinking. The triangulating signal is that OpenRouter's leaderboard, which samples a different population, last showed OpenAI ahead of Anthropic in December 2025; both samples now agree on the directionality. Kharazian published a blog post arguing the lead may not durable, which is a useful counterweight โ the next twelve months depend on whether GPT-5's enterprise rollout pulls back share or whether Anthropic's Cowork-and-MCP-and-Claude-for-Legal expansion compounds the lead.
The ecosystem read here connects threads that have been moving in parallel all spring. Anthropic shipped Claude for Legal expansion via MCP into DocuSign/Box/Westlaw, multi-agent orchestration tooling at Code with Claude, and Cowork as an enterprise-native deployment surface โ each of those moves landed on infrastructure that's now generating market share against OpenAI's mostly direct-API approach. OpenAI counterprogrammed with the Daybreak cybersecurity initiative, the Operator agent line, and broader consumer plays; the result is two labs that have visibly diverged in customer focus. Anthropic's lead is concentrated in builders, developers and regulated industries; OpenAI retains broader cross-industry and consumer reach. The MCP standard, which Anthropic shipped and most of the wrapper economy adopted, is part of why technical customers are choosing Claude โ it lowers the integration friction every other commitment locks in. For the wrapper economy, the implication is the choice between standardizing on Anthropic's Claude/MCP stack and OpenAI's GPT/Function Calling stack just got harder for tech-heavy customers.
For builders: if you're choosing a primary model provider for a new enterprise product right now, the Ramp index is one data point among several to weigh, not a verdict. Two concrete things to watch over the next two quarters: (1) whether OpenRouter's leaderboard share gap widens or closes โ it samples a different population so independent confirmation matters; (2) whether GPT-5's enterprise rollout (the rumored June/July window) lands as a counterpunch big enough to flip the trajectory. Honest hedging: the 26-point Anthropic gain in twelve months is unusual enough that you should expect some reversion regardless of model quality โ enterprises buying AI subscriptions often do so on a trial basis before consolidating, and the second-half-2025 onboarding cohort will go through its first renewal cycle soon. The "Anthropic leads enterprise" headline is true today; the question is whether it's true in 12 months.