Analysis
New spending data from corporate card and expense platform Ramp shows OpenAI narrowing the enterprise gap with Anthropic, even as Anthropic still leads on overall share. Ramp analyzed spend patterns across more than 70,000 American businesses and found Anthropic's share of AI spend grew from 41% in May 2026 to nearly 44% in July, while OpenAI grew from 39% to nearly 40% over the same stretch. But "OpenAI is currently growing faster among this segment in Q3 to date than Anthropic," Ramp economist Ara Kharazian said. The share of Ramp's business customers paying for any AI product at all climbed from just over 50% in March to nearly 56% by July -- the category is still expanding, which means both companies can grow revenue simultaneously even as their relative share shifts month to month.
The data has real limitations: it excludes large enterprises that route spend through American Express or other card providers, and Ramp shared percentage splits rather than absolute dollar figures, so a shift from 41% to 44% doesn't tell you whether the underlying spend pool grew or shrank for either company. Ramp's own analysis flags that businesses are willing to switch AI providers based on new model releases, which is the more consequential finding: enterprise AI spend looks considerably less sticky than the annual-contract, high-switching-cost SaaS relationships investors are used to valuing.
The zero-retention play
OpenAI moved on the product side the same day. On August 20, the company announced Private Safety Processing, an automated system that scans customer interactions for potential misuse and returns limited safety signals without exposing the underlying prompts or responses to OpenAI personnel. The pitch is explicitly aimed at the objection enterprise buyers raise most often about handing sensitive data to a foundation model provider. It's a direct contrast with how Anthropic currently handles safety review: Anthropic retains prompts and outputs for 30 days on its covered models for safety purposes, with human review triggered when automated systems flag potential harm. OpenAI says human intervention under its new system happens only when child exploitation material is detected, with all other safety flagging staying fully automated and unretained -- though OpenAI has not yet published the detailed documentation it says is coming next month, so the claim isn't independently verifiable yet.
For enterprise buyers actually making the decision, the practical calculus increasingly comes down to workflow fit rather than brand loyalty to either lab: legal and compliance teams evaluating Anthropic for its safety framing, engineering teams pulling toward OpenAI for tooling breadth, and finance teams comparing both against per-seat cost. That fragmentation by function, more than any single company's messaging, is what's driving the month-to-month share swings Ramp is recording.
A three-point swing in market share over three months is well within the range of normal quarter-to-quarter noise for a fast-growing category, not proof that OpenAI has structurally closed the gap. Anthropic still leads outright, and Ramp's own data shows businesses churning between providers based on whichever model shipped most recently -- which means this week's OpenAI gains could just as easily reverse the next time Anthropic ships a model update. The zero-retention pledge is also a claim OpenAI has made about its own systems, unverified by a third party, at exactly the moment it has the most incentive to make that claim sound credible to enterprise buyers evaluating Anthropic.
