A year ago, one in every 25 businesses on Ramp, the enterprise spending platform, was paying for Anthropic's AI models. Today it is one in four. That number is counterintuitive for one reason: the Pentagon banned Anthropic from federal contracts during the same period when the company's enterprise business was accelerating fastest.
The data comes from Ramp's AI Index, a monthly digest of spending patterns across more than 50,000 businesses that use the platform to manage operational costs. The numbers are striking. In February, Anthropic's customer share grew 4.9 percent month over month, its largest single-month gain since Ramp started tracking business AI adoption. OpenAI, in the same period, declined 1.5 percent, the steepest single-month drop for any AI model company in Ramp's dataset. Anthropic now commands roughly 30.6 percent of the combined Anthropic-OpenAI customer base on Ramp, compared to OpenAI's 35.2 percent. The gap has nearly vanished in 12 months. A year ago Anthropic held about 10 percent of that combined spend; OpenAI dominated.
The trajectory matters more than any single snapshot. Among businesses buying AI services for the first time, Anthropic wins roughly 70 percent of head-to-head competitions against OpenAI, according to Ramp's economist Ara Kharazian. That is the pipeline. Those first-time buyers become repeat customers. Anthropic leads OpenAI among venture-backed companies, a cohort that is simultaneously the most aggressive AI adopter and the most likely to set enterprise norms for the next decade: 66 percent of VC-backed firms on Ramp now pay for Anthropic, versus 59 percent for OpenAI.
Why is this happening? The wire angle is the ban, and the ban probably matters, but not in the obvious way. The Pentagon controversy made Anthropic more visible, not more trustworthy. What seems to be driving the shift is something more durable: Anthropic has positioned itself as the default choice for companies that want to signal they take AI safety seriously, without sacrificing capability. That positioning mirrors what Amazon did in cloud. AWS did not win because it was the cheapest or the most technically sophisticated. It won because it arrived at the right moment in the enterprise adoption cycle and became the implicit standard.
The pattern in the Ramp data mirrors that logic: Anthropic holds the top position in the three highest-adoption sectors, information and software, finance, and professional services. These are exactly the industries that drove early AWS adoption and that are now writing the playbook for AI procurement. When a VC-backed software company chooses an AI model provider today, it is making the same kind of infrastructure commitment that its predecessor made when it chose AWS in 2008.
There is also a compute constraint that the data cannot capture. Anthropic has at times struggled to secure enough of Nvidia's advanced chips to meet demand. A 70 percent win rate against OpenAI means little if Anthropic cannot scale fast enough to serve the customers it is winning. OpenAI has its own problems: a declining share of first-time buyers is a lagging indicator of a positioning problem that compounds.
The main caveat is that Ramp's data skews toward tech-native companies. Traditional industries like healthcare, manufacturing, and logistics are underrepresented. If those sectors follow the same pattern, the consolidation trend is real. If they diverge, Anthropic's win is concentrated in a slice of the market that, while influential, is not the whole economy.
OpenAI is not ignoring the problem. The company has been aggressive on pricing, partnerships, and model releases. But the Ramp data suggests that among the cohort setting adoption norms, Anthropic has already done something OpenAI has not: made its brand synonymous with the right kind of AI risk. Not the risk of models being too capable, but the risk of models that are not safe enough for consequential use. That framing is expensive to build and difficult to copy.
What to watch: whether Anthropic can convert its enterprise position into durable revenue at scale, whether the compute constraint becomes a binding limit, and whether traditional industries follow the VC-backed cohort or diverge. The numbers say Anthropic has won the first round. The game is not over.
