A year ago, this would have been unthinkable.

Anthropic has passed OpenAI in paid business adoption for the first time, according to the latest Ramp AI Index published Tuesday. The numbers: 34.4% of businesses on Ramp's platform now pay for Anthropic services, compared to 32.3% for OpenAI. Overall AI adoption crossed the 50% mark, hitting 50.6% of all tracked businesses.

To appreciate how dramatic this reversal is, rewind 12 months. OpenAI was the undisputed king of enterprise AI. ChatGPT had brand recognition that no competitor could touch. Anthropic was the safety-first company that executives vaguely knew existed. In January 2026, Ramp data showed OpenAI "far ahead" of rivals across software development, research, finance, and customer support.

Then Anthropic quadrupled its business adoption in a single year. OpenAI grew by 0.3%.

The timing makes this particularly brutal for OpenAI. Sam Altman is in the middle of the Musk v. Altman trial, closing arguments are expected Thursday, and the company is racing toward a late-2026 IPO at a $1 trillion valuation. Losing the enterprise adoption crown to your primary rival during your biggest legal battle and just months before your IPO is the kind of narrative that makes investors nervous.

What Drove the Flip

Claude Code. That is the short answer. Anthropic's developer-focused coding agent launched in late 2025 and became the tool that engineers refused to stop paying for. The longer answer involves Anthropic's broader enterprise push: $44 billion in annualized revenue (up from $9 billion at the end of 2025), gross margins that doubled from 38% to 70%, and more than 1,000 customers spending over $1 million per year. Eight of the Fortune 10 are Anthropic customers.

But Ramp's lead economist Ara Kharazian, who published the index, is not ready to declare Anthropic the winner. In fact, he flagged three specific headwinds that should worry the company.

First, incentive misalignment. Anthropic makes more money when businesses buy more tokens, which means the company is incentivized to push users toward expensive models even when cheaper ones would work. Uber's CTO already announced the company blew through its entire 2026 AI budget. If cost-cutting drives firms to route toward cheaper models, Anthropic loses.

Second, Claude has been getting worse. Users have reported frequent outages, rate limits, and declining response quality in recent weeks. Anthropic responded with a public postmortem and usage limit reset in April, and the new SpaceX Colossus compute deal should help. But reputational damage from reliability issues is hard to undo.

Third, Anthropic's latest model update would triple token costs for any prompt that includes an image. When your customers are already worried about costs and you are already struggling with compute capacity, making things 3x more expensive is a bold move.

What Happens Next

The most interesting data point Ramp surfaced is not the Anthropic number. It is the rise of AI inference platforms selling access to cheap, open-source models. Some of the fastest-growing vendors on Ramp's platform last month were these inference brokers. And OpenAI's Codex does similar tasks to Claude Code, often more cheaply, with minimal switching costs.

Here is the uncomfortable truth for both Anthropic and OpenAI: this market may not have a durable leader. Kharazian wrote that he has "never seen a software industry as dynamic, where newcomers can disrupt market leaders in a matter of months." The switching costs are trivially low. The models are converging in capability. The next breakthrough from either side could flip the numbers again by July.

This is the scariest possible outcome for anyone trying to value these companies. Anthropic is raising $30 to $50 billion at a $950 billion valuation. OpenAI is targeting a $1 trillion IPO. Both valuations assume a company that can hold market position for years. The Ramp data suggests neither can hold it for months.

First reported by Ramp. Additional reporting from Business Insider (Alistair Barr), Axios, and TipRanks.