Anthropic Passes OpenAI in Enterprise Adoption, but There Is a Catch

It took three years to get here. In June 2026, corporate card platform Ramp released its latest AI Index, showing that Anthropic reached a 41% paid adoption rate among US businesses. That edged out OpenAI's 39.5% for the first time. Ramp tracks transaction data across more than 70,000 companies, reflecting actual card spend rather than self-reported survey answers.

Yet in that same report, Ramp Chief Economist Ara Kharazian pushed back against the celebration: "Anthropic's incentives are not aligned with enterprise customers." Because Anthropic sells tokens, its revenue grows when customers burn more compute. Structurally, that encourages vendors to route workloads to expensive frontier models even when lighter, cheaper models handle the job just as well. Kharazian's conclusion was blunt: a single month's lead does not constitute a moat.

From 0.03% to 41% in Three Years

The reversal has been swift. In June 2023, Anthropic accounted for just 0.03% of corporate AI spend in Ramp's database—essentially a rounding error. At the time, OpenAI dominated the market, and ChatGPT Enterprise was the default choice for corporate procurement teams.

The shift started in technical communities. Claude gained traction among engineers and technical leaders by offering longer context windows, reliable instruction-following, and an explicit emphasis on safety and governance. Anthropic then launched Claude Code, a developer-focused agentic coding tool that became the fastest-growing product in the company's history. That engineering credibility soon spread to finance, legal, and research teams. Over the past year, Anthropic's enterprise adoption grew fourfold, while OpenAI grew by just 0.3%.

As Kharazian noted, Anthropic started with a highly technical user base, focused on their specific needs, executed well, and then expanded into broader business workflows with products like Cowork.

Smart Buyers Are Done Betting on a Single Provider

Another data point in Ramp's June report reveals a deeper shift. Across the 70,000 businesses tracked, the top 1% of AI power users spend $7,500 per user each month on AI tools. The median business spends just $11.38—a 680-fold gap.

The companies spending the most and integrating AI most deeply follow a distinct pattern: they mix multiple frontier models while pairing them with low-cost open-source options. In Ramp's June data, one of the fastest-growing AI providers was DeepSeek, an open-source model provider known for competitive pricing. The most advanced buyers now treat models as interchangeable infrastructure rather than platforms requiring deep lock-in. Ramp noted that this directly threatens any vendor relying on customer stickiness.

At the same time, an IDC survey of enterprise CIOs conducted in March presents a different angle. OpenAI is still used by roughly 42% of organizations and Google by 38%. Only 19% report using Anthropic widely, with another 25% still evaluating it. Both datasets are consistent because they measure different things: Ramp tracks corporate card transactions at small and mid-market growth firms, while IDC measures deployment depth inside large enterprises. Anthropic has won the initial entry point, but the battle across enterprise workflows is far from finished.

The Starbox AI View: The Winner Is Neither Anthropic nor OpenAI

For enterprise leaders, market share swings point to an operational reality: if your workflows and proprietary knowledge are hardcoded to a single provider, every vendor price cut or model update forces an expensive rebuild.

At Starbox AI, we see a clear pattern across client deployments: the teams generating the highest ROI from AI are not those searching for a single "best model." They are the ones building model-agnostic architectures. Starbox AI uses multi-model routing to dynamically assign tasks based on performance and cost: Claude handles complex reasoning, lightweight models process high-volume routine tasks, and vertical models step in for specialized domain work. Enterprise knowledge stored in our four-dimensional knowledge system remains owned by the business, never locked inside a third-party API.

This matches the strategy used by the top 1% of enterprise AI spenders: treat foundation models as interchangeable utilities, avoid single-vendor dependence, and keep workflows and knowledge assets under your own control. Kharazian noted that a temporary market lead is not a defensible moat for Anthropic. For enterprise buyers, the takeaway is the inverse: never let a vendor's temporary advantage set the boundaries of your own competitive advantage.

Regulatory Context

On July 1, Colorado signed the Chatbot Safety Act (HB 26-1263). Following its earlier law governing AI in employment decisions, this marks the state's second AI statute this year and will take effect in January 2027. With US federal AI legislation still stalled, state-level regulations are moving quickly to fill the void. Enterprise compliance teams should begin preparing now.