Anthropic passed OpenAI in business AI spending in April 2026, according to Ramp's monthly AI Index, which pulls from actual payment data across more than 50,000 U.S. companies. This is the first time any company has led that measure since ChatGPT launched in late 2022.
The gap is narrow: 34.4% of businesses paying for Anthropic versus 32.3% for OpenAI. But the trajectory behind it is steep. Anthropic went from 9% of businesses a year ago to 34% now, while OpenAI's share has been sliding since it peaked near 36.5% in mid-2025.
The engine behind most of that growth is a single product. Claude Code, Anthropic's AI coding tool, became the fastest-growing product in the company's history and pulled a wave of technically-minded employees into the platform. Those employees tended to work in software, finance, and professional services, and they brought their organisations with them.
Here is what makes this data more reliable than most industry reports: it is based on what companies actually paid, not what they said in a survey. Ramp processes corporate card and invoice payments. There is no self-reporting bias.
That said, OpenAI still leads in overall revenue and in large enterprise contracts, which typically do not get paid on a corporate card. The Ramp data skews toward mid-market and growth-stage companies in the United States.
More telling than the market share number is what is happening inside the buying patterns. By February 2026, 79% of Anthropic's customers were also paying for OpenAI. This is not a mass defection. It is companies adding a second or third AI provider alongside the first. The average large enterprise now maintains 3.2 AI vendor relationships, according to Gartner.
Organisations with the most AI experience have settled on a practical position: stop trying to pick one winner, and build your setup so you can swap tools without pain. The AI models will keep leapfrogging each other. What a company builds around those models, including its own data, its workflows, and its internal processes, is where the actual advantage sits.
There is a cost problem building underneath all of this. The same Ramp data shows that the average business is spending 13 times more on AI tokens than it was in January 2025. Uber's CTO disclosed that the company burned through its entire 2026 AI budget in four months, mostly on Claude Code. Anthropic makes money when its customers use more tokens, which means the company is financially incentivised to push users toward its more expensive models, even when a cheaper one would do the job.
This creates a pattern worth watching in your own organisation. AI costs are usage-based, which means they grow faster than a traditional software licence. The bills can arrive as a surprise. The companies handling this well are tracking which team or workflow is generating each cost, tying that cost to a measurable business result, and using cheaper AI options for simpler tasks.
The last shift worth noting is about what AI tools are starting to do. Until recently, the question was whether AI could produce good enough output. That question is largely settled. The new question is about AI systems that take actions: searching for information, calling external services, running code, making decisions in sequence without a person reviewing each step. These systems require a different kind of oversight than a tool that just generates text. The evaluation moves from "is this output good?" to "what can this system do inside my organisation, and what can I stop it from doing?"
For anyone not yet spending on AI: half of all U.S. businesses are now paying for it, and the number is climbing every month. The question is less whether to start and more how to set up a structure that keeps costs visible and data protected from the first day.