Enterprise AI dollars concentrate
Enterprise AI spending is still growing but buyers are becoming choosier, favoring vendors that deliver low‑friction deployment, clear economics and usable agents rather than purely flashy models. OpenAI says enterprise now accounts for over 40% of its revenue, even as reports suggest a growing share of new enterprise AI spend is flowing to Anthropic and similar vendor platforms. (decrypt.co) (businessinsider.com)
A funny thing is happening in artificial intelligence: the flashiest consumer chatbot is no longer the whole game, and OpenAI now says business customers generate more than 40% of its revenue. In an April 8 note, Chief Revenue Officer Denise Dresser said enterprise sales are on track to match consumer revenue by the end of 2026. (openai.com) That means the center of gravity is shifting from millions of curious users to a smaller number of companies signing bigger contracts. OpenAI says its application programming interfaces now process more than 15 billion tokens per minute, and it named Goldman Sachs, Phillips, State Farm, DoorDash, and Thermo Fisher among customers using its business products. (openai.com) But the money is not spreading evenly across the field. Axios reported on March 18 that Anthropic was capturing more than 73% of spending from companies buying artificial intelligence tools for the first time, using data from Ramp, the corporate card and expense software company. (axios.com) That 73% number is about new buyers, not all buyers, and that distinction matters. It suggests the newest enterprise dollars are picking a narrower set of vendors even while OpenAI keeps a very large installed base and a large overall revenue stream. (axios.com) (openai.com) Venture firm Menlo Ventures saw the same pattern in a wider 2025 enterprise survey. Menlo estimated Anthropic at 40% of enterprise large language model spend, OpenAI at 27%, and Google at 21%, with those three vendors together accounting for 88% of enterprise model usage. (menlovc.com) The important part is where companies are spending, not just which model wins a benchmark chart. Menlo said more than half of enterprise artificial intelligence spending in 2025 went to applications rather than base models, which means buyers are paying for tools that plug into work instead of raw intelligence on its own. (menlovc.com) Coding has become the cleanest example of that shift. OpenAI said Codex reached 3 million weekly active users, while Anthropic is pushing Claude Code as a managed enterprise product with centralized controls, security features, and customer case studies claiming 79% faster feature delivery and 89% employee adoption in one deployment. (openai.com) (claude.com) This is why “agents” keep showing up in every enterprise pitch. OpenAI’s Denise Dresser told Decrypt that customers are moving from single chatbots toward teams of software agents that can handle connected tasks, which is closer to hiring a junior digital staff than opening a smarter search box. (decrypt.co) Enterprise buyers are also acting more like procurement teams than fans. Anthropic’s business pitch now leans heavily on secure deployment, admin controls, and measurable output, and OpenAI’s business note leans on throughput, named customers, and workflow adoption rather than consumer virality. (anthropic.com) (openai.com) So the market is getting bigger and narrower at the same time. More companies are buying artificial intelligence, but a growing share of the newest checks is going to vendors that make deployment feel less like a science project and more like ordinary software procurement. (axios.com) (menlovc.com) OpenAI is not losing relevance in that world; a business that gets 40% of revenue from enterprise is already deep inside it. The change is that enterprise customers now seem less impressed by dazzling demos and more interested in whether a model can ship code, fit compliance rules, and produce a bill that a chief financial officer will sign. (openai.com) (decrypt.co)
