OpenAI: enterprise now >40% revenue

OpenAI says business customers now bring in more than 40% of its revenue, and the company says that share could catch up with consumer revenue by the end of 2026. In the same April 8 note, OpenAI said its application programming interfaces now process more than 15 billion tokens per minute and its coding product Codex has reached 3 million weekly active users. (openai.com) That is a different picture from the one most people still have in their heads. OpenAI’s public image was built on ChatGPT going viral with consumers, but the company said on March 31 that it is now generating $2 billion in revenue per month and described enterprise deployment as one leg of a four-part flywheel alongside consumer use, developer use, and compute. (openai.com) The business shift started before this week’s update. In June 2025, OpenAI said it had 3 million paying business users across ChatGPT Enterprise, ChatGPT Team, and ChatGPT Edu, up from 2 million in February 2025. (cnbc.com) Those business products sell something different from a consumer chatbot subscription. A company is not just paying for answers on a screen; it is paying to connect artificial intelligence to Google Drive, Microsoft SharePoint, Dropbox, Box, OneDrive, meeting transcripts, internal files, and the permission systems that decide who can see what. (cnbc.com) OpenAI’s own language shows how the pitch has changed. Denise Dresser, the company’s chief revenue officer, wrote that customers are “past the experimentation phase” and want a unified operating layer instead of isolated tools that “don’t talk to each other and just create chaos.” (openai.com) That is where “agents” come in. In OpenAI’s February 5 launch of Frontier, the company described agents as tools that can independently complete tasks on behalf of a user, and it said Frontier is meant to stitch together a company’s separate systems so those agents can work with shared context instead of acting blind. (cnbc.com) The important detail is that OpenAI is not talking about one agent doing one trick. Frontier was presented as a layer that can manage OpenAI agents, customer-built agents, and third-party agents from Google, Microsoft, and Anthropic, which turns the product into traffic control for a mixed fleet rather than a single car. (cnbc.com) Once companies buy that kind of system, the meter keeps running in a steadier way. Every employee query, document search, code generation job, workflow handoff, and automated task uses inference, which is the paid step where a trained model actually does work, and OpenAI says that work is already happening at industrial scale with 15 billion tokens per minute through its application programming interfaces. (openai.com) OpenAI is also naming the kinds of companies now buying in. Its April 8 enterprise note listed Goldman Sachs, Philips, and State Farm as new customers, while saying existing demand is growing from Cursor, DoorDash, Thermo Fisher, and LY Corporation. (openai.com) That customer mix helps explain why enterprise revenue is valuable. Banks, insurers, healthcare suppliers, and large software companies usually do not rip out core systems every quarter, so once artificial intelligence is tied into their files, workflows, and security rules, the relationship starts to look less like a viral app download and more like a long software contract. (openai.com; cnbc.com) OpenAI’s own forecast is blunt about where it thinks the business is heading. The company said enterprise is on track to reach parity with consumer by the end of 2026, which would mean the company best known for a mass-market chatbot is increasingly being paid like back-office infrastructure. (openai.com)