Enterprise AI now sells like software
Enterprise buyers are demanding clear pricing, not mystery usage bills, so model vendors are publishing rate cards and per-feature price lists. OpenAI released a Codex rate card showing tiered, credit-based consumption and Anthropic emphasised model-and-feature pricing in its API docs, making costs more forecastable for procurement teams (help.openai.com) (platform.claude.com). That shift matters because finance teams now expect predictable economics before approving pilots, and OpenAI’s move—plus its comment that enterprise makes up 40% of revenue—signals this is becoming a core enterprise-software buying problem, not just a research purchase (cnbc.com).
A lot of companies loved artificial intelligence demos until the bill arrived. The new fight in enterprise artificial intelligence is not over model quality first; it is over whether a finance team can predict next quarter’s spend before signing a contract. (help.openai.com) (platform.claude.com) OpenAI changed its Codex pricing on April 2, 2026, and said the product now aligns with application programming interface token usage instead of per-message pricing for new ChatGPT Business and new ChatGPT Enterprise plans. That is a move away from “how many chats happened” and toward the kind of metered unit procurement teams already see in cloud software. (help.openai.com) OpenAI’s Codex page is now literally a rate card. It lays out credit rates by plan, says credits can extend usage beyond included limits, and separates legacy plans from the new structure so buyers know which rules apply to which contract. (help.openai.com) Anthropic is doing the same thing from the application programming interface side. Its Claude pricing page breaks costs out by model and feature, lists prices in United States dollars, and points customers to current pricing as a standard reference instead of forcing them into custom back-and-forth for every estimate. (platform.claude.com) Anthropic has also published a Usage and Cost application programming interface for organizations. That gives a company’s administrators programmatic access to historical usage and cost data, which is the kind of control finance teams expect when software starts spreading across multiple departments. (platform.claude.com) This looks boring until you remember how enterprise software gets bought. A chief information officer can love a coding agent, but a procurement team still wants a price sheet, a budget owner wants a forecast, and a finance chief wants to know what happens if usage doubles in month three. (platform.claude.com) (help.openai.com) OpenAI’s own revenue mix shows why it is changing its posture. On April 8, 2026, CNBC reported comments from OpenAI executive Denise Dresser that enterprise now makes up 40 percent of the company’s revenue and could match consumer revenue by the end of 2026. (cnbc.com) CNBC had already reported on January 21, 2026, that enterprise customers accounted for roughly 40 percent of OpenAI’s business and 80 percent of Anthropic’s business. When that much revenue comes from companies instead of individual users, pricing stops being a side note and becomes part of the product. (cnbc.com) That is why these pages read more like software catalogs than research lab experiments. A model vendor is no longer just selling intelligence in the abstract; it is selling a budget line that has to survive legal review, procurement review, and a chief financial officer asking for a spreadsheet. (help.openai.com) (platform.claude.com) The companies that win the next round of enterprise artificial intelligence deals may not be the ones with the flashiest demo. They may be the ones whose pricing page lets a Fortune 500 buyer estimate spend in 10 minutes and take that number straight into an approval meeting. (help.openai.com) (platform.claude.com)
