Enterprise AI shifts to controls and pricing
Model vendors are pivoting from novelty demos to procurement‑friendly products, pushing enterprise‑grade controls and lower pricing to win corporate budgets. Reports say OpenAI and Anthropic are packaging governance, auditability and cost predictability to appeal to large buyers that care more about integration and control than headline model performance. (siliconangle.com, aboutchromebooks.com)
The race in artificial intelligence has moved from “which chatbot feels smartest” to “which vendor can survive a procurement meeting,” and that is why Anthropic spent April 2026 talking about organization-wide controls while OpenAI cut the cost of Codex access for business customers. (siliconangle.com) OpenAI’s own pricing pages now frame ChatGPT Business at $20 per user per month on annual billing, while Enterprise is sold through custom contracts with centralized billing, security features, and service terms aimed at larger companies. (openai.com) OpenAI also changed how new ChatGPT Enterprise workspaces are billed, shifting them to token-based rates instead of the older message-based system, which gives finance teams a meter they can model the way they model cloud-computing spend. (help.openai.com) That pricing shift landed next to control features that information-technology departments already understand: OpenAI supports Security Assertion Markup Language single sign-on, System for Cross-domain Identity Management provisioning, and audit logs for workspace and application programming interface activity. (help.openai.com, help.openai.com, help.openai.com) OpenAI’s Compliance Platform goes one step further by exporting workspace logs and metadata into eDiscovery, data loss prevention, and security information and event management tools, which is the language of legal, security, and compliance teams rather than software demos. (help.openai.com) Anthropic is making the same pitch from the other side. Its Enterprise plan advertises single sign-on, System for Cross-domain Identity Management, audit logs, role-based permissions, and a promise not to train on Claude for Work customer data by default. (anthropic.com) Anthropic’s newer push is to manage artificial intelligence agents the way companies manage employees: Claude Cowork is sold as a desktop system for multi-step office work, and Anthropic is adding group-based access controls, spend caps, usage analytics, and connector controls for broad rollouts. (anthropic.com, anthropic.com) Microsoft has been training buyers to think this way for a while with Azure OpenAI Service, which sells not just model access but predictable capacity through Provisioned Throughput Units, a reserved slice of compute meant to avoid the cloud version of a crowded highway at rush hour. (azure.microsoft.com, techcommunity.microsoft.com) The result is that “best model” is becoming only one line item in the buying decision. A bank, hospital, or manufacturer usually cares just as much about who can set seat policies, lock down connectors, export logs, cap spending, and fit into existing identity systems on day one. (help.openai.com, anthropic.com, anthropic.com) That is also why Anthropic put $100 million behind its Claude Partner Network in 2026: large companies rarely buy a new platform from a website alone, and systems integrators are the people who turn a pilot into a signed rollout across finance, human resources, and security teams. (anthropic.com) The market still talks about benchmark scores and model names, but the money is starting to follow the vendors that can answer dull questions with exact answers: Who can see the data, who approved access, what happened last Tuesday, and what the monthly bill will be before the quarter ends. (siliconangle.com, help.openai.com, anthropic.com)
