AI access is a utility

A year ago, picking an artificial intelligence assistant still felt like picking a favorite app. In April 2026, it looks more like choosing an electricity plan: free, mid-tier, premium, team, enterprise, plus add-ons for heavier use. (openai.com) (claude.com) (gemini.google.com) OpenAI now sells ChatGPT in consumer and workplace layers, with free access at one end and Business and Enterprise at the other, and its pricing page says the paid tiers buy more features, more model access, and faster response times. Its help pages also describe separate plans like Go, Plus, and Pro, which turns “use the chatbot” into a menu of service levels. (openai.com) (help.openai.com) Anthropic has moved the same way. Claude now has Free, Pro, Max, Team, and Enterprise plans, and its pricing page says Max comes in two usage bands, “5x” and “20x,” which is the language of capacity, not just software features. (claude.com) Google has gone even further by folding artificial intelligence into the place many offices already work all day. Google Workspace now lists Gemini features inside Starter, Standard, Plus, and Enterprise plans, and Google also sells separate “Expanded Access” and “Ultra Access” add-ons for higher limits on image generation, video generation, automation, and speech translation. (workspace.google.com) (knowledge.workspace.google.com 1) (knowledge.workspace.google.com 2) That changes the buying decision. If one vendor’s assistant lives in your email, documents, browser, phone, and desktop app, switching stops being “which model writes better” and starts becoming “which company runs a bigger share of our workday.” (workspace.google.com) (claude.com) (openai.com) The pricing pages now read like utility schedules because the products are sold around limits. Google’s consumer Gemini plans list 50, 200, and 1,000 monthly artificial intelligence credits depending on tier, while Anthropic sells higher usage explicitly as Max 5x or Max 20x. (gemini.google.com) (claude.com) OpenAI’s workplace plans do the same in a different style. Its Business and Enterprise material separates “unlimited” access for some models from “flexible” access for others, and its rate-card help page describes model usage in credits and message equivalents rather than a single flat bucket of chat. (openai.com) (help.openai.com) Once access is sold in layers, companies get a new management problem. Two employees with the same job title can quietly end up with different model quality, different speed, different context windows, and different automation limits if one has a premium seat and the other does not. (openai.com) (workspace.google.com) (knowledge.workspace.google.com) That inequality is easy to miss because it does not look like old software licensing. One person gets a stronger reasoning model, another gets more video credits, a third gets desktop extensions or workspace connectors, and all three still say they are “using artificial intelligence.” (claude.com) (gemini.google.com) (help.openai.com) The next argument inside companies will not be whether to allow artificial intelligence at all. It will be who gets the equivalent of residential service, who gets commercial service, and who gets the industrial plan with the bigger meter. (openai.com) (claude.com) (workspace.google.com) The practical fix is boring and specific: write down which roles get which tier, which tasks justify premium access, and which limits trigger an upgrade. If that policy stays informal, the loudest teams will get the best tools first, and the gap will hide inside subscription menus and admin consoles. (openai.com) (claude.com) (knowledge.workspace.google.com)