Pricing is shifting toward compute
Industry chatter shows vendors moving from flat subscriptions toward compute‑aware or usage‑sensitive pricing because a small group of power users consumes a disproportionate share of compute — one post cites power users taking roughly 50% of capacity. VC subsidy tactics have kept low sticker prices alive for now, but several voices warn that those subsidies could end and trigger multi‑fold price increases after they stop. (x.com) (x.com)
Artificial intelligence pricing is moving away from one flat monthly fee and toward meters, credits, and stricter limits as companies tie bills more closely to computing use. (openai.com) OpenAI now sells ChatGPT Business at $20 per user per month, but it also says “unlimited” access is subject to abuse guardrails, and its Business, Enterprise, and Education plans use a credit rate card for higher-end models and tools. The current card prices some actions at 10 credits for a GPT-5.4 Thinking message, 50 credits for a GPT-5.4 Pro message, 30 credits for an Agent message, and 50 credits for one Deep Research task. (openai.com) OpenAI has also added pay-as-you-go credits for consumer plans. Its help center says Plus and Pro subscribers can buy extra credits for Codex after hitting plan limits, and Sora video generation uses a credit schedule that rises with length and quality, from 10 credits for a 10-second Sora 2 video to 500 credits for a 15-second high-resolution Sora 2 Pro video. (help.openai.com) Google is using the same playbook with both subscriptions and usage caps. Its consumer Gemini help page says Google AI Pro and Google AI Ultra come with daily prompt, image, agent, and Deep Research limits, and warns that those limits can change based on demand and capacity. (support.google.com) The current Gemini consumer tiers show how detailed those limits have become. Google lists Google AI Pro at $19.99 a month with 1,000 monthly artificial intelligence credits for Flow and Whisk, while the help page says Google AI Ultra users can get up to 1,500 “Thinking” prompts a day, 120 Deep Research reports a day, and 1,000 images a day. (gemini.google.com) Google’s business products are moving in the same direction. Workspace’s AI Ultra Access add-on promises the “highest access” to video generation, image generation, NotebookLM, and Gemini app features, while a separate Workspace help page says some limits will start being enforced on May 1, 2026. (workspace.google.com) Anthropic has kept a subscription wrapper, but it is also segmenting access by intensity. Its Claude Max plan offers “up to 20x more usage per session than Pro,” and the company’s Opus 4.6 page says the premium model is priced in the application programming interface at $5 per million input tokens and $25 per million output tokens, with lower prices available through caching and batch processing. (anthropic.com) That split reflects how these systems are built. A simple chatbot reply, a long-running coding agent, a deep-reasoning query, and a high-resolution video generation all consume different amounts of graphics processing unit time, memory, and model inference, so vendors are increasingly charging by task, token, or credit instead of pretending every user costs the same. (anthropic.com) The low flat-fee era was always under pressure from those compute-heavy features. When OpenAI introduced ChatGPT Pro on December 5, 2024, it said more advanced capabilities “take significantly more compute” and priced the plan at $200 a month for scaled access to its most capable models and tools. (openai.com) The result is not the end of subscriptions so much as a new stack on top of them: a base fee for access, then limits, credits, or overages when heavier users push deeper into the system. The more artificial intelligence products look like metered infrastructure, the harder it becomes for vendors to keep selling every workload at one simple sticker price. (help.openai.com)
