In a move that could fundamentally redefine the economics of artificial intelligence, OpenAI is reportedly preparing to phase out its traditional “all-you-can-eat” subscription model. According to internal sources and recent public statements from leadership on March 18, 2026, the company plans to transition ChatGPT toward usage-based pricing, effectively treating AI intelligence as a metered utility like electricity or water.
The “Unlimited” Problem
The shift comes as AI models become exponentially more resource-intensive to run. Nick Turley, VP and Head of ChatGPT, recently compared unlimited AI access to “unlimited electricity,” suggesting that as compute costs for flagship models like GPT-5.4 Pro soar, flat-rate subscriptions are no longer sustainable.
- Rising Compute Costs: Each “Thinking” query or “Deep Research” task consumes significantly more GPU cycles than the chat interactions of 2024.
- The “Power User” Gap: Currently, a user paying $20/month might use $2 worth of compute or $2,000, creating a massive imbalance in the company’s margins.
- Industry Trend: Microsoft and Anthropic have already begun implementing “pay-as-you-go” or “per-agent” billing structures for their most advanced tiers.
Sam Altman’s Vision: “Intelligence by the Meter”
Speaking at the BlackRock US Infrastructure Summit in Washington, D.C., on March 11, 2026, OpenAI CEO Sam Altman doubled down on the idea of “metered” intelligence.
“We see a future where intelligence is a utility… and people buy it from us on a meter, and use it for whatever they want to use it for,” Altman stated.
Under this proposed model, users wouldn’t pay a flat $20 fee. Instead, their monthly bill would be calculated based on:
- Model Grade: Queries to “Nano” models would be nearly free, while “Pro” or “Thinking” models would cost more per token.
- Reasoning Effort: Users could choose to pay more for “Extended Thinking” for complex problems or opt for a cheaper, “Light” response.
- Task Complexity: Background agents running multi-step workflows would accrue costs based on their duration and the tools they utilize.
Current 2026 Pricing Tiers (The Transition Phase)
While the “metered” transition is expected to fully roll out by late 2026, OpenAI has already fragmented its plans to capture different segments of the market:
| Plan | Price (Monthly) | Best For |
| ChatGPT Free | $0 | Casual use (limited to GPT-5.4 Mini) |
| ChatGPT Go | $8 | Students & light users (10x more messages than Free) |
| ChatGPT Plus | $20 | The standard “all-rounder” (160 messages / 3 hours) |
| ChatGPT Pro | $200 | Power users & researchers (unlimited flagship access) |
Impact on Users
The proposed change has already drawn sharp criticism from consumer advocates, who argue that “metering” intelligence creates a “wealth gap in wisdom.” * The “Stupid” Tax: Critics suggest that if people are charged “per thought,” they may hesitate to use AI for learning or creative exploration, limiting the technology’s benefits to those with deep pockets.
- Predictability: For freelancers and small businesses, a fluctuating “utility bill” for AI makes monthly budgeting significantly more difficult compared to the current $20/month predictable expense.
Despite the rumors of metered billing, OpenAI has confirmed that it will always maintain a free tier to ensure basic AI accessibility, though it will likely be supported by the new ChatGPT Ads ecosystem launching later this spring.
