AI is protecting itself... and blocking us

Generative AI arrived for the end user wrapped in a clear message: open access, free of charge, and without barriers. For months, millions of people used these models without any visible restrictions. However, that phase has ended. Today, limits, usage windows, and temporary blocks are part of the normal operation of the major platforms. This is not a glitch: it is a change in the business model. And it affects every key player in the sector.

Gemini (Google): Usage based on compute and dynamic limits

Google has moved from counting messages to measuring compute. Consumption depends on the complexity of the prompt, the duration of the conversation, and the features used (image, video, advanced analysis). The limits act as a compute budget that replenishes overtime and also has a weekly cap. The goal is to maintain service stability and avoid bottlenecks in the most resource-intensive features.

ChatGPT (OpenAI): Usage windows and model-based limits

OpenAI applies usage windows that automatically renew after a certain period. Limits depend on the plan (Free, Plus, Team, Enterprise), the model (GPT-4o, GPT-4o mini, reasoning models), and system load. There is no public, fixed table with exact figures, but the reality is evident: the more powerful models have stricter restrictions, and access can be adjusted based on demand.

Claude (Anthropic): The most restrictive system… even in paid plans

Claude offers a free version with access to an advanced model, but with tight limits. Paid plans expand capacity and allow the use of more powerful models, though Anthropic does not publish specific numbers. The company acknowledges that there are limits per period, stricter restrictions on advanced models, and automatic adjustments when demand is high. Many analysts agree that it is one of the services with the most stringent controls.

Grok (xAI): Access conditioned by region and availability

Grok adjusts access based on user region, available model version, and system load. xAI does not publicly detail all limits, but it does admit that access can vary and that some features are rolled out progressively. In practice, this results in a service whose behavior is not always consistent across all users.

Why all AIs are imposing limits

The blocks are not a technical error, but a design decision. They respond to several factors:

• Limited capacity: Advanced models consume enormous GPU resources.
• More demanding models: Deep reasoning costs significantly more than a quick response.
• Abuse prevention: Unintended automated processes that could overwhelm the systems must be curbed.
• Regulation: Some models are subject to export restrictions or regional availability.

From “free and open” to “essential service”

When AI first reached the public, it was presented as an open tool without barriers. Today, limits are part of our daily routine. Sam Altman has argued that AI will eventually function as an essential service, comparable to water or electricity. But no essential service is unlimited or free.

And that is the shift: AI is protecting itself, regulating its consumption, and deciding how much each person can use. The question is no longer whether AI can provide an answer, but who controls access to that capacity. And that debate has only just begun.