The end of cheap unlimited AI

• Big Tech is quietly capping access to its best AI models because running them loses money.
• OpenAI's newest models jumped from open access to just 50 messages a week, then near-unlimited use moved behind a $200/month tier — 10 times the old price.
• The cause is "tokenmaxxing": models now think longer and burn far more computing power per answer, so costs keep climbing.
• Throwing more compute at models stopped paying off — a 10x jump now buys only 10–15% better results, and the human-written text needed to train them is running out by around 2028.
• Power is the next wall — data centers in Virginia now need as much electricity as nuclear plants, and the grid can't connect them for years.

Outlook: The best AI is heading toward pricey enterprise and premium tiers, while cheap subscribers get the weaker models — universal, unlimited AI is fading.