In his recent piece [*AI Subscriptions Get Short-Squeezed*](https://ethanding.substack.com/p/ai-subscriptions-get-short-squeezed), Ethan Ding delivers a sharp reality check on the economics of today’s AI startups. The dream was that models would get cheaper and margins would follow. The reality is the opposite.
Yes, token costs have dropped. But every time a new model is released, like ChatGPT 5, it becomes the default. Users are not choosing to upgrade. They are being pushed to the most expensive model with no real alternative. Companies are setting the top-tier model as the only choice while still pricing access like it is 2023.
At the same time, AI usage is changing. Simple chats are being replaced by long compute-heavy workflows. That $20 per month plan cannot support this kind of usage. Claude Code tried higher prices and smarter scaling, but even their $200 plan collapsed under the load.
Too many AI startups are jumping onto a fast ship without checking if it can stay afloat. This is not the chip race of the 90s. Cost improvements are real, but usage is outpacing them fast. Without a shift to usage-based pricing or smarter business models, this entire category is on track for a margin collapse.
Credit to Ethan Ding for highlighting this clearly. Full article: [AI Subscriptions Get Short-Squeezed](https://ethanding.substack.com/p/ai-subscriptions-get-short-squeezed)
Labels: AI business models, AI startup strategy, Anthropic, ChatGPT 5, Claude, Ethan Ding, LLM pricing, OpenAI, subscription collapse, token costs