The LLM Cost Illusion: How Scaling Killed the Flat-Rate Subscription

Many AI companies bet on the trend of LLM costs dropping 10x per year, assuming early losses would be offset by future high margins. Reality is different. While model costs are decreasing, user demand for the best models continues to grow, leading to an explosion in compute usage. The length of responses from models like ChatGPT has dramatically increased, resulting in exponential growth in token consumption. This means that even with cost reductions, overall spending far exceeds expectations. The article analyzes three counter-strategies: usage-based pricing from day one, creating insane switching costs for high margins, and vertical integration to profit from infrastructure. The author concludes that sticking to a flat-rate subscription model will ultimately lead to bankruptcy.