AI is for rich people
In a Nutshell
AI is becoming prohibitively expensive for frontier models, with prices rising rather than falling—high-performance tokens now cost significantly more than before, and the richest companies and individuals can afford usage levels that everyone else cannot. This creates a structural divide where top AI talent and compute cluster inside a few organizations like OpenAI and Anthropic, leaving outsiders and new graduates with little access or opportunity. The practical takeaway is to use cheap AI strategically for rapid experimentation and iteration, but accept that meaningful scale remains gated behind wealth and institutional affiliation.
These notes were generated by AI and may contain inaccuracies.
Peter Steinberger, the inventor of Open Claw who works at OpenAI, shared a screenshot of his dashboard showing his token consumption. In the last 7 days, he spent $250,000 on tokens. In the last 30 days, he spent $1.3 million on tokens. This spending occurred during a science experiment where he explored how software would be built in the future if tokens didn't matter.
The notion of building technology without considering costs is compared to a crack dealer asking how society would look if crack were free. Similar logic is applied to housing construction—if wood didn't matter, the housing crisis would supposedly be solved. The same flawed reasoning is used regarding Elon Musk's rocket ships: if gravity didn't matter, the cost problem would disappear.
The common belief that AI tokens will follow a Moore's law trajectory and become free is identified as one of the biggest deceptions in the industry. Two separate pricing charts exist: lower-end models have seen prices decrease or stay roughly the same, but the high-performance models required for meaningful work have seen prices increase significantly.
The price of tokens for advanced models has been rising. From GPT 5.4 to 5.5, token prices doubled. With Anthropic's Opus 4.7, the price remained the same as 4.6, but a new tokenizer was introduced that produces 35% longer responses, effectively increasing cost. OpenRouter conducted a study showing that GPT 5.5 costs 49 to 92% more than GPT 5.4.
Sign in to read the full notes
Get access to AI-generated notes, topic timestamps, and more.