AI Bubble or Not? Why Token Costs Aren't What You Think

Subscribe to the FullStack HR newsletter

An impromptu between-episodes take on the AI bubble panic. Johannes cuts through the headlines about rising token costs, Microsoft cutting Claude Code, and Uber's COO questioning AI ROI, and explains what's signal and what's noise.

In this episode:- Why nobody can actually call a bubble, and what the market is and isn't telling us right now- The arguments for a bubble (circular money between NVIDIA and OpenAI) versus against it (Anthropic's first profitable quarter)- Why "rising token costs" is a mislabel. Per-token prices are flat or lower, total spend is what's climbing- How Moore's Law underpins the whole thing: more powerful models for the same or lower price- The full context on Uber blowing its AI budget, and why the COO is far more nuanced than the headlines suggest- Why early subsidize-then-correct cycles are normal, same as railroads, electricity, and the internet- How to do the token economics yourself: if an agent solves a problem cheaper than a human at equal quality, you have your answer- Why the model companies are targeting a ~$58 trillion global labor market, and what that means for their valuations

Links mentioned:- Rapid Response podcast, Andrew Macdonald (Uber COO) interview - Fortune coverage of the Uber AI budget story