The speaker detailed how AI inference for leading models (OpenAI's GPT, Anthropic's Claude) is massively loss-making under current subscription plans, with power users burning thousands of dollars in compute on $200/month plans. He explicitly compared the funding frenzy and unsustainabile economics to the dot-com bubble. The fundamental business model is broken because competitive pressure forces models to burn exponentially more tokens for useful outputs, eroding hardware efficiency gains. The path to profitability via per-token pricing would crater demand and is untested. WATCH because the setup for a major sector dislocation is clear, but the timing of a bust is uncertain (akin to the NASDAQ doubling after 1998 before crashing in 2000). The upcoming IPOs of OpenAI and Anthropic could be pivotal events. A breakthrough in inference efficiency or a massive, sustained subsidy from vendors/governments could prolong the unsustainable model, deferring the reckoning.