The people who cite BuTWhAt AbOuT AmAzOn?? generally forget that the magnitude of their losses were close to their revenues (not a 10x CapEx gap like we’re seeing in many of the hyperscalers), that AWS was essentially repackaging and selling cloud infrastructure they already had to build for their own core business, and that it became profitable much quicker than many people remember.

One of the many bummers of this situation is if (when?) the bubble deflates, it will discredit AI more broadly, even though this represents a massive doubling down on one narrow type of architecture (transformer-based LLMs and generative AI) at the expense of many other useful techniques. In the hangover period, it will be hard to attract investment for more productive AI development, much like we stopped building houses after the 2007 mortgage crisis