The AI Endgame is Heavy Industry

It’s funny how AI is eating the moats of the very companies spending billions to create it. For years, cloud lock-in was real and painful. Migrating a complex web of services from AWS to Google Cloud was a project measured in months and anguish. Now, you can just tell an LLM to handle it. The friction is vanishing, taking the stickiness of the platforms themselves with it.

The software layer is becoming as portable as a shell script, but the physical layer is doing the opposite. Our era’s megaprojects are data centers full of GPUs that depreciate to zero in about six years. The hyperscalers are in a frantic arms race, spending sums that dwarf the Apollo program on hardware that’s obsolete by the time the next keynote starts.

The recent news of xAI pushing for a "Terafab" reveals the brutal endgame of this logic. When software is instantly commoditized, the only defensible moats left are heavy industry—it resides in the physical chips, massive tracts of land, and dedicated power generation.