Excellent piece, but it does hinge on the idea that what we’re building is actual AI (still hotly debated), if it can be successfully commoditized like utilities (very unlikely given compute continuously evolves and the hardware cycles of datacenters relative to power plants or water filtration systems), and if such a healthy cycle - should it develop at all - survives the multiple decades needed to effectively provide ROI.

Which is an awful lot of “ifs” to be hedging a bet on, especially at nation-state scales.