Microsoft plans $329B data center leases

The author argues that the massive data center buildout for artificial intelligence is driven primarily by monetization and renting models rather than technical necessity. While training frontier models requires large data centers, most inference workloads can run efficiently on local hardware like modern laptops. Furthermore, the author notes that major outages at AI labs stem from software failures rather than memory exhaustion, suggesting that future infrastructure spending is disproportionate to actual demand. Open article →