Infrastructure Chokepoints: The Real Leverage Points Over Frontier AI
This page is a structured working draft — real analysis, not yet expanded with the full expert sourcing given to the flagship pages. Safe to build on; treat specifics as provisional until sourced.
Why infrastructure, not confrontation
Frontier AI depends on a small number of physical chokepoints: advanced chip fabrication (concentrated in a handful of facilities worldwide), the cloud data centers that house training runs, and the electrical grids that power them. This concentration is a systemic risk in normal times, but it’s also the most tractable control point available if a coordinated slowdown or shutdown were ever genuinely required.
What responsible practice looks like today
Organizations deploying advanced AI systems should maintain tested, regularly exercised rollback and shutdown procedures — not a policy document that’s never been run in practice, but infrastructure drilled the way disaster-recovery failover is drilled in other critical industries. A killswitch that’s never been tested is not meaningfully different from having none.
The societal-level version
Critical infrastructure — power, water, financial settlement — that can fall back to a degraded, human-operated mode functions as a societal circuit breaker: it means a worst-case AI-driven disruption to automated systems doesn’t cascade into a total loss of essential services.
Why this is fighting back, not just defense
Controlling and hardening chokepoints is the collective-action equivalent of the household hardening covered in Protecting Your Family — it doesn’t require confronting a system directly, only ensuring humans retain a physical, verifiable path back to control if oversight ever meaningfully fails.