Infrastructure runs on hyperscaler servers. In a crisis, the hyperscaler cannot order new GPU servers — TSMC is down. It maintains existing capacity but cannot extend it. If the organisation needs to scale, it hits rationing. It is in the queue — after the hyperscaler's own needs, after Enterprise Agreement customers. Without guaranteed contractual priority, it waits.

It also cannot repatriate its infrastructure — it no longer has the skills (documented in The Newspeak), the engineers (−34% Linux jobs), and migration takes 6 to 12 months (documented in The Accompanied Exodus).

Infrastructure runs on physical servers it owns. These servers were purchased and delivered before the crisis. Their chips are there, functional, independent of future production. In a crisis, it can maintain normal activity. It can even purchase additional Refurbished on the secondary market — whose value rises with chip scarcity but remains available since the world's existing server park is massive. It is outside the cloud rationing queue.

Core infrastructure runs on its own servers — critical services, sensitive data, baseline workloads. Cloud is used for temporary peaks or non-critical services. In a crisis, it can rebalance cloud workloads to physical infrastructure. It is not entirely dependent on the vendor's queue. This is the model documented in "The Sovereign Interface" — the Anti-Corruption Layer enabling switching without application rewrite.