all cities
Post-Geography2030 — ?·Distributed

The Compute Empire

The first center that is not a city.

Kernel

The plausible 2035 successor to every center in this archive is not a city but a distributed infrastructure: 200,000-GPU training clusters in Phoenix, Memphis, Inner Mongolia, the Gulf, Pori, the Hebrides; the agent-economy interfaces those clusters expose; the legal and capital structures that own them. If the 21st-century pattern holds — and it might not — the "center" of civilization for the next several decades may be the polity that owns the most compute.

§ 01

What it could look like

The Compute Empire is structurally different from every historical center: it is not a city, it has no commute, it has no marketplace, it has no agora. Its visible surface is regulatory and capital — which jurisdiction it incorporates in, which sovereign signed off on the next training run, which AI lab's safety case the regulator accepted. The cathedral-as-data-center essay applies: this is a center organized around producing artifacts, not around housing citizens.

§ 02

Where it might fail

Concentration risk. If compute centralizes around a small number of vendors and jurisdictions, the resulting fragility is comparable to the 17th-century Dutch peat industry — one shock and the substrate collapses. Decentralized alternatives (local inference, open weights, agent-to-agent commerce) may erode the empire faster than centralized infrastructure can extend. The history of every center in this archive suggests the empire phase ends.

Civilization OS · layers opened or extended
Energy
100 TWh+ class training clusters
Computational
GPU + TPU + future architectures
AI
Frontier models · agent economy
Ideological
Sovereign-AI as legitimacy claim