Blocs, Hubs, and Choke Points

We can now assemble the full picture. Three forces have been described: the strategic value of compute that makes it worth controlling, the export controls that weaponize the choke points, and the national strategies driven by sovereignty and constrained by energy. When these forces act together on the physical world, they redraw the map of where AI infrastructure is built and who is allowed to reach it. The old logic of globalization, in which production flowed frictionlessly to wherever costs were lowest, is being overwritten by a new logic in which compute flows to wherever it is permitted, powered, and politically safe.

The most visible result is the emergence of compute blocs. On one side sits the United States and its network of allies, who together control the most valuable choke points: the leading chip designs, the critical manufacturing equipment, and the most advanced foundries. Access to frontier compute within this bloc is coordinated and increasingly conditional, extended to friends and denied to rivals. On the other side sits China, pursuing self-sufficiency across the entire stack, building a parallel ecosystem that does not depend on the choke points it cannot control. Between and around these two poles sits everyone else, choosing sides, hedging, or trying to trade with both.

Within the favored bloc, compute is concentrating into a limited number of hubs, chosen not by cost but by a combination of political safety, energy availability, and strategic alignment. A region with abundant power, stable governance, and good standing with the controlling nation can attract enormous AI investment. A region lacking any of these, however cheap its labor or land, is passed over. This is why we see massive data center buildouts clustering in specific American states with available power, in Gulf nations offering energy and capital, and in a handful of allied countries deemed safe, while vast portions of the globe are simply not in the running for frontier compute at all.

Access, meanwhile, has become a lever of statecraft in its own right. The controlling nations do not merely deny chips to rivals; they ration access to the ambitious middle tier, using it to draw nations into their orbit. A country seeking to build sovereign AI capacity may find that the price of admission is aligning its broader technology policy, its security posture, or its stance on the rival power with the wishes of the nation that controls the chips. Compute access becomes a diplomatic currency, spent to build coalitions and withheld to punish drift. This is the interdependence lever applied not as a blunt embargo but as a graduated system of privileges.

The choke points remain the fulcrum of the entire arrangement, and their distribution determines the balance of power. The nation or coalition that controls a choke point holds a lever over everyone downstream of it. This is why the deepest choke points, advanced lithography above all, are the most strategically sensitive assets on Earth, and why so much energy goes into controlling them, defending them, and, on the other side, into building the capability to route around them. A choke point successfully bypassed loses its power; a choke point that remains unique retains it. Much of the coming contest is a race between the effort to hold the choke points and the effort to render them obsolete.

There is a physical vulnerability layered atop the political one, and it centers on Taiwan. Because the most advanced chip manufacturing on Earth is concentrated on a single island subject to a territorial dispute with China, the entire global compute map has a single point of catastrophic fragility. A disruption there, whether through conflict, blockade, or accident, would reverberate through every layer of the stack and every bloc on the board. This concentration is the reason the subsidize playbook exists at all: the frantic effort to build alternative fabrication capacity elsewhere is, in large part, an insurance policy against the day the island choke point fails.

My read is that we are watching the birth of a compute geography every bit as consequential as the oil geography that shaped the twentieth century. Oil created petrostates, drew borders, spawned alliances, and triggered wars. Compute is beginning to do the same, but faster and with capability that compounds rather than depletes. The nations that host the hubs, control the choke points, and command the energy will hold a structural advantage in the defining technology of the century. The map is being drawn now, in subsidies and export rules and power contracts, and once the concrete is poured and the clusters are running, that geography will prove as durable and as contested as any that came before.