The singularity centralizes and decentralizes at the same time
The core tension of the AI age is that centralization and decentralization are happening on the same clock. Frontier training keeps getting bigger and rarer, while useful intelligence keeps leaking outward to laptops, edge devices, and small teams.
Why the frontier stays centralized
Training a frontier model requires giant pools of compute, energy, capital, talent, and coordination. These inputs do not fit inside a garage. They fit inside sovereign wealth funds, hyperscalers, and a handful of national champions.
That gravity is not going away. The most capable systems of the next decade will be built by a small number of organizations that can assemble those inputs at scale.
Why useful intelligence spreads
Inference costs fall on a curve that makes yesterday's frontier tomorrow's default. Open and open-weight models keep landing months behind the closed frontier, on hardware ordinary people can afford. Agents, portable workflows, and local runtimes let a single operator do work that used to require a team.
The result is a strange topology: a few enormous towers at the top and an ever-widening plain of capable systems everywhere else.
The test that actually matters
Centralization is not inherently the enemy, and decentralization is not automatically liberating. A well-run centralized system can serve billions. A poorly designed decentralized system can recreate every old gatekeeper with a new interface.
The real test is whether people retain choice, ownership, portability, auditability, and the ability to exit. Any system that fails those tests, however open its branding, is not on the side of the sovereign singularity.
"Centralization is not the enemy and decentralization is not the cure. The test is whether people can choose, own, audit, and exit."
Where this shows up