Ownership concentration × Governance coordination

The fight doesn't stay on ownership — it migrates immediately to allocation. Once shared rents become a meaningful income stream, whoever controls the disbursement formula becomes the most powerful institution in history: not by owning the rails but by deciding who gets how much. Coordinated governance also requires verification at scale, which builds a surveillance apparatus owned by everyone and therefore dismantlable by no one. And safe mainstream systems push frontier experimentation to whatever grey zone coordination can't reach — which is where the actual research agenda gets set. The gains are genuinely shared; the infrastructure is genuinely creepy.

Incumbents lobby sincerely for heavy safety rules, because compliance cost is the best barrier to entry ever invented. The solution and the problem don't fight — they fuse into a single self-stabilising structure in which good governance strengthens feudal capture. Citizens stop being customers with exit options and become wards with access tiers; regulators and owners converge in personnel and crisis vocabulary until international safety standards function as cartel treaties, allocating markets under the language of public benefit. What makes the corner durable is that it isn't a lie: some catastrophic risks genuinely are easier to monitor through a small number of actors.

Millions of small stakeholders each have marginal upside from deploying harder, and there's no chokepoint with both the power and the standing to halt — diffuse ownership means diffuse responsibility, and the race accelerates precisely because nothing brakes it. Standards wars become identity wars: populations running on incompatible, differently-tuned models drift into incompatible perceived realities faster than under concentrated ownership, where one dominant model at least imposes a shared baseline. Insurers end up the de facto global regime — pricing some tail risks and excluding the uninsurable, which is precisely where the genuinely dangerous things live.

Competition converges on the same architectures, datasets, and chip supply lines, because that is what wins the race — so the world looks maximally competitive while accumulating maximally correlated failure modes. Concentrated labs without coordination become quasi-sovereign actors; states host them like client kingdoms, trading sovereignty for capability. The corner is also unstable: catastrophe is the only coordinator it responds to, disaster triggers belated rules the owners help write, and the whole configuration drains toward the Safety Cartel. The chaos is the last reservoir of optionality — and it closes fast.