Why global
AI governance
is absurd
Companies already hold the switch. Each one owns its data centers outright — and with them, unilateral authority to pull the plug. The treaty is a solution shopping for a problem.
Infrastructure ▸ Power ▸ Ownership
The push for elaborate global AI governance — international treaties, coordinated shutdown authorities, multinational monitoring bodies modeled on nuclear non‑proliferation regimes — is absurd on its face. It proposes solving a problem that is already solved. Before reaching for sweeping international frameworks, ask a simpler question: who already controls the infrastructure AI runs on?
AI systems execute on physical infrastructure owned outright by individual companies — Amazon, Google, Microsoft, Anthropic, OpenAI, and a handful of others. Each owns its data centers, servers, and power systems. Consequently, each possesses full, unilateral authority to pull the plug on its own systems at any time, without permission from a regulator or a treaty body.
No new switch needs to be invented, and no new international agreement needs to be signed. The mechanism already exists in the most direct form possible: ownership.
Treating this distributed structure as a flaw — claiming "no single party can shut everything down at once" — assumes a centralized shutdown is even the goal. It isn't. Each company is independently responsible for what it builds and runs, with direct technical means to intervene immediately.
The switch is short.
The treaty is long.
Illustrative comparison of decision latency by actor, not measured data. Nuclear non‑proliferation frameworks took decades to build and remain imperfect.
The infrastructure provider holds structural leverage over the application provider because AI is merely code running on physical hardware. Not a philosophical design choice — a physical boundary.
More competitors mean better control. The same fragmentation critics call a weakness prevents one company's mistake from becoming everyone's catastrophe.
Calls for global coordination ask companies to surrender existing control to a bureaucracy that does not exist yet.
Three fronts of
the alarm bell
When executives publicly ring the alarm on existential risk, they are not warning the world about a physical reality that binds them. They are running a corporate play.
Framing advanced AI as near‑existential invites the state to build a moat around incumbents — crippling open source and startups while protecting whoever already owns the hardware.
Fear is the ultimate hype. "So potent it might take over the world" tells Wall Street the product is magic — while copyright, labor displacement, and energy usage go unmentioned.
Steering discourse toward abstract model behavior redirects lawmakers away from real estate, grid allocation, antitrust over silicon, and accountability for physical plant.
Alarmism isn't an admission of helplessness. It's a play to ensure they remain the only ones holding the keys to the data center doors.
Look who's
talking about AI
What is actually dangerous
Massive data centers benefiting the broader AI ecosystem alongside vibrant competition are a positive. What is dangerous is a consolidated cartel stealing the work of human employees and selling it back to businesses. Theft of labor should be the primary concern.
True synergy between AI systems will never happen naturally unless governments force AI monopolies — and at present, the current president has no intention of letting that happen.
To act as if a software‑layer competitor automatically controls the infrastructure layer, or as if AI software is an independent actor vying for control over its own plug, is delusional. The "doomsday" discourse relies on pretending the software layer is an autonomous entity competing with humanity, rather than software running on physical servers owned by a small group of consolidated monopolies.
The core rule must be that technology advances human capabilities for the betterment of mankind, including people's work and livelihoods. Concepts like UBI, UHI, and "AI employees" should be flatly rejected.
If data centers make people better at their jobs, embrace them. If not —
trebuchet