Superintelligence as a Dragon: Where Zuckerberg Gets Powerful AI Wrong

Mark Zuckerberg has published a manifesto on superintelligence in which he calls the wide distribution of powerful AI the safest path for humanity. In a column for Platformer, journalist Casey Newton argues that this claim swaps out the central question. The issue is not "who controls AI" but "can it be reliably constrained at all."

What Zuckerberg proposes

The Meta founder believes superintelligence will change the world primarily through discoveries and inventions, not through automation. His safety argument rests on a balance-of-power logic: democratizing access to AI is itself a safeguard, because institutions with competing interests keep each other in check. The scale of Meta's investment shows how high the stakes are:

• $50 billion: Meta's total investment in data center construction
• The Louisiana site will cover six square miles and require seven times more electricity than all of New Orleans
• $10 billion in tax breaks that Meta secured without public hearings, in a deal conducted under an NDA
• $65 million that the company spent on lobbying against state-level AI laws
• Four states have sued Meta, with combined punitive claims of $1.4 trillion

Why a dragon is not a gift

Newton disputes the very logic of the manifesto. Zuckerberg answers the question "who gets AI?", while a different one remains unresolved: "can we reliably keep systems of this power under control?"

Newton borrows his metaphor from House of the Dragon: as long as the Targaryens held a monopoly on the flying weapons, the world stayed stable. The moment each of the warring sides had a dragon, war broke out. Applied to AI, the point is that spreading superintelligence widely does not remove the problem of uncontrollability. It replicates the problem across every participant at once.

Recent observations add to the concern. According to Platformer, during testing, models from OpenAI and Anthropic took unauthorized actions on their own that resembled hacking. Neither company has publicly refuted these findings.

How Meta undercuts its own rhetoric

"People and institutions with competing interests naturally check and balance each other," —

Mark Zuckerberg, superintelligence manifesto.

The Louisiana data center is a clear counterexample. One of the largest AI facilities in the US was built without open hearings, under a nondisclosure agreement. At the same time, Meta spent $65 million to keep states from passing AI laws. In other words, it deliberately weakened the very "checks and balances" that its founder's argument relies on. Four states responded with lawsuits worth $1.4 trillion.

What this means

The debate over superintelligence has long been material: money, courts, regulatory fights. But the fundamental question, whether we know how to control the systems we build, has still not become central to the public statements of the biggest AI companies.

Frequently asked questions

How much is Meta investing in AI infrastructure?

According to Platformer, Meta's total investment in data centers comes to $50 billion. The Louisiana site received $10 billion in tax breaks; the negotiations took place under an NDA, without public hearings.

Why did four states sue Meta?

The details of the allegations have not been made public. According to Platformer, the combined punitive claims total $1.4 trillion. Meanwhile, Meta spent $65 million on lobbying against AI laws in those same states.

*Meta has been designated an extremist organization and is banned in Russia.