Somewhere between the race to build artificial intelligence and the call to regulate it, the competitors began helping design the track.
When the most powerful companies in an industry ask government to restrain that industry, the first question is not whether their warnings frighten us. It should be: who benefits from the restraint?
Investor Michael Burry, who saw the 2008 housing collapse before most of Wall Street, has raised exactly that question. He calls the recent chorus of AI warnings self-serving “hype and puffery” and asks why companies racing to build ever more powerful systems would suddenly want the government to regulate them.1 He may be right, or he may not. Either way, the warnings are only half the story. The rules written in response will shape not only what AI becomes but also who gets to build it.
That does not prove the warnings are false. AI is advancing rapidly, and credible researchers have raised serious questions about misuse, cybersecurity, biological risk, autonomous systems, and the ability to control ever more capable models. Those concerns deserve examination on their merits. So do the incentives of those raising them. A real danger and a commercial advantage can coexist.
Why everyone wants guardrails now
Artificial intelligence did not become powerful last week. The acceleration has been visible for years. Generative AI reached the public in late 2022, and the largest technology companies have since poured tens of billions of dollars into models, chips, cloud capacity, data centers, and talent.
Yet September 2026 brought a sudden concentration of public warnings. Anthropic CEO Dario Amodei called for stronger oversight of frontier systems. OpenAI’s Sam Altman, Google DeepMind’s Demis Hassabis, and Elon Musk backed stronger guardrails or a slower pace on some development. President Donald Trump rejected the push, arguing that new restrictions could weaken the United States against China.2
The debate is unfolding just before the November 2026 midterms, so its timing invites scrutiny. However, timing alone is not evidence of coordination.
The better question is what the proposed rules would actually do. Who could afford to comply, and who could not? Would regulation restrain the companies already at the frontier, or make it harder for the next competitor to catch up? Those questions move us from motive to something we can examine: incentives and consequences.
We have seen this playbook before
Big companies asking for regulation is an old tactic, not a confession. Regulation is not the enemy of concentrated power. The right statute can preserve rivalry; the wrong one can price competitors out. Antitrust history records attempts to distinguish those two statutes.
Railroads became the first federally regulated industry under the Interstate Commerce Act of 1887, after their control of interstate commerce gave them enormous economic power. Congress required their charges to be “reasonable and just” and banned discriminatory rates.3 Three years later, the Sherman Antitrust Act made combinations that restrain interstate commerce unlawful and prohibited monopolization and conspiracies to monopolize.4
Then came Standard Oil, American Tobacco, the Clayton Act, the Federal Trade Commission, and eventually AT&T's breakup. The details differed; the problem recurred. What happens when economic power becomes so concentrated that those already inside can control who gets through the door? The Clayton Act gave the government more power to challenge mergers that could substantially reduce competition, and Congress created the FTC in 1914 to police unfair methods of competition.56
The law does not make being big illegal, and a few enormous competitors do not automatically create an unlawful oligopoly. That’s important.
A monopoly involves a firm with market power that, for Sherman Act liability, maintains that power through unlawful exclusionary conduct. An oligopoly is a market dominated by a small number of firms. Concentration alone does not establish an antitrust violation.
The question is what those firms do with their position. That brings us back to AI.
The cost of entry is already high
You do not have to speculate about concentration in artificial intelligence. The federal government has already examined it. In January 2025, the Federal Trade Commission issued an inquiry into the relationships among major cloud providers and leading AI developers, including Microsoft-OpenAI, Amazon-Anthropic, and Google-Anthropic.7
The findings deserve far more attention than they have received. The FTC identified more than $20 billion in cumulative investments across those partnerships. It described billions in cloud commitments, equity stakes, discounted computing, access to technical and business information, and certain control and exclusivity rights.8 The agency did not call these arrangements illegal, but it did identify possible competitive consequences: less access to computing power and engineering talent, higher switching costs, and advantages that flow from information rivals cannot see.
Consider the many young AI companies trying to win a place in that market. Building a frontier model already demands staggering capital, computing power, chips, electricity, talent, and data. Now add a regulatory structure so expensive that only firms with armies of lawyers, compliance officers, and government-relations teams can satisfy it.
The regulation may restrain the giant. It may also become the giant’s moat. That is the possibility Burry’s argument puts on the table.
Who benefits when the rules get expensive?
This is where the debate reaches people who will never build a large language model. Competition matters more than corporate valuations. It shapes which products reach us, their cost, the choices we have, the privacy terms we accept, and how easily we can switch providers. When competition thins, consumers feel it directly through fewer options and higher prices. The FTC describes the purpose of antitrust law as protecting the competitive process for consumers, preserving incentives to become more efficient while keeping prices down and quality up.9
The AI market poses two risks that should not be confused. One is under-regulation, in which genuinely dangerous capabilities develop without safeguards. The other is incumbent-protecting regulation, in which rules meant to control powerful technology raise the cost of entry so high that only today’s largest companies can comply.
Both shrink human agency.
The first leaves people exposed to technology they cannot understand or control. The second leaves them dependent on a small group of institutions that hold the keys to that technology.
That is why both “regulate AI” and “do not regulate AI” are inadequate positions. The real questions are harder. Regulate what? Against which proven harm? At what cost, and who bears it? Who gains the advantage? And does the rule open the market or close the gate?
Follow the incentives, not just the warnings
The most revealing clue comes from the industry next door. A federal court found that Google unlawfully maintained monopolies in general search and search advertising. In 2025, it barred certain exclusive distribution deals and ordered Google to share specified data and syndication access to help rivals compete.10 A separate federal case found that Google unlawfully monopolized key advertising-technology markets, and a court ordered additional remedies in September 2026.11
Google’s conduct proves nothing about Anthropic, OpenAI, Microsoft, Amazon, or anyone’s current AI proposals. It establishes something more basic: digital markets are not immune to the longstanding problem of gatekeeping power.
Technology changes. The economic temptation does not.
Safety has a blind spot
An ancient story captures this blind spot. In 2 Samuel 12, the prophet Nathan comes to King David with a case for judgment. A rich man with large flocks takes the one beloved lamb of a poor man rather than give up one of his own. David burns with anger and declares that the man who did this deserves to die.
Then Nathan turns the story around.
“You are that man.”
David had condemned himself without realizing it. Injustice is easy to see when someone else commits it. That is the trap in every debate like this one.
For more than a century, Americans have looked back on railroad barons, oil trusts, telephone monopolies, and search gatekeepers and understood why concentrated power deserved scrutiny. It is obvious in hindsight. It is obvious when the power belongs to someone else.
The names have changed. Today, the critical inputs are compute, chips, energy, models, talent, capital, and access. Some of the companies with the greatest command of those resources are also helping shape the rules that will govern everyone who comes next.
That does not make them guilty of monopolization. It does not make every AI warning false. And it does not mean government should ignore real risks. It means we should bring the same scrutiny to the story unfolding in front of us that comes so easily when we encounter it as history.
The question isn’t whether AI needs guardrails. It’s whether guardrails shaped by today’s leaders protect us from danger—or protect today’s leaders from tomorrow’s competition.
The hardest concentration of power to see may be the one forming while our attention is fixed somewhere else. Nathan’s accusation still confronts every generation—especially the generation certain it would have recognized the danger sooner.
What if you are looking at that man?