Artificial intelligence has crossed the line from private technology into sovereign infrastructure, and the fight now beginning in Washington is not about whether AI will be powerful, but who will own the power it creates.

That is the new argument forming around the most important companies in the world. It is no longer a debate about whether chatbots can write emails, produce code, summarise documents or generate images. That phase already feels old. The larger question is whether the next economic order will be owned almost entirely by private capital, partly claimed by the public, or governed through some uneasy hybrid of markets, state power and democratic demand.

The trigger was Anthropic.

On 1 June 2026, the company behind Claude announced that it had confidentially submitted a draft registration statement to the Securities and Exchange Commission for a proposed initial public offering. The statement was cautious. No share count. No price range. No guarantee of listing. The offering would depend on regulatory review, market conditions and other factors.

But the significance was obvious. One of the frontier artificial intelligence laboratories was preparing to cross from private strategic actor into public market asset. Once that happens, artificial intelligence stops being merely a technology story. It becomes a capital formation story. It becomes a stock market story. It becomes a question of who gets to own the infrastructure of cognition.

Anthropic is not alone. OpenAI has already become one of the most valuable private companies in the world. The frontier AI sector is no longer populated by speculative start ups hoping to find product market fit. It is being built around companies whose valuations increasingly resemble national economic instruments. They require vast quantities of capital, electricity, chips, data centres, talent and political permission. They compete not only for customers but for access to the basic machinery of modern power.

These firms do not simply sell software. They build systems that may sit inside search, education, coding, law, medicine, defence, intelligence, public administration and the labour market itself. They are becoming a layer between human intention and institutional action. Once that layer becomes normal, the question is no longer whether people use AI. It is whether society can function without it.

The evidence of that shift is now visible in Washington.

President Donald Trump’s executive order on advanced artificial intelligence, signed in early June, did not nationalise AI. It did not require government permission before frontier models could be released. It did not create a formal licensing regime. Those caveats matter. The order explicitly frames the new process as voluntary.

But politically, the order is still important. It treats the most advanced AI models as national security objects. It directs federal agencies to create classified benchmarking processes and develop a framework through which developers may give the government access to covered frontier models before wider release. The stated justification is innovation, security, cyber defence, intellectual property protection and American strategic advantage.

A private product cycle is becoming a matter of national statecraft. The government is not building the models, but it is beginning to inspect them, secure them, measure them and fold them into the logic of national power. That is how infrastructure becomes sovereign before it becomes public. Not through formal ownership at first, but through dependency, security classification, strategic necessity and political leverage.

Then came Bernie Sanders.

His proposal is crude, radical and politically explosive. It is also clarifying. Sanders has argued that the public should own half of the major American AI companies. His proposed American A.I. Sovereign Wealth Fund Act would create a sovereign wealth fund through a one time 50 percent tax, paid not on profits but in stock, by the largest AI companies. He argues that because AI is built on collective human knowledge, the wealth generated by it should benefit the public rather than simply enrich billionaires and investors.

The instinct behind the proposal is not difficult to understand. AI models are trained on vast bodies of human production: books, articles, code, images, websites, conversation, culture, research and institutional knowledge. The infrastructure they require depends on energy systems, planning decisions, public tolerance, water use, grid expansion and national industrial policy. Their security is increasingly treated as a state interest. Their economic impact is expected to fall heavily on workers whose jobs are transformed, downgraded or displaced.

In that context, the old model looks increasingly strained. Private investors fund the firms. Private firms own the models. Private shareholders capture the upside. The public supplies the culture, absorbs the disruption, funds much of the basic science, grants the permits, maintains the legal order, secures the grid and then waits to see whether any of the gains are shared through wages, prices, taxes or philanthropy.

That bargain may hold for normal companies. It becomes more unstable when the company is building something closer to a general purpose economic layer.

OpenAI’s scale explains why the issue is now combustible. ChatGPT has reportedly crossed one billion monthly active app users globally. That is not the adoption curve of a niche tool. It is the adoption curve of mass infrastructure. Workers use it. Students use it. Companies use it. Lawyers, doctors, programmers, journalists, teachers and civil servants use it, or are preparing to be judged against those who do.

The public is not encountering AI as an abstract technology. It is encountering it as pressure. Pressure on work. Pressure on wages. Pressure on skill. Pressure on education. Pressure on credibility. Pressure on what counts as human expertise.

If AI merely made better office software, the political question might remain narrow. But if AI reorganises labour, captures economic value, concentrates market power and becomes embedded in state capacity, then ownership becomes a constitutional question in economic form. Who owns the machines that mediate the next economy? Who owns the profits? Who decides the rules? Who bears the losses?

There is, however, a serious counterargument. It should not be dismissed.

Sanders’s 50 percent proposal is not simply a public dividend. It is partial nationalisation by another name. A stock tax of that scale, combined with voting rights and board representation, would give the federal government direct influence over private AI firms. That may sound democratic in theory. In practice, it could create a different kind of concentration.

State ownership does not automatically mean public control. It may mean political control. It may mean congressional bargaining, regulatory favouritism, industrial patronage and pressure on companies to align with whichever administration holds office. It could deter investment, punish risk taking after the fact and encourage frontier firms to structure themselves offshore. It could also entrench the very giants it claims to discipline, because only the largest companies would have the legal, financial and political machinery to negotiate with the state.

There is another danger. If the government owns shares in AI companies, does it regulate them more strictly because the public has an interest, or more gently because the Treasury has one? Does public equity reduce corporate power, or fuse corporate power with state power? Does it democratise AI, or create a permanent alliance between Washington and the dominant laboratories?

That is the hardest problem for the Sanders position. It identifies a real injustice but offers a remedy that may produce a new one.

The same warning applies to Trump’s version of the question. A small public stake in AI giants may sound pragmatic. It may even be sold as patriotic capitalism: private innovation, public benefit, American leadership. But the danger is that public ownership becomes a decorative phrase for state sanctioned monopoly. The government takes a symbolic stake. The companies receive legitimacy. Competitors face a higher wall. The public gets a slogan while the structure of power remains intact.

If AI increases productivity and lowers costs, private ownership may look efficient. If AI destroys jobs, hollows out professions and concentrates income, private ownership will look predatory. If AI becomes essential to national defence, private ownership will look strategically fragile. If AI becomes essential to public services, private ownership will look democratically incomplete. If AI firms become too central to fail, they will start to resemble banks before the crash: private in profit, public in consequence.

The companies understand this. So does Washington. So do investors.

Anthropic’s IPO filing matters because it marks the financial packaging of frontier AI for the market. Trump’s order matters because it marks the entry of the state into the security logic of frontier models. Sanders’s proposal matters because it marks the arrival of an explicit public ownership claim. OpenAI’s scale matters because it proves that the technology is already embedded in daily life at a planetary level.

AI is no longer waiting for society to decide what it is. It is being capitalised, deployed, secured and normalised before the political settlement has been written.

The first winners are already being built. The valuations are already being imagined. The infrastructure is already being laid. The state is already moving closer. The wealth claims are already forming. The dispute has simply arrived late.

The fight over AI is therefore no longer about chatbots. It is about whether the next economy will be treated as a private enclosure, a public asset, or a contested zone between capital, state power and society itself.

That is the question now facing Washington. Not whether artificial intelligence will change the world, but whether the world it creates will belong to the few who financed it, the state that protects it, or the public whose knowledge made it possible.