Sam Altman offers a Trojan Horse to American taxpayers

Sam Altman just offered the US – meaning every American – US$42.6 billion (S$55 billion). (Not each, sadly). He said giving the government a 5 per cent stake in the company he runs, OpenAI, is the best way to ensure that Americans share in the promised bounty from artificial intelligence.

Before the Trump administration takes him up on the offer, it should organise a group trip to see Christopher Nolan’s new movie The Odyssey, whose opening act is the most famous gift in Western literature: a giant wooden horse, wheeled through the gates of Troy by the people it destroyed. The lesson translates. Beware of CEOs bearing gifts.

Start-up CEOs usually guard their equity with a fierceness that lions standing over a fresh kill would envy. So why was this offer made?

That question was answered 55 years ago in a 1971 paper by the economist George Stigler, who would later win a Nobel prize in economics. He described the theory of regulatory capture, which is the argument that regulation tends over time to be acquired by the industry it governs and operated for that industry’s benefit.

While capture tends to affect all regulatory regimes, it’s far more likely when the regulations were requested by industry in the first place. Capture is a tendency rather than a law, and economists still argue over how far it reaches. The incentive it names, though, only runs one way.

OpenAI’s “gift” is a way to capture not just regulators, but the government itself. Equity makes the government an owner. So the same government that is dictating which models OpenAI can ship would suddenly have an economic stake in the company’s success. That stake, like all government investments, would become a political football.

In the 2012 election, the Obama administration’s failed loan guarantee to Solyndra, a solar-cell maker, became a major line of attack by Republicans. The same Energy Department loan office also backed Tesla, now one of the most valuable companies in the world. But in political terms, only the failure mattered. And those were loans – finite things that get repaid.

A permanent equity stake in OpenAI would be a permanent tie to a single company, and would put enormous pressure on politicians going into any electoral cycle to keep the value of its equity up. And in America, major elections happen every two years. Regulator, gatekeeper, and shareholder would all collapse into a single party, but only one of those roles could be assessed every day by anyone with a phone.

The latest shifts in the economics of AI suggest why OpenAI might be willing to give up 5 per cent of its equity in the hope of acquiring a government backstop.

Leaked audited statements reported by the technology writer Ed Zitron and verified by the Financial Times show that the company lost roughly US$21 billion from operations in 2025 on about US$13 billion in revenue, against US$1.4 trillion in computing commitments over the next eight years.

Its public offering has slipped towards 2027, which means the US$852 billion valuation behind that $42.6 billion still rests on its last funding round, a number set by OpenAI’s investors and no public market has tested. And Oracle, which is building much of the company’s future capacity, just had its worst week since the dot.com bust of 2001 on fears that its marquee customer might not be able to meet its future commitments. A company in that position has every reason to tie the government to its fate. Insurance is at its cheapest if you buy it before you need it.

If 5 per cent of OpenAI’s equity in government hands is a bad idea, then 50 per cent is an even worse one. In June, Senator Bernie Sanders of Vermont introduced a bill to move half of the largest AI firms into public hands through a one-time tax paid in their own stock. There’s a reason the government taxes companies on their profits rather than by taking their equity, and the recent exceptions, like its stake in Intel, were struck one company at a time, not levied on a whole industry. Holding equity in every AI firm and the government acquires a unique incentive to favour that industry over others.

And this is a sector that hardly needs favours. AI-related capital spending has grown so large that in the first half of 2025 it added more to America’s economic growth than consumers did. The public interest is not served by tilting the field even further towards an industry already distorting the rest of the economy.

This insurance might also have costs for the companies buying it that their leaders haven’t considered. A company partly owned by the US government is, to everyone outside America, an arm of the government. Before the Trump administration’s new approach to foreign policy, that might have been fine. US allies were willing to trust their security to the American nuclear umbrella. Compared to that, relying on American technology was a small risk.

They no longer feel that way, and it’s started to affect what they buy. In June, France’s domestic intelligence service moved to replace Palantir Technologies with a domestic alternative, and Prime Minister Sebastien Lecornu said why in plain terms: A country cannot rely on a partner capable of “turning off the access tap”. He was pointing at Washington’s decision, days earlier, to cut foreign users off from Anthropic’s most advanced AI models.

Germany’s domestic spies and Britain’s health service are edging the same way. When capable Chinese models that can be run on your own servers are available at a fraction of the price of the leading American models, other countries will have every reason not to tie themselves to the American technology stack.

So let’s keep Washington’s gates closed to this particular gift. These companies should stand or fall based on their skills in the marketplace, not on Capitol Hill. BLOOMBERG