Uncle Sam considers buying a seat on the Titanic

AI + ML

L’etat, c’est AI

OPINION The US government is reportedly weighing whether to take a financial stake in AI companies, which looks a bit like negotiating for a seat on the Titanic.

Neither OpenAI nor Anthropic, the marquee brands in US AI, are profitable yet. While Anthropic may be nearer to that point if its accounting survives scrutiny, OpenAI’s $1.4 trillion in financial commitments over the next eight years have been interpreted as a red flag for investors.

This raises (at least) two questions: Should the US government be picking winners? And should the US government be picking losers?

The first question appears already to have been decided. As noted by the US Council on Foreign Relations, since January 2025, the feds have invested $20.9 billion in sixteen deals that involve direct ownership. This represents a change from more hands-off financial arrangements involving grants, loans, and tax incentives.

The Department of Commerce, for example, has taken a 10 percent stake in Intel, once a symbol of American technical prowess and now a national security backstop. The Development Finance Corporation had invested in minerals, energy, and infrastructure. And the Department of Defense has undertaken at least seven similar deals.

Neoliberal US notions about competition and the separation of church, state, and private industry have succumbed to the new world disorder.

When economists from Harvard and Yale looked at the issue in a 2021 paper titled “The Dance Between Government and Private Investors: Public Entrepreneurial Finance around the Globe,” they were cautiously optimistic.

Authors Jessica Bai (Harvard), Shai Bernstein (Harvard), Abhishek Dev (Yale) and Josh Lerner (Harvard) looked at 755 entrepreneurial finance policies in 66 countries during the period from 1995 to 2019. They concluded that “government funding programs are associated with subsequent increases in innovation,” as measured by “top patents.”

They offered some caveats, such as the observation that “government programs frequently rely on private capital markets through capital matching requirements, where private capital groups are often allowed to invest in more preferential terms than the public funds.”

And they also noted that economists have long recommended government investment in response to market failures – areas where private funding has chosen not to invest, presumably due to the uncertainty of returns.

A recent example of that would be the US Commerce Department’s decision to invest $2 billion in quantum computing in exchange for a minority controlling stake in nine technology companies. Pure-play quantum computing companies like D-Wave, Quantinuum, IonQ, and Rigetti Computing are not making a profit. But concern that quantum computing might some day do meaningful computing not possible with classical computers is enough to keep the funds flowing for now.

The US government’s reported interest in AI companies might be interpreted in a similar light, as a bailout for companies that have committed to spend heavily on data centers before demand has been demonstrated and pricing has stabilized. With OpenAI and Anthropic preparing to go public, the White House would do better to wait before placing its bet.

OpenAI CEO Sam Altman is said to have pushed for federal investment last year but publicly repudiated the idea after CFO Sarah Friar suggested federal loan guarantees. If the feds were to buy into OpenAI, the deal might take the form of a public wealth fund – so the public would receive revenue from intellectual property that AI firms have captured and are reselling.

US Senator Bernie Sanders (I-VT) last week said he planned to introduce a bill called the American AI Sovereign Wealth Fund Act. Funded by a one-time 50 percent tax paid in AI company stock, it would give the public a say in how AI is used and a portion of the revenue generated by AI companies (which, again, follows from the largely uncompensated capture of public content).

Meanwhile, the White House last week issued an executive order directing “the national security enterprise to accelerate AI adoption to meet surging demand, adapt the best commercial and open-source technologies for mission use, assure that fielded systems are robust, steerable, controllable, and preserve clear lines of accountability under the Constitutional chain of command.” And the order promises “new partnerships with willing private-sector companies to secure America’s cutting-edge AI against global threats.”

Buying into these companies doesn’t make a lot of sense if they can deliver on their promises at a viable price. The market would ensure plenty of good options for federal procurement.

But if leading AI models are priced like Claude Mythos, reported to run $25 per million input tokens and $125 per million output tokens, or about 5x Opus 4.8, there may be some concern that leading edge AI will be too costly for much of the market. Uber’s $1,500 monthly token spending cap per employee AI tool suggests companies won’t reward the AI industry for over-investing. If cutting-edge AI is going to be priced out of reach for most industries and if it really can accomplish things that lesser models cannot, the case for federal involvement gets stronger.

It would be a shame if the feds rewarded OpenAI and its peers with taxpayer money because that would reward fiscal irresponsibility and hinder startups hoping to innovate. Worse still, it would commit funds prematurely and unnecessarily for some notional national security edge that’s razor thin and is being dulled by evolving open weight models and foreign model providers. 

The jury is still out – there are at least 115 lawsuits against AI companies – on whether there’s a broad, sustainable market for AI services outside of software development and perhaps a few other knowledge work markets. The government should wait for the courts, the public, and the market to weigh in before riding to the rescue. ®