Sam Altman may just be beating lawmakers at their own game. Reports the chief executive of OpenAI is discussing giving a five per cent stake in the ChatGPT maker to the US government suggest an AI sovereign wealth fund is not out of reach. After all, having the world’s best-known AI lab on board gives the concept a huge credibility boost.
Details of Altman’s proposal are thin, but should be viewed as a counter-offer to Bernie Sanders’ proposal in June. The senator wants to capture AI-generated wealth, on behalf of the public through a mandatory 50 per cent stock transfer from the largest frontier AI companies into a federally controlled trust. He estimates the vehicle would be worth $7tn at current valuations — making it large enough to pay every American roughly $1,000 a year from a 5 per cent dividend.
Altman’s alternative is well timed. While Sanders’ plan gets the AI wealth diagnosis right, it gets the prescription wrong.
Correct diagnosis
Critics assert that AI is the product of human ingenuity and private investment, not a passive resource sitting in the ground waiting to be taxed. They are right that the latter analogy is imperfect. But a meaningful share of the wealth generated by AI should probably become public capital.
Sanders argues AI is trained on a public resource more valuable than oil — the knowledge and labour of humanity, continuously renewed by the data billions of people generate every day.
What’s more, the energy needed for its build-out — which is the foundation of Nvidia CEO Jensen Huang’s five-layer AI stack — is being co-ordinated at the state level on both sides of the Pacific.
If the [energy] layer of the AI stack already warrants sovereign co-ordination, the wealth generated above it deserves the same logic
America’s AI Action Plan seeks a major expansion of power generation while China’s new Rmb2tn ($294bn) sovereign bond-financed compute network foresees huge grid upgrades. If the base layer of the AI stack already warrants sovereign co-ordination, the wealth generated above it deserves the same logic.
Yet the country in the middle of the stack — South Korea with its memory chips and semiconductor substrate — offers a valuable lesson about resisting the temptation to treat a structural technology supercycle like a depleting oil well, to be tapped for citizens’ cash.
On May 12, presidential policy chief Kim Yong-beom posted on Facebook that excess tax revenue from the country’s AI and semiconductor boom should fund a “citizen’s dividend”. The Korea Composite Stock Price Index (Kospi) fell 5.1 per cent intraday, before recovering. A rival presidential aide labelled the idea “extremely dangerous and irresponsible” and Nobel laureate Peter Howitt called it “premature”.
Deputy prime minister Koo Yun-cheol subsequently clarified that the government wouldn’t simply “distribute and spend the money” but “build it into an asset for the future”. The Kospi held firm, showing the market did not reject the idea of public benefit from AI wealth, but the absence of a plan for delivering it.
Since then, Seoul has added further momentum to the concept of an AI SWF. It has proposed a fund seeded with tax revenues from soaring memory chip sales to invest in fabs alongside Samsung Electronics and SK Hynix, alongside other projects.
Wrong prescription
This brings us to the prescription for AI wealth, and where Sanders’ bill goes wrong. AI profit is generated inside private companies, not extracted from state-owned ground. A forced transfer risks the government regulating companies it owns — precisely the tension that turned a 10 per cent US stake in Intel into a political controversy.
The institutional solution to the AI wealth debate already exists. It simply isn’t the one most proposals reach for. The usual inspiration is Norway’s sovereign wealth fund: hold the equity, keep managers at arm’s length. But Norway’s fund invests almost entirely abroad, meaning Oslo never owns a controlling stake in companies it regulates.
The better template for holding equity in AI labs is Singapore’s Temasek which has held majority stakes in domestic champions for 50 years without collapsing into regulator-shareholder conflict. As a former chair put it, when Singtel wanted to buy a major rival “they did not consult Temasek”.
A nation that taxes AI profits only to cut one-off household cheques treats a structural dawn like a sunset industry
Temasek returns up to half its expected long-term gains to Singapore’s budget, funding a fifth of annual spending and compounds the rest. That is the template Seoul is now studying for its own fund in the AI era.
The Alaska Permanent Fund has long treated direct dividends and reinvestment as a spectrum. Its annual payment to citizens from its resource royalties builds little future capacity, but is popular. Notably, the structure envisioned by OpenAI is based on this model — equity donations from AI labs pooled into a single unit paying dividends to the American public.
Ownership of AI champions is already under way on both sides of the Pacific. How those benefits will be distributed is the phase no government has fully designed.
A nation that taxes AI profits only to cut one-off household cheques treats a structural dawn like a sunset industry. The ones that get the governance right — and choose how proceeds are shared between citizens today and future capability — will not just avoid that mistake. They will have built the rarest thing in public finance: a windfall that compounds.