If a global dividend system is to be more than a thought experiment, we have to answer a hard question: where will the money come from?

A common critique is that there simply isn’t enough money to make something like this work globally. But in an AI-driven economy, two things seem increasingly likely. First, the removal of labor from production across key sectors could generate enormous new wealth—value that doesn’t exist today. Second, in a world of low marginal costs, goods and services could become significantly cheaper. In that world, even modest transfers well below the scale of our current welfare systems might go a long way.

Creating economic value is what markets do. Capturing a portion of that value for public use is a political and institutional challenge.

One early answer to this came in the form of the Windfall Clause, a report by researchers at the Center for the Governance of AI. It proposed AI labs voluntarily pre-commit in a legal contract to redistributing a share of their profits once they crossed a certain economic threshold—say, 1% of global GDP. The concept remains interesting, but the original formulation is likely insufficient for today’s landscape. The mechanism struggles against familiar headwinds. Profits are pliable. They can be shifted, hidden, reinvested. Thresholds can be gamed or endlessly deferred. Reviving or building on the Windfall Clause would require significant redesign: tighter definitions, stronger enforcement, and external levers to reinforce accountability.

There are ways to strengthen and diversify the toolkit. The Windfall Clause was a proposal tailored to AI labs, but they may not be the only entities capturing transformative value. And voluntary profit-sharing alone may not offer the enforceability or durability needed. What follows are several approaches that respond directly to those limitations: offering legal footholds, broader targets, and potential paths toward more stable institutional buy-in.

One idea is to secure equity stakes in the firms or infrastructure providers most likely to benefit from AI-led transformation. Unlike profit-sharing pledges, equity provides a legal claim on future value, and can be harder to evade or obscure. But equity-based models assume early action and leverage: that we know where value will accrue, and that public actors can move before capital lock-in. That window may close quickly and the value chain is likely to splinter across labs, cloud providers, chipmakers, and downstream companies that are quietly automating labor, often without public scrutiny or oversight.

Another approach might shift focus to the users of AI systems rather than their producers. Imagine a framework where large companies that automate away significant portions of their workforce are required to pay into a global fund as a condition for continued access to advanced AI imposed by the providers—licensing not just the technology, but the responsibility. But that, too, relies on a fragile foundation: a willingness from AI labs to coordinate, a willingness from firms to comply, and enough shared incentive to keep the system from unraveling the moment it becomes inconvenient.

Then there’s the state. Governments could implement taxes on AI-driven productivity gains or capital income, and route a portion of the proceeds into an international fund. These policies will be hard to coordinate but the incentives may change. If AI accelerates job loss and weakens demand, governments may find themselves scrambling to sustain consumption and prevent collapse. Under those conditions, redistribution could become a tool of macroeconomic stabilization, not just social justice.

And yet, even fragile foundations have a way of hardening under pressure. Because the history of social policy is not a history of foresight. Many of the major welfare systems we rely on today didn’t exist before the Industrial Revolution. They emerged in response to breakdown. As factories replaced farms, as urban poverty swelled and wealth concentrated, societies created public schools, labor laws, pensions, and insurance systems. Not because they envisioned them in advance, but because the cost of doing nothing became too high.

The same pattern holds at the global level. In 1944, as World War II neared its end and the Great Depression had shattered confidence in free-market capitalism, 44 nations gathered in Bretton Woods to design a new economic order. Out of that meeting came the IMF, the World Bank, and a global monetary framework. It wasn’t consensus that was the driving force. It was crisis.

If AI drives a similar rupture, then crisis will once again force the question. And when it does, the difference will be whether we have something to reach for: a policy framework, an institutional prototype, a draft on the table. It is a question that deserves far more attention, experimentation, and collective foresight.

Capturing a share of AI-driven economic surplus for public benefit won’t be easy. There are no turnkey solutions, and many mechanisms face serious technical and political challenges. Some of these proposals may sound idealistic but history reminds us that today’s political nonstarters can become tomorrow’s economic necessities. In a world where labor becomes less central to value creation and there are simply fewer jobs to go around, the boundary of what’s politically and economically feasible could shift—possibly faster than we expect.