How an AI Commons Fund can deliver UBI for All

Bernie Sanders made headlines this week with a proposal to take a 50% equity stake in leading AI companies - OpenAI, Anthropic, xAI and others - and use the returns to fund a sovereign wealth fund that pays direct dividends to the American people.

It was reported, predictably, as a radical socialist provocation. I think it is actually one of the most logical policy ideas to surface in years, and it points toward something bigger and more urgent than even Sanders has framed it.

Here is why, and what a truly transformative version of this idea could look like.

The commons argument is not radical - it’s obvious

Every large language model that currently exists was built by training on the accumulated writing, code, art, knowledge, and conversation of billions of human beings, the vast majority of whom received nothing for their contribution, were never asked for consent, and will not share in the returns. The knowledge that powers these systems - every Wikipedia article, every digitised book, every published paper, every forum thread - is a collective human asset, generated over centuries of intellectual labour and cultural production.

When an oil company extracts value from land that a community has always lived on, we at least have the concept of mineral rights, however imperfectly they are applied. When AI companies extract value from the entire accumulated output of human civilization, we have not yet developed the equivalent legal or economic framework to reflect what has actually happened. Sanders' instinct - that this constitutes a taking of a public resource, and that the public should share in the returns - is correct, and it follows the same logic that underpins Alaska's Permanent Fund Dividend, and more recently the Marshall Islands' sovereign wealth fund, which both distribute a basic income to every citizen.

The displacement reality that makes this urgent

What makes this more than an abstract question of ownership is the speed and shape of what is happening to labour markets right now. The first wave of automation displaced manual and routine clerical work. This wave is different - it is hitting knowledge work, white-collar professional services, legal research, financial analysis, coding, journalism, marketing, and customer service all at once, and it is hitting them faster than retraining systems, social protection systems, or labour markets can deal with.

This is already happening, with major firms across law, finance, and technology quietly reducing graduate intake and restructuring teams around AI-augmented workflows. Just look at the sudden rug pull experienced by Meta staff in recent weeks - a class of worker once considered at the top of the tech-status pyramid suddenly made redundant. 

The result is a simultaneous acceleration of capital accumulation at the top - the companies and shareholders profiting from AI productivity - and a hollowing out of the middle, as the white-collar salary base that historically anchored consumer spending and our tax base begins to shrink. The economic logic of this is straightforward: if you compress the wage bill significantly across the knowledge economy while concentrating returns among a small number of equity holders and technology companies, you destroy demand at a scale that no conventional stimulus policy can easily fix.

Governments are stuck doing the same old things they’ve always done: provide welfare for transitional periods; protect workers' rights; provide education and retraining services, and increase minimum wage. These are all fine in isolation. But they are fundamentally unsuited to cope with the approaching crisis. 

A universal basic income funded by AI returns is not just an equity argument. It is a macroeconomic stabiliser for an economy about to undergo significant structural disruption, ensuring that productivity gains translate into spending power distributed across the population rather than concentrated in a handful of balance sheets.

What a Global AI Commons Fund would look like

The Sanders proposal is national, but the logic applies globally. AI is being developed and deployed internationally, its training data was extracted internationally, and its job displacement effects are being felt internationally - including, and in some ways especially, in lower-income countries where labour markets are less mature and protected. 

A Global AI Commons Fund, modelled on existing sovereign wealth fund architecture but designed from the outset to distribute a universal dividend, would work as follows:

Revenue would be raised through a suite of complementary mechanisms and pooled into a governed fund with a mandate to distribute a proportion as a direct, unconditional payment to every eligible citizen (either at state, national or even global level), while investing the remainder in transition support - retraining, community economic development, and AI-proof public services and infrastructure.

The distribution architecture already exists as a proof of concept. Alaska has paid its Permanent Fund Dividend to every resident for over forty years. The Marshall Islands recently became the first country to legislate a national UBI funded through a sovereign wealth fund. What’s missing is the political decision to apply this model to the most significant new source of concentrated wealth the world has seen in a generation.

How to actually capture the wealth: practical mechanisms

This is where most commentary stops at the level of principle without getting specific. Here are several concrete approaches, each targeting a different node in the AI value chain, which could be used individually or in combination.

An AI Revenue Levy on frontier model companies. A percentage tax on gross revenues above a defined threshold for companies whose primary business is developing or deploying AI foundation models - OpenAI, Anthropic, Google DeepMind, Meta AI, xAI, Mistral and their successors. Set at, say, 10-15% of revenues above $1 billion, this would generate tens of billions annually at current scale and hundreds of billions as the sector grows.

A Compute Tax on semiconductor manufacturers and data centre operators. Nvidia's dominance of AI chip supply makes it an obvious collection point - a tax on the sale of high-performance AI accelerator chips (GPUs, TPUs, custom silicon) at the point of manufacture or first sale. Similarly, a surcharge on the electricity consumption or revenues of hyperscale data centres, which are the physical infrastructure of AI at scale, would capture value from the "picks and shovels" of the industry regardless of which model company ultimately profits.

A Data Royalty for training on the commons. The most direct response to the "essentially stolen" argument Sanders raises is to formalise a fee for using public-domain or scraped human-generated content to train commercial AI models. This could be structured as a per-token or per-parameter levy, collected at the point of model release or deployment, and paid into the fund. It mirrors the logic of compulsory licensing in music copyright, and would create an ongoing financial relationship between AI companies and the knowledge commons they depend on.

An Automation Displacement Contribution. Companies that demonstrably reduce their wage bill by replacing workers with AI - measured as a significant fall in headcount-to-revenue ratio above a baseline - pay a contribution proportional to the wages no longer paid. This directly links the source of displacement to the cost of supporting displaced workers, and creates an incentive structure that does not simply penalise productivity but asks those capturing the gains to contribute to managing the transition. Germany's Kurzarbeit scheme offers a partial precedent for this kind of contributory logic at scale.

Equity acquisition (the Sanders approach). Taking a direct ownership stake in leading AI companies - either through a one-time acquisition, a tax payable in stock, or a requirement that new AI companies above a valuation threshold issue a proportion of shares to a public fund on incorporation - gives the fund a long-term claim on returns rather than a one-off revenue event. This is structurally the most powerful mechanism but also the most politically contested, particularly in the US context.

A Windfall Profit Tax on AI-dependent incumbents. Google, Microsoft, Amazon, Meta, and Apple are not AI-first companies in the same way as OpenAI, but they are generating extraordinary windfall profits from AI integration - in advertising revenues, cloud services, and productivity software - that would not exist without the technology. A targeted windfall levy on profit growth attributable to AI across these sectors would capture a significant additional stream.

These mechanisms are not mutually exclusive, and a well-designed fund would combine several of them to avoid being overly dependent on any single company or sector, and to ensure the revenue base grows as the technology proliferates rather than concentrating around a small number of early dominant players.

You can take a small step to support these ideas by signing the AI Pledge for Humanity here.

What do you think is the most politically feasible entry point - national sovereign wealth funds like Sanders proposes, an international levy mechanism, or something else?

PS as proof that AI comes for us all, I used Claude to structure and draft this post.