Bernie Sanders wants the public to own half of the big AI companies. His bill would hit any AI firm that crosses $200 million in revenue with a one-time tax — payable not in cash but in stock, 50% of it — handed over to a federally managed fund. The fund would pay every American a dividend of around a thousand dollars a year, and a presidentially appointed panel would vote those shares to block company decisions it judged harmful to ordinary people.

My first instinct was to dismiss the whole thing. But I don’t think that’s right, because buried inside it is a question I think is worth asking — and that’s what makes the rest of it so frustrating.

The part worth taking seriously

The premise I’ll concede is this: the economy keeps consolidating. A smaller and smaller number of very large companies capture a bigger and bigger share of the value. And within those companies, the equity is concentrating too — an outsized portion of the upside accruing to a relatively small set of founders and early executives. If you own a meaningful piece of that, you ride the wave. Most people don’t, in any real amount, and they watch it go by. It’s a fair bet that a huge share of the wealth this AI build-out creates will end up in very few hands.

So the question — how do regular people get a stake in the upside of a concentrating economy, and how do we share the value being created more broadly across society, not just with the people who already own shares? — is a good one. And the answer Sanders reaches for — the public holds equity and shares the returns — isn’t some fringe socialist fever dream. It’s basically what Alaska does. Every Alaskan gets a check each year from a fund built on oil royalties. It’s what Norway does, on a much bigger scale — a $2 trillion fund that owns a slice of nearly every public company on earth. Singapore’s been a major shareholder in real companies for decades. These are market economies where the state is, among other things, a shareholder. The idea has a track record.

So I’m not opposed to the public sharing in the upside. It’s a serious question worth taking seriously. The problem is that almost every specific design choice in this particular bill is wrong.

Why only AI?

The first thing that doesn’t hold up is the target itself. If the argument is “concentrated capital should be broadly owned,” that argument has nothing to do with AI. It applies to every large company in America — Amazon, Walmart, the banks, the energy majors. The concentration that should worry us shows up wherever a small number of people capture an outsized share of the gains, whatever the underlying driver. Singling out AI makes the whole thing feel less like a principle and more like a tax on whatever happens to be the hot, slightly scary thing this year. The AI label here is both unclear and beside the point. Wherever concentration shows up, on whatever factor, we should think about it the same way.

And the label invites a game. The tax triggers off a company’s “AI-related revenue” — so the whole thing hinges on a definition nobody has pinned down. What even counts? Almost every company is racing to call itself an AI company right now, and plenty have a real case, because AI is genuinely lifting their margins and accelerating their growth. They aren’t foundational AI labs, but they’re being reshaped by deploying AI inside their businesses. Now watch how fast they’d race to stop claiming the label the moment it cost them half their stock. That line will be hard to draw and harder to hold — fought in court, lobbied in the drafting, cheated in the accounting. A principle that applies to everyone is hard to dodge. A penalty aimed at one buzzword is just an invitation to litigate the label.

Fifty percent, one time, is the wrong mechanism

Then there’s the structure of the take.

Fifty percent is too much. There’s a real difference between the public participating in the upside and the public owning the company. Ten or twenty percent gets you the former. Fifty gets you the latter — it’s not a stake anymore, it’s control. If the goal is for people to share in the returns, you don’t need half.

“One-time” doesn’t hold together. A company crosses $200 million once and gets hit once? So the firms that already crossed it are grandfathered, and the next startup to succeed pays a toll the incumbents never will. That’s backwards — you’ve built a tax that punishes the act of catching up. If you genuinely think the public should share in this, you want something ongoing and small, not a single confiscatory event triggered by success.

And someone has to get diluted. Where do the shares come from? If you’re handing 50% to a federal fund, the people who owned that 50% before — founders, employees who took stock instead of salary, early investors, the pension funds and endowments that backed them — just got cut in half. You can think some of those people are too rich and still recognize that “we’ll take half of what you were promised” is a real cost that needs a real answer. The bill doesn’t give one. History actually offers a cleaner version of this mechanic: in the 1970s, Sweden’s Meidner Plan proposed forcing companies to issue new shares — worth a slice of annual profits — into collective funds, gradually, over decades. Whatever else you think of it, that at least confronted the dilution question honestly and spread it over time. A one-time grab of existing shares doesn’t.

You can’t force a growth company to pay you a dividend

The bill requires the fund to pay out a 5% dividend, and on the consumer side that instinct makes sense. Regular people don’t just want equity that appreciates on paper — they want income, money they can actually use. A stake you can’t spend isn’t much of a stake.

The trouble is that these particular companies can’t easily produce it. Most of the big AI firms are losing money on purpose — pouring every dollar back into compute, research, and growth. You can’t squeeze a 5% dividend out of a company that isn’t throwing off distributable profit without forcing it to stop investing in the thing that made it valuable in the first place. And the bill closes the obvious escape hatch: the fund is barred from selling its shares, so it can’t raise the cash that way either. The income the bill promises and the companies it targets are pulling in opposite directions. Notice that the funds that actually work avoid this bind: Alaska pays its dividend out of the investment returns of a diversified fund, not by reaching into an operating company and demanding cash. Norway doesn’t pay a citizen dividend at all — it accumulates. The instinct to give people income is right; this structure can’t deliver it.

The line you don’t cross: letting Washington run the company

And now the part I think is genuinely a bad idea, not just a badly tuned one.

The bill hands the fund’s voting shares — and equal seats on the board — to a new federal body: a seven-member commission, nominated by the president and confirmed by the Senate from bipartisan lists, with seats set aside for labor, pension-fund, and AI expertise. The commission would use those shares and seats to block company decisions it deems harmful to Americans and push the ones it likes. That’s not the government acting as a passive shareholder. That’s the government in the room, making operating decisions about products, research, and strategy.

There’s a line here, and this crosses it. The government already has a tool for protecting people from companies: laws and regulation. Rules of the road, applied to everyone, out in the open, that you can see coming and plan around. That’s the legitimate lever, and I’m all for using it. Telling companies what they can’t do is the government’s job. Telling companies what to do — sitting on the board, voting the shares, steering the calls — is a different thing, and a much worse one.

And it’s worse for reasons that go well beyond fine-tuning. To its credit, the bill tries to insulate this commission — bipartisan nominations, Senate confirmation, designated seats — so it isn’t simply the White House’s plaything. But that design doesn’t rescue the idea; it just dresses it up. You’re still handing a small panel of political appointees veto power and board seats over the most valuable companies in the country. That’s an open invitation to capture and corruption, and a recipe for decisions made for political reasons rather than for the business, its customers, or its long-term health — no matter how balanced the nominating process looks on paper. And at bottom it cuts against what this country is. America is a capitalist, free-market society — that’s a feature, not a bug. We regulate where regulation is warranted, we set the rules of the road, but we do not put the government in charge of running companies and industries. This bill does exactly that. It’s worth remembering that the one serious modern attempt to go down this road — Sweden’s Meidner Plan — collapsed precisely on the question of control, triggering the largest business protest in the country’s history and getting scrapped within a decade. Of all the bad parts of this bill, this one is by far the worst.

So where does that leave us

Strip the bill down and you find two different things mashed together. One is a defensible idea: in an economy where value keeps concentrating, give regular people a real stake in the upside. Broad-based, passive, ongoing, shared returns. Alaska, Norway, and a stack of serious proposals show it can be built.

The other is the thing history keeps rejecting: pick a politically convenient target, seize a controlling share of it in one shot, and put the government in the operator’s chair. Sanders took the first idea and bolted on the second.

I don’t have the perfect blueprint for how to address these challenges, and I’m not sure anyone else does yet either. But I’m fairly confident about what could make this plan more viable: a smaller stake, taken gradually, spread across the whole economy rather than one trendy slice of it, sharing the returns without seizing the wheel. The government can regulate AI. It shouldn’t run it. The question underneath is real and worth taking seriously — but it needs a lot more debate and honest research to find the right design before anyone’s ready to write it into law.