AIPolitics 20: Everyone Agrees the AI Dividend Should Happen...and That's Kinda the Problem
Trump, Bernie, Altman, and Musk all just decided AI wealth has to be shared. They agree on nothing else, and the fight over who controls it runs from the Federation to the Butlerian Jihad.
Something happened over the last week that I’ve been waiting and thinking about for a while, and I almost can’t believe how fast it arrived.
That I’ve been waiting on it at all tells you how much science fiction I’ve put away over the years, because somewhere in all that Trek and Asimov I kept circling “ye olde technology dividend,” the promise that eventually the machines make so much surplus that we all just share it and live happily ever after, and the nagging question of if and how and when we’d ever land a not-so-ridiculous policy version. Turns out the Writers’ Room handling reality’s recent episodes has been working hella hard at original, surreal, and convoluted, and this week they finally aired “The Dividend” episode, plotted out by people who clearly never read the source material.
Welcome to the path to the Federation, amiright?
First, the happenings…
President Trump told reporters he’s exploring having the federal government take equity stakes in OpenAI, Anthropic, and xAI. He called it a “partnership with the American people,” the idea being that “pieces could be given to the American public” so that the public “essentially becomes a partner with the companies.” A US stake in the AI giants, he said, “could be a beautiful thing.” And it didn’t come out of nowhere. Days earlier he’d signed an order pulling the frontier labs into a voluntary security review before they release new models, and a memo telling the military to adopt their products while forbidding the companies from interfering. In one week the administration moved to bind the state to the labs on three fronts at once: security, defense, and now ownership.
In the same stretch of days, Bernie Sanders introduced a bill to slap a one-time tax on the stock of the biggest AI firms and pour the proceeds into a public wealth fund. Elizabeth Warren proposed an excise tax on the electricity data centers burn, plus a wealth tax on AI billionaires, with the money routed to displaced workers. Congressman Greg Casar published an op-ed arguing we should tax AI itself and spend the revenue making jobs. New York’s legislature passed the first statewide moratorium on new data centers in the country. And underneath the politicians, the men who build the machines were saying the loudest things of all. Sam Altman now wants every American to hold “an ownership share in whatever the AI creates.” Elon Musk wants the federal government to mail everyone a check he calls “universal high income.”
Read it all in one sitting and it looks like something crystallized overnight. The most powerful technology in human history is going to throw off an enormous amount of wealth, and across the entire political spectrum, people have decided that some of it has to be shared with everyone else somehow.
I want to tell you that's good news. Mostly I want to tell you it's a trap, because the agreement that something is leading us to bountiful cornucopias is real, and yet it's almost entirely on the wrong question.
The one thing they actually agree on
Strip away the proposals and look at the premise underneath them, which is the genuinely new thing.
A year ago the AI policy conversation in Washington was about two things: safety, meaning will the models do something catastrophic, and competitiveness, meaning will we beat China. Distribution barely registered. Now it’s the center of gravity. Sanders and Trump, Warren and Hawley, Altman and Amodei and Musk, and the leader writers at the Financial Times have all converged on a single proposition. AI is going to concentrate wealth and power so severely that it threatens to come apart in our hands, and something has to be done to spread the gains before that happens.
You can hear how seriously the people closest to the money take this in how they talk about it. Dario Amodei, who runs Anthropic, made what he called a “pragmatic argument” to his fellow billionaires: back a good version of AI taxation now, because if they don’t, they’ll “inevitably get a bad version designed by a mob.” He warned that the concentration AI produces could “break society.” Gavin Newsom put it with less varnish. “The pitchforks are here, they’re not just coming.”
This isn’t only a feeling. Last week the MIT AI Risk Initiative published an assessment from 272 experts of which AI dangers carry better than a ten percent chance of catastrophic outcomes even if we mitigate them well. Two of the five that made the list were “inequality and unemployment” and “power centralization.” The risk people, the ones usually preoccupied with rogue models and engineered pandemics, now rank distributional collapse alongside cyberweapons.
The chorus isn’t everyone, though. The loudest checkbook in AI politics is dead against any of it: the hundred-million-dollar super PACs the labs’ investors built to elect deregulators, the donor class that would like the wealth left exactly where it’s collecting. The convergence is among the principals and the politicians, the people who have to win votes or launder a reputation. The people who only have to write checks are funding the other side.
So the diagnosis is shared, and it’s likely correct, and it’s a real event that it’s shared. Hold onto that, because it’s about to do the opposite of what you’d expect. Agreement on the diagnosis is just the starting gun for a fight, not the beginning of a settlement.
Cash, stock, or a job
The agreement ends the instant anyone has to name the cure. And the cures don’t line up the way a lifetime of American politics has trained you to expect.
There are three of them (so far), really.
The first is cash. Musk’s “universal high income” is the purest version: the government issues checks, funded somehow, and as the robots take over production “everyone can have a penthouse if they want.” This is the old universal basic income idea wearing a bigger number. Early Altman was here too, back when he funded a study mailing people a thousand dollars a month.
The second is ownership. The idea here is to hand people a stake instead of an income, a piece of the thing itself. Altman has actually moved to this position, and the move is telling. He’s backed away from UBI on the grounds that “cash payments will not be enough,” and now argues for “universal basic wealth, better than universal basic income,” an ownership share in what the AI builds. OpenAI’s policy paper sketches a national fund, seeded by the companies, that owns a slice of the whole AI economy. Trump’s equity-stake idea is the same shape from the other direction. So is Sanders’s sovereign wealth fund. Which means the strangest sentence I’ll write this year is true: on the question of whether the public should own a piece of the AI companies, Donald Trump and Bernie Sanders are standing in the same place. Or rather, they’re standing on the same instrument. What they would do with it could not be more different, and that gap is where the real fight lives.
The third cure is the oldest one: a job. Not a check, not a share, but work. This is what the public actually says it wants. Poll after poll finds people uninterested in being paid to sit idle; they want to earn their living. It’s also where the most rigorous policy thinkers have landed, and I’ll come back to them. For now just notice that “human jobs first” is a slogan you can already hear forming on both the populist left and the populist right.
Three cures: cash, stock, a job. Now watch where the parties fall. Trump and Sanders on ownership. Musk and early Altman on cash. The public and the populists of both sides on work. The lines don’t sort by party. They cut across it.
This is the thing I've been arguing since the start of this series, finally visible in the daylight of actual legislation. AI doesn't easily slot into the old left-right cleavage (especially since our current politics is cut so markedly by education, gender, and race). Instead, it cuts a new line at a right angle to it, the line I tried to draw a few weeks ago.
You can see it in the geography. Brookings mapped local AI exposure against how counties vote, and the most exposed places, the ones where the most jobs sit in the machine’s path, are the bluest. When NBC asked Americans how they felt about a list of political things, AI’s net favorability landed below the Democratic Party and just above ICE. The public has already filed AI in the same drawer as the other things it resents.
One caveat, the same one I made last time: this scramble is clearest among the elites, the people who write the bills and sign the checks. Whether ordinary voters sort the same way, cash here and jobs there, or just take the cue from their party, is still a guess. The elite room is bisected. The electorate is a question mark.
Why everyone suddenly wants to placate you
Worth asking why the Placation ReflexTM kicked in across the whole spectrum at the same moment, before most people have actually lost a job to a model. The answer is the most useful idea I know for thinking about this, and it comes from Peter Turchin.
Turchin studies why societies fall into what he calls ages of discord, and his model runs on three gauges. The first is popular immiseration, where the living standards of ordinary people stagnate or slide. The second is elite overproduction, where a society mints more would-be elites than it has positions to hold them, and the surplus curdles into capable, well-connected people with a grievance. The third is the state running low on money and legitimacy. When all three climb together, things break.
AI pushes on all three. But the one that matters most here is the second, and it’s where AI does something new. The classic version of elite overproduction is too many law graduates for too few partnerships. AI doesn’t just overproduce the aspirants. It demotes the credentialed class itself. The lawyer watching the model do discovery, the analyst watching the junior roles get cut, the coder watching the agent ship the feature. These are exactly the people Turchin warns about: educated, articulate, networked, and suddenly told the ladder they climbed leads nowhere. Even Leicht and Ball, two writers I’m about to defend, put it in nakedly Turchin terms. Young people “raised to believe they could join the knowledge-work elite,” now “functionally barred” from it, and “prone to rebellion.” The dangerous figure in the AI economy isn’t the displaced warehouse worker. It’s the displaced associate.
And this is what the dividend actually is, underneath the policy language. It’s not generosity. It’s elite self-preservation. When Amodei tells billionaires to fund a good version before they get a bad one designed by a mob, he is describing, almost word for word, the move ruling classes make to defuse a structural-demographic crisis: redistribute something, widen the circle, buy time. The people building the disruption have read enough history to know what tends to come after disruptions, and they would like to schedule the concession themselves.
They can see the other pole from here, and it isn’t a policy disagreement. It’s the Butlerian Jihad: the oldest anti-machine story we’ve got, thou shalt not make a machine in the likeness of a human mind, and its blunt modern cousin, the EMP and the server farm that goes dark and stays dark. That’s the “bad version designed by a mob” Amodei keeps warning his fellow billionaires about. The dividend is the wager that you can buy your way to the Federation before the mob decides to burn the machines instead. And the mob is not a thought experiment: ask the Indianapolis councilman who woke up to thirteen bullets and a “No Data Centers” note on his door.
Which sets up the problem this whole piece is really about.
Quick intermission in the regularly scheduled dread. This whole series, following AI through the money, the coalitions, and the institutions downstream of both, is free. If it's your kind of trouble, the subscribe button keeps it landing in your inbox.
The fight isn’t whether to share it. It’s who gets to spend it.
Let’s suppose, against the odds, the dividend actually happens in the short-medium term. Most of these proposals will die in committee; America has floated public wealth funds before and built none of them, and that skepticism is worth keeping in your pocket. But suppose one survives. Suppose we get past cash and checks and land on the ownership model that Trump and Sanders, improbably, both want, a public fund that holds a piece of the AI economy and pays out to citizens. You might think that’s the moment the fighting stops, the rare patch of ground where left and right built the same thing.
It’s the moment the real fight starts.
A fund like that is a permanent, enormous pot of public money, and whoever controls how it gets spent controls a redistribution machine bigger than anything in American government since Social Security. The convergence on creating the pot guarantees a war over governing it, and the two visions of what it’s for aren’t a difference of degree. They’re a difference in kind.
The left wants the fund to be the foundation of a new social contract. Money that decommodifies the basics, the health care and housing and child care the market keeps failing to deliver. A sovereign wealth fund as the financing arm of social democracy, finally cut loose from the payroll taxes that fund the welfare state today. The dividend, in this vision, is collective, and it builds public goods.
The right wants the opposite thing wearing the same clothes. The model on the right is the Alaska Permanent Fund: oil money paid out as an annual check to every resident, no strings, no bureaucracy, no program. Trump’s framing gives it away. He doesn’t say “public investment,” he says the public “becomes a partner,” that “pieces” get “given to the public.” That’s ownership as individual property, a citizen’s dividend that deliberately routes around the state instead of through it. The whole appeal, on the right, is that it hands people cash and a stake while starving the administrative apparatus the left wants to build with the very same dollars.
These do not meet in the middle. A fund spent on a national child-care entitlement and a fund paid out as a no-strings check to every American are rival theories of what government is for. We have spent a century fighting over exactly this in smaller denominations. AI proposes to raise the stakes by an order of magnitude and hand the winner a money-printing machine.
There’s a wrinkle here that should worry everyone. A country sitting on a giant claim to AI’s output starts to rhyme with a petrostate. The analogy isn’t clean, of course, since the rent here is domestic and still has to be pried out of a working economy rather than pumped out of the ground, but the political shape carries over, right?
Political scientists have a sturdy literature on what happens to places that live off a rent instead of off their own work. The politics curdles into a fight over dividing the rent, accountability weakens because the government no longer needs your taxes, and patronage does the work policy used to. A national AI fund could stabilize the country by buying everyone in. It could also corrode it, by making the central question of American politics not what we build but who gets a cut. South Korea is already there in miniature, floating a tax on the excess profits of its chip champions that an official called a “national dividend.” We will get there at continental scale, and we have no idea how to govern it.
And look at what the same administration is already doing. The week it floated taking an ownership stake, it also ordered the labs to submit their models for government security review and told the military to deploy their products while barring them from interfering. A government that regulates these companies, buys from them, and now wants to own a piece of them is not a neutral steward of anyone’s dividend. It’s three conflicting interests wearing one hat, refereeing a game it’s also playing. We’ve run the regulator-and-customer version of this for eighty years with the defense contractors, and managed it, badly, with procurement rules and recusals. Adding owner to the same hat, over the same few firms that now sit at the center of the whole economy, is the new and unmanaged part.
So no, agreeing to share the wealth doesn’t end the cleavage AI opened. It moves it, from whether to redistribute to who owns the redistribution, which is a deeper and more durable fight, because it’s a fight about the shape of the state itself.
The catch: the people writing the check caused the problem
Now the part that makes me least optimistic any of this has a chance of happening.
Look again at who’s leading the charge. The loudest, most detailed dividend proposals aren’t coming from labor or from a mass movement. They’re coming from Altman, from Musk, from Amodei. The people manufacturing the disruption are also drafting the compensation for it, and in OpenAI’s case the structure is explicit: the company would donate equity to seed the very public fund the administration is now negotiating to accept. The disrupter writes the check and addresses it to a fund it helped design.
I wrote a whole piece a couple of months ago arguing that Sam Altman is not FDR, and this is why the distinction matters. The New Deal worked as a settlement because labor extracted it. It got wrung out of a frightened ownership class by strikes and votes and the credible threat of something worse. A concession you fight for binds differently than a gift you’re handed, because the gift can be withdrawn, and because the people receiving it know the giver is still in charge. Andy Hall, whose “Politics of Jobless Prosperity” I leaned on earlier in this series, makes the sharper academic version of the point: a social contract offered from above, by the powerful to an unorganized public, isn’t really a contract at all. It’s a press release with a budget.
Yes, I can hear the historical objection, because it’s the best one against this whole piece. The most durable welfare state in history was handed down from the top by a man terrified of a revolt. Bismarck built German social insurance in the 1880s precisely to cut the legs out from under the socialists, granted from above, wrung out of no one, and it held for a century.
So “a gift never binds” is too strong. Sometimes it binds fine. The difference between Bismarck’s settlement and a press release comes down to whether the state has the capacity and the spine to make the concession stick over the objections of the people footing the bill. That’s the thing in doubt here, because the people designing the AI dividend and the people who would have to be bound by it are the same people. Bismarck wasn’t Krupp. Here, Krupp is writing the pension plan.
And there’s a credibility problem sitting on top of it. Sam Altman wants to distribute AI’s wealth; he also lobbied to weaken the very measurement that would show how much wealth there is and where it went. Bernie Sanders found the cleanest way to expose the gap, turning to Elon Musk and asking how exactly he plans to fund “universal high income” when he won’t support even a five percent tax on his eight hundred billion dollars. The dividend offered by the man who refuses to be taxed is not a dividend. It’s a brand.
Turchin would tell you why this fails even when it’s sincere. The credentialed class AI threatens wants its role back, its status, the ladder it was promised, more than it wants a check. This is the one place the dividend has something to offer, and it’s why ownership beats cash: a stake is a kind of status, which is exactly why Trump reaches for “partner” instead of “recipient.” But a stake handed to you is still not a rung you climbed. It answers the wallet, gestures at the dignity, and restores neither the work nor the purpose. You can placate a hungry population with money. You cannot so easily placate a class that has lost its function, and that’s the class AI is coming for first.
The one serious attempt to thread it
I’ve laid out a lot of scrutiny on the dividend, so let’s try to be fair to the people trying to do it well, because they exist and they’re worth reading.
A few days ago Anton Leicht and Dean Ball published a piece called “Betting on Humans,” and it’s the most careful thing anyone has written on this. Their core move is humility. Nobody actually knows whether AI ends work or just reshuffles it, so the worst thing policymakers can do is lock in a guess. They argue against the instinct to freeze today’s labor market in place with automation bans and job guarantees, on the well-supported grounds that rigid labor rules strangle the very reinvention a technological shift demands. Instead they want to keep options open. Stop tilting the tax code to favor a robot over a worker. Build real measurement, so we actually know what’s happening to jobs close to real time, which means forcing the labs, who hold the usage data and have every commercial reason to spin it, to share. And their one bold bet: subsidize junior jobs directly, paying firms to keep hiring and training the young workers the machine is pricing out, so we don’t lose the talent pipeline and the on-ramp to a career in one swing.
It’s the best version, and I want to flag two things about it. The first is generous. Their measurement proposal quietly closes a loop this series has been tracking for months. They’re prescribing exactly the public measurement infrastructure that OpenAI was busy dismantling and that the new federal architecture keeps private. They’ve named the precondition for any honest policy, and they’re right that we don’t have it.
The second is a worry, and it’s the same worry as everything else here. Leicht and Ball reject the equity stake and reject UBI; they’d rather raise the corporate income tax, broadly and agnostically, and run the junior subsidy. That’s probably the most defensible package on offer. But it is still a technocratic fix, designed at a think tank, handed down. It answers the economics and it doesn’t answer the legitimacy. A junior-jobs subsidy is a fine policy and a terrible rallying cry. It will not feel, to the displaced associate, like a thing the public won. It will feel like a thing the smart people arranged. Even the best dividend has a hole where the politics should be.
A caveat that matters: none of this has actually happened yet
Now the quiet part, because the whole conversation is running ahead of the evidence.
The mass immiseration all this redistribution is meant to head off has not, so far, shown up in the numbers. Ezra Klein laid this out plainly a few weeks ago. Unemployment is 4.3 percent, almost exactly where it was in 2020; wages are stable; demand for software engineers, the people supposedly first against the wall, is still climbing. Andy Hall’s read is the same. The backlash isn’t here yet because the displacement isn’t here yet. The argument over whether it’s coming fast or slow is genuinely unresolved, with Anthropic warning the machines are starting to improve themselves and serious skeptics answering that the bottleneck on human progress was never raw intelligence in the first place. And the single most visible grievance, the one about data centers wrecking your town and spiking your power bill, turns out to be mostly thin when you actually chase the numbers, as Andy Masley has spent a year patiently documenting.
So that’s the shape of things. The politics of the AI dividend is sprinting toward a crisis the economics hasn’t yet produced. That doesn’t make the politics fake or disingenuous. It’s been the recurring lesson of this whole series that the politics of AI moves at the speed of fear, not the speed of unemployment. But it does mean we’re watching a settlement get negotiated for a disaster that may arrive late, or differently, or in some sectors and not others. Whether that’s wise preparation or premature panic depends entirely on a question none of us can answer yet, which is how fast and how deep the displacement actually goes.
Won, not granted
Put it all together and the political prize comes into focus, and it isn’t the one everyone’s racing for.
Everyone is racing to propose the dividend. Within a single week, Trump, Sanders, Warren, Casar, Altman, and Musk all planted a flag on some version of it. Proposing it is table stakes now. The prize goes to whoever can make the dividend feel won rather than granted, claimed by the public rather than gifted by the people who caused the problem. The tribune the country believes wrested a share of the AI economy away from the men who built it will own the coalition AI is busy assembling. The one who looks like he’s handing out the labs’ charity on their behalf will not.
The geography of AI anxiety is Democratic, the exposed counties are blue, and the party has, on paper, the most to gain from becoming the home of the AI-displaced. And yet the move that would actually capture this moment, a president nationalizing a slice of the AI giants and mailing the country its cut, is one a Republican is better positioned to make. He can do it as ownership and patriotism instead of as welfare, and he’s already taken quiet equity stakes in twenty other companies while no one was watching. If Trump claims the dividend as something he took from the billionaires on your behalf, he completes a realignment the Democrats have spent this decade sleepwalking through, and he does it on ground the Democrats thought was theirs.
Except his own donors are the catch. The same week he floated the stake, the hundred-million-dollar super PAC his investors built was busy trying to elect the candidates who will kill it. The move that captures the AI-displaced is the one Andreessen Horowitz is spending a fortune to prevent, which means the dividend splits Trump’s coalition at least as badly as it splits mine. Maybe that’s the only mercy in here: the thing cuts so cleanly across the old lines that nobody gets to own it without breaking something of their own. Or maybe it just means the prize goes to the first politician willing to cross his own donors to reach it.
The dividend is coming. I’m decently sure of that now; the week we just lived through convinced me that it might have to.
What I’m less sure of is which story it drops us into.
Roddenberry’s utopia didn’t open on the replicators. It opened on the Bell Riots in 2024 San Francisco, the displaced fenced into sanctuary districts until the whole thing went up, and a world war not long after. The Federation was the back half of that story. The peace got scraped out of the wreckage of the front half. And the front half is Herbert’s: the Butlerian Jihad, the moment a frightened humanity decided the machines were the enemy and tore them all down, leaving a universe that still flinched at a thinking mind ten thousand years later.
That’s the real fork, and it was never about whether AI makes us rich. It’s about which sci-fi we end up writing. The dividend is the switch. Make it feel like something the public won, and it’s the first clumsy step toward the replicators. Hand the arsonists a hose and call it a partnership, and you find out the pitchforks were never a figure of speech. The people with their hands on that switch right now are the same people who built the machine.
Roddenberry bet we’d choose the Federation. He just had the decency to show us the real immiseration and riots and wars first.
If you enjoyed (or hated) this one, restack it or forward it to the person in your life who keeps insisting AI is all hype, and the one who keeps insisting it’s the apocalypse. They’re both wrong in the same interesting way.
New here, or want to catch up? Some of the AI & Politics series so far:
Part 1: Artificial Intelligence and Politics — where it started, and where the Butlerian Jihad first came up
Part 5: The Credential Is the Coalition — why AI’s threat lands hardest on the credentialed class
Part 9: Sam Altman Is Not FDR — the labs’ New Deal cosplay
Part 15: The Backlash That Hasn’t Arrived...Yet — Andy Hall and the trigger
Part 18: The Cleavage Rotated — the four-quadrant map
Part 19: Comments on Arnold Kling’s Comments — the two-level model this one builds on
If I may quote Spock, this was a dazzling display of logic. Well done, sir.
What a time to be alive.
Interesting points. Outcome-wise, I am very pessimistic about any of this going well. It's simply a worthless tax and cash grab. It's like a cigarette tax that ends up funding the governor's mansion.