A Piece of the Problem

00:00:04 On Friday, on Air Force One, a reporter asked President Trump about an idea that had been circulating for a few days, and he didn't bat it away. He said he'd been talking to AI executives about, in his words, "concepts where pieces could be given to the American public, where the American public essentially becomes a partner with the companies." By Saturday, his account of it had grown a little.

00:00:27 He said the administration was exploring giving Americans ownership stakes in AI companies so that, as he put it, the American people can benefit from the success of AI. Let me lay out what we actually know, because the words promise a lot and the structure delivers almost nothing yet.

00:00:44 CNBC reported that the administration has been discussing an equity stake specifically with OpenAI. Some of that equity, the reporting goes, could seed a so-called Public Wealth Fund that OpenAI itself has been pitching. OpenAI's own description of that fund is that proceeds "could be distributed directly to citizens, allowing more people to participate directly in the upside of AI-driven growth, regardless of their starting wealth or access to capital." Bloomberg adds a detail worth holding onto: Sam Altman has reportedly been floating the idea of a government stake in major AI companies since early 2025.

00:01:20 So this isn't a thing the White House sprang on the labs. At least one lab has been quietly inviting it for over a year. If this feels familiar, it should. The government took a ten percent stake in Intel last year when the chipmaker was struggling. Trump's interest in the state holding equity in private companies isn't new, and it's not confined to one party.

00:01:41 Senator Bernie Sanders, this week, proposed a one-time fifty percent tax that companies like OpenAI, Anthropic, and xAI would pay in the form of stock. Sanders argued it would, quote, "give the public a direct role in determining the future of this technology" and "guarantee that the trillions of dollars potentially generated by A.I.

00:02:01 are used to improve the lives of all of us." So you have a Republican president and a democratic-socialist senator arriving, from opposite ends, at the same instinct: the public should own a slice of these companies. The most interesting reaction came from David Sacks, the investor who until recently was Trump's AI and crypto czar and now co-chairs the President's Council of Advisors on Science and Technology.

00:02:26 Sacks said he could see why Sanders' idea resonates, including with many on the right, but he warned it would, quote, "accelerate the corporate-government fusion we're already sliding toward." That's a striking thing for an administration ally to say about his own side's proposal.

00:02:42 And it names the real tension. The pitch is simple: the public gets a share of the upside. The risk Sacks is pointing at runs the other way — the state and the labs end up holding each other's risk, each with a financial reason to protect the other. Once the Treasury owns a piece of a frontier lab, the government is no longer a neutral referee on that lab.

00:03:03 It's an investor with a position to defend. And there's a darker read circulating too. The former Microsoft employee Dare Obasanjo, watching all this, suggested that the groundwork is already being laid for a government bailout of OpenAI. That's worth holding next to the optimistic pitch, because the same mechanism cuts both ways.

00:03:23 A public stake can be a way for citizens to share the gains. It can also be the pipe through which public money flows in if one of these companies, burning cash at the rate they are, hits a wall. Which one you're looking at depends entirely on terms nobody has written down.

00:03:39 I promised yesterday to track whether the government-equity talk turned into real terms or stayed a concept. As of today it's still very much a concept. There's no announced company, no mechanism, and no number. But it moved from a leak to the President saying it on the record, twice in two days.

00:03:57 That's not nothing.

00:03:58 Here's the part that makes today worth an episode rather than a headline. The same week the government is talking about buying into these companies, the most conservative pool of money on the planet just declined to. S&P Dow Jones Indices, which decides what goes into the S&P 500, was asked to change its rules to let SpaceX in quickly after its planned public offering.

00:04:19 SpaceX wanted to float only about three percent of its shares, and it's currently unprofitable, carrying a debt load that's reached twenty-nine billion dollars, much of it from its own spending spree on AI infrastructure. The rule changes it wanted would have applied to all the mega-cap newcomers.

00:04:37 And in its final decision, S&P said no. The exact language: "no changes will be made to the eligibility criteria including financial viability screens, seasoning period, or minimum investable weight factor." Translation: you have to be consistently profitable, and you have to wait.

00:04:53 SpaceX doesn't clear that bar today. Neither, the reporting notes, do OpenAI or Anthropic. Why does index membership matter so much? Because roughly seven and a half trillion dollars sits in funds that passively track the S&P 500. When a company enters the index, those funds have to buy it, in proportion, automatically.

00:05:11 Bloomberg Intelligence estimated that fast entry would have triggered about fourteen billion dollars of passive buying for SpaceX, more than eight billion for OpenAI, and four and a half billion for Anthropic. That's money that buys your stock not because anyone judged it a good bet, but because a rule said it must.

00:05:30 S&P just declined to switch that machine on for unprofitable AI firms. Now, the rest of the index world blinked differently. The Nasdaq exchange changed its rules to let SpaceX into the Nasdaq-100 within fifteen trading days instead of the usual three months. FTSE Russell decided to give SpaceX accelerated entry to its top-500 index after just five trading days.

00:05:51 So the doors aren't all locked. But the S&P 500 is the one that matters most to ordinary retirement accounts, and it's the one that held the line on profits. Sit with the contrast for a second. On one side, the federal government is musing about taking an equity position so the public can share in the upside.

00:06:09 On the other, the index that actually governs most of the public's existing exposure to the stock market just said these companies aren't financially viable enough to include on the normal terms. The denial landed days after Morningstar's analysts called SpaceX significantly overvalued going into its offering, pinning a fair value around seven hundred and eighty billion dollars, against the company's roughly one-and-three-quarter-trillion-dollar target.

00:06:35 Less than half. So when you hear "let the American people benefit from the success of AI," it's worth asking which success, measured how. The private valuations are enormous and the public-market gatekeepers are skeptical, and both of those things are true at the same moment the government wants to wade in.

00:06:53 When Trump's ownership-stake comment hit social media, it pulled a quarter of a million views and several hundred replies in a few hours, and the replies are a decent map of where the public actually is on this. They split cleanly. One camp loved it. A reply that got traction called it, quote, "an America First move that actually shares the upside," arguing that regular Americans shouldn't just absorb the disruption from AI, they should own a piece of the gains.

00:07:20 That's the optimistic read, and it's not a stupid one. If this technology really does generate the wealth its boosters promise, then spreading the ownership is more attractive than letting it pool in a few hundred private cap tables. The other camp went straight at the mechanism, which is exactly the right instinct, because the mechanism is where these things live or die.

00:07:41 One reply put it plainly: the market question is the mechanism, not the headline. Do citizens get equity, options, or a fund stake? Who gets diluted to pay for it? "Owning a piece of the AI buildout sounds great," he wrote, "the structure is everything. Until that's defined it's a slogan, not a policy." I'd go further.

00:07:59 There are at least three completely different things hiding inside the phrase "ownership stake." There's the government holding shares on the public's behalf, the way it held Intel. There's a sovereign-style fund that pays dividends to citizens, the Alaska-oil-check model.

00:08:15 And there's individuals being handed actual shares. Those have wildly different consequences for who controls the companies, who bears the downside, and whether this is wealth distribution or just a new entanglement between Washington and three or four labs. And then there was the reply that stuck with me, because it cut to the politics underneath.

00:08:35 The administration, this person noted, is in voluntary talks, not seizing anything. But, he wrote, fifty-five percent of Americans think AI does more harm than good, and the fix being sold is, quote, "own a piece of the problem instead of stopping it." Whether or not that poll number is exact, the point is sharp.

00:08:53 An equity stake doesn't slow the buildout, address the job displacement, or touch the energy strain. It offers people a financial interest in the thing they're anxious about, which is a way of converting opposition into shareholding. That's clever politics. It isn't, by itself, a safety policy or a labor policy.

00:09:11 It's an ownership policy, and those aren't the same thing.

00:09:14 While Washington debates who gets the upside, New York spent this week arguing about who absorbs the cost. On Thursday, the state legislature passed a one-year moratorium on large data centers. It now sits on Governor Kathy Hochul's desk. If she signs it, New York becomes the first state to hit pause on the facilities powering the AI boom.

00:09:34 The bill targets what it calls hyperscale data centers, anything over twenty megawatts, and it largely aims at the biggest builders. It won't apply to facilities that already hold their state permits. The senator who co-authored it, Kristen Gonzalez, gave the clearest version of the argument I've heard from a state official.

00:09:53 "Big tech has been used to writing their own rules," she said, "or not having rules that they have to play by. This is one of the first times that we're really drawing a line in the sand." She laid out the stakes in numbers: there are at least twenty-eight large data centers currently being evaluated by the state for their grid impact, and together they would add roughly nine thousand, six hundred and eighty-two megawatts of demand onto a grid she described as already constrained and aging.

00:10:21 And she didn't soften the why. "We should not have to sacrifice our water, our energy, our green space and local communities for big tech," she said, "and specifically for generative AI, which is oftentimes used for things like AI slop." The bill would require an environmental impact report documenting water and electricity use, plus new labor, efficiency, and transparency standards, and ratepayer protections meant to keep ordinary New Yorkers' energy bills from rising to subsidize the buildout.

00:10:52 The original proposal asked for a three-year pause. It got cut to one year as a compromise. Now put a national number behind the local fight, because a paper out this week does exactly that. Researchers compiled facility-level data on four hundred and three US hyperscale data centers operating between May 2024 and April 2025.

00:11:11 Their finding: those centers drew somewhere between sixty-eight and ninety-nine terawatt-hours of electricity, which works out to about one-point-eight percent of total US electricity consumption. They were associated with thirty-seven to fifty-four million metric tons of carbon dioxide, with a little over half the power coming from fossil fuels.

00:11:31 And the carbon intensity of the electricity these centers pull is about five hundred and forty-five grams of CO2 per kilowatt-hour, roughly forty-eight percent above the national grid average. So they're not just big consumers. They're drawing dirtier-than-average power, on average, right now.

00:11:48 The industry's response came through the Data Center Coalition, which warned that a statewide moratorium would discourage investment and signal that New York is closed for business. And there's a real economic argument there; these are large capital projects. But Gonzalez had an answer for the people who say leave it to local governments.

00:12:07 "It's an abdication of our responsibility to ask a local government to engage and take on the wealthiest companies in the world," she said. "That is what state government is for." She told the reporter that a neighbor said: if my electric bill goes up another fifty dollars, I can't live here.

00:12:37 Cordes isn't an anti-government type. Her line was, "I'm not a person who's about big government, but come on, please help us here in these small rural towns." That tells you the politics are moving. Note also: Maine passed a similar pause in April and its governor vetoed it.

00:12:53 Hochul has been cool to a statewide approach before. So this isn't decided. But the binding constraint on the AI buildout keeps turning out to be the same one we talked about Wednesday — power, and now the people who live next to where it gets spent.

00:13:07 There's a parallel move happening in healthcare, and it runs the opposite direction from New York's caution. According to a Washington Post report by Elizabeth Dwoskin, the Trump administration is building the regulatory groundwork for AI chatbots that can diagnose illness and prescribe medicine, including an FDA fast track for digital health tools.

00:13:27 The person tasked with bringing AI into the health system is Amy Gleason, who took over the DOGE service from Elon Musk and is now advising Health and Human Services Secretary Robert F. Kennedy Junior. The direction of travel is to speed these tools through approval.

00:13:43 And I want to be fair to the upside here: a well-built diagnostic aid that's actually accurate could extend care to people who can't easily see a doctor, and triage tools have clear promise. The deployment ambition isn't crazy on its face. The problem is the evidence the fast track would be running ahead of.

00:14:01 The Post cites a study where researchers had twelve hundred volunteers pose as patients in conversations with chatbots built on OpenAI's GPT and Meta's Llama. The result: the chatbots identified medical conditions accurately just thirty-four percent of the time, and were, in the researchers' assessment, essentially no better than Google at guiding users.

00:14:21 Thirty-four percent. For a tool whose job description, in this push, includes diagnosing and prescribing. Physicians quoted in the piece warn that AI can introduce more problems than it solves in a clinical setting, and you can see why. A search engine that's wrong two times in three doesn't write you a prescription.

00:14:40 A chatbot on this path might. Here's the gap that worries me. Regulatory fast tracks exist for good reasons; they get breakthrough therapies to dying patients faster. But a fast track is a bet that the thing works well enough that speed beats caution. Applying that to a general-purpose chatbot whose measured diagnostic accuracy is a coin flip and change is a different kind of bet.

00:15:02 It moves the risk onto the patient who trusts the output, and onto the doctor who has to clean up after it. If the FDA does build this lane, the detail that matters is whether it requires the tools to clear an accuracy bar specific to diagnosis and prescribing, or whether it just shortens the calendar.

00:15:19 Those are very different policies wearing the same name.

00:15:22 A shorter one, but it belongs in this set, because it's about who actually writes the rules we've been talking about. The Information reported today that Sriram Krishnan, one of the top AI policy advisors in the Trump administration, plans to leave his White House role at the end of June.

00:15:39 According to the sources, he intends to start a pro-Trump AI policy institution. On its own, a staffer leaving government is routine. What makes this worth thirty seconds is the pattern it sits in. We talked Thursday about a bipartisan House bill that would set up a federal AI risk framework and preempt state laws.

00:15:57 We've talked all week about who holds the levers — the administration's classified benchmark, the equity-stake idea, and the Pentagon's fight with Anthropic. The policy is being written right now, and the people writing it are starting to move from inside the government to outside institutions that will keep shaping it from the other side of the door.

00:16:16 A pro-administration AI think tank, staffed by the person who just held the government pen, is a way of making sure the influence outlives the appointment. I'm not assigning a motive here; people leave jobs for ordinary reasons, and building an institution is legitimate work.

00:16:32 But if you're trying to track where AI rules actually get made, the answer is increasingly: not only in the agencies, but in the orbit of well-funded advocacy shops that the rule-writers themselves go on to run. That's the structure to keep an eye on as the state-preemption fight and the equity-stake idea both look for a permanent home.

00:16:51 I want to end somewhere different, because it cuts against the day's money-and-power register and gets at something harder to price. The Guardian ran a piece by Kenneth Mohammed about a speech that went viral across Africa and the Caribbean — a fierce and defiant address attributed to Namibia's president, Netumbo Nandi-Ndaitwah.

00:17:11 It denounced corruption and foreign exploitation, and declared that Africa's resources belong to its people, not politicians or multinationals. People shared it with real admiration. Finally, here was a leader speaking with moral clarity. There was one problem.

00:17:26 She never said it. Nandi-Ndaitwah rejected the speech as an AI-generated fabrication. And yet it keeps circulating, more than a year after it first appeared. Mohammed's argument is the part that's stayed with me. The speech wasn't embraced because people are gullible, he writes.

00:17:43 It was embraced because it, quote, "articulated truths many believe their leaders are afraid to say." The fake didn't spread by fooling people about a fact. It spread by satisfying a hunger that real institutions weren't feeding. We usually talk about deepfakes as a deception problem — someone fabricates a politician saying something damaging, and the danger is that you believe a lie.

00:18:06 This is the inverse, and it's stranger. Here the fabrication says something people wish were true, and they pass it on knowing, or not much caring, that it's synthetic, because the sentiment lands even when the source is fiction. That's a harder thing to fight than a forgery, because the vulnerability isn't technological, it's political.

00:18:26 Mohammed ties it to a leadership vacuum that runs from gerontocratic presidents who've held power for decades to what he calls performative leadership in the Caribbean. The synthetic speech became a kind of Rorschach test for everything people felt their actual leaders refused to say.

00:18:43 For the AI conversation, here's the uncomfortable takeaway. We spend a lot of energy on detection — watermarking, provenance, and whether you can prove a given clip is fake. That work matters. But it assumes people want the truth and just need help finding it. A speech that goes viral because the audience prefers the fake to the real won't be solved by a label saying "AI-generated." They already know.

00:19:08 The demand was for the words, and a machine supplied them when the institutions wouldn't. That should worry anyone who thinks the answer to synthetic media is purely a matter of better forensics.

00:19:19 Pull the day together and it lands on a single question: when the AI buildout pays off, or fails to, who's holding the receipt. The government is feeling around for an equity stake so the public shares the upside. The S&P 500 just refused to let the most reliable money buy in until the profits show up.

00:19:34 New York is trying to make sure the people living next to the substations don't pay the bill for a buildout they didn't ask for. The FDA may speed a diagnostic tool to market faster than its accuracy earns. And a fake president gave a speech better than the real ones, because the audience wanted the words more than they wanted the truth.

00:19:51 None of these is settled. The equity idea is a sentence on an airplane, not a term sheet. The moratorium needs a signature. The fast track needs rules. The open question for tomorrow is the mechanism on that ownership stake — whether anyone in Washington can answer what those replies kept asking: own a piece of what, paid for by whom, and with what say over how it's run.

00:20:10 Until that's defined, it's a slogan. The moment it gets defined, we'll learn whether the public is being made a partner or just handed a share of the risk. That's the one I'll be on tomorrow. Jonas.