June 2026: The Month AI Became a Capital Game

Transcript

Sam: Okay, here's the number that reordered my whole picture of where this is going. In one month, a rocket company sold about seventy-five billion dollars of stock — the single biggest stock-market debut in human history — and it wasn't really about rockets.

Alex: It was about renting out the machines that run AI. That one deal made the first trillionaire on Earth. And it's only the fourth-biggest money story of the same four weeks.

Sam: The fourth.

Alex: The fourth. June was the month AI quietly stopped being a research race and became a capital one.

Sam: And that — that's the whole show today. Not one of those.

Alex: Welcome back to Dan's AI Intel. And Sam's right, so quick heads up before we dive in: this is not one of our usual deep dives.

Sam: Right — normally on this show we take one question. One weird, nagging question about AI, and we go all the way down the rabbit hole with it. One thing, all the way to the bottom.

Alex: This is the opposite of that. Once a month, we do the inverse: we step back from the firehose and just tell you everything that actually moved in AI. The whole month, in one sitting. This is the news roundup.

Sam: So if you've never heard a single earlier episode of ours — honestly, perfect. You don't need any of them. Start right here, today.

Alex: This is Dan's AI Intel, and the reason the show exists is pretty simple. The AI revolution is moving faster than any normal human being can keep up with. The knowledge horizon is brutally short — the thing that's true this month might be old news the next. So Dan built a whole stack of tools to chase it all down and actually understand it, and we turn that into a show you can keep pace with — on a walk, on a commute, in one sitting.

Sam: I'm Sam, that's Alex, and here's how the next forty-odd minutes work. We're going to move through the whole of June across the five lenses that, between them, basically decide where AI goes next. The big ideas and horizons. The frontier labs and their models. The compute and the power they all run on. The money. And then how governments and society are reacting to the whole thing.

Alex: And here's the spine of the entire month — the one thread to hold onto as we go. Capital became the product. For two years, the moat in AI was who has the cleverest model. As of June, the moat is who can raise, build, and rent the machines. Brains stopped being the bottleneck. Money and megawatts became it.

Sam: And stick around to the very end, because after all the news, we've got a fast, upbeat recap of the deep dives we dropped this month — the full rabbit holes — in case one of them turns out to be exactly the one you've been looking for.

Alex: One quick thing before we start the clock. If you follow the show — on Spotify, on Apple, wherever you're listening — this monthly roundup just lands in your feed automatically. It's free, it's one tap, and it's genuinely the easiest way to never fall behind on any of this. Okay.

Sam: Let's do the headlines first. Hit me with the month at a glance — the whole thing in thirty seconds.

Alex: Fast version. One: history's largest IPO turned out to be an AI compute bet in a rocket costume. Two: both of the leading AI labs filed to go public in the very same month. Three: a rocket company became a top-tier AI landlord, eighty billion dollars committed. Four: the talent war became an actual market event — a stock moved because two people quit. Five: Washington reached into a live product and switched it off, worldwide. And six: AI security crossed a line, from finding the bugs to fixing them, at planetary scale.

Sam: Six headlines. And four of them are basically about money.

Alex: Which is the tell. That's the month. Let's get into it.

Sam: So we start with big ideas — the horizon stuff, the where-is-all-this-going stuff. Which feels almost funny in a month this dominated by money. But the single most striking thing in June wasn't a model release at all.

Alex: It wasn't. It was an essay. And the big-ideas headline for the whole month is that the field's centre of gravity moved. For two solid years, the question everyone obsessed over was "how good is the model?" In June, it flipped to "what is the model allowed to do — and who pays for it?" That's a different universe of questions.

Sam: Okay, the essay. Tell me.

Alex: Dario Amodei — he runs Anthropic, one of the very top AI labs — published a piece called "Policy on the AI Exponential." And the shift inside it is the actual story. Because this is a founder who used to argue for light-touch rules. Just transparency, just tell people what you're doing. In June, he flipped — to asking for binding, enforceable regulation. On his own industry.

Sam: Wait, hang on — what does he actually want? Because "please regulate me" from a CEO can very easily mean a whole lot of nothing. It can be PR.

Alex: That's exactly the right suspicion, and it's why the specifics matter so much. This isn't vague. He wants FAA-style third-party testing for any model over a certain size threshold — meaning an outside body has to sign it off, like an aircraft. He wants the government to have actual authority to block an unsafe deployment. Mandatory incident reporting when something goes wrong. Wage insurance for the workers who get displaced. And a domestic ban on fully autonomous weapons.

Sam: Okay. That's not vibes. That's a genuine regulatory agenda, with teeth.

Alex: And his framing was about as blunt as a CEO gets. He basically said: now that a frontier model can demonstrably go out and find real cyber-threats on its own, it is time to go beyond transparency to serious, binding regulation. The capability got real, so the rules have to get real. Quick aside — if Amodei himself is the rabbit hole here, we did a whole episode on his worldview earlier this month: "Dario Amodei: The Scaling Thesis and AI Stewardship," number twelve. That's where this regulatory turn actually comes from.

Sam: So let me try the "so what" and you tell me if I've got it. When the one person with the most to lose from regulation is the one sitting down to draft it —

Alex: — the whole "just trust the labs" era is over. That's the read. For two years the implicit deal was, these companies are responsible, they'll self-govern, leave them be. The moment a founder stands up and says "no, actually, bind us by law," that deal is dead. He just told you the labs can't govern this alone.

Sam: It's the fox asking for a fence around the henhouse.

Alex: It is — but here's the nuance I'd hold, because it's the part that'll matter later. He's asking for a fence and he's also drawing the fence. Holding the pen. That's not nothing. When the regulated industry writes its own rulebook, you watch very carefully which gaps it leaves. But the headline stands: he reframed every government move we're going to talk about today. Because if even the builders are admitting there's a vacuum, then everything Washington did in June reads as someone rushing to fill it.

Sam: There was a second big-ideas story, and it's nerdier, but honestly I kind of love it.

Alex: The benchmarks one. Yeah.

Sam: Right. So on June 12th, a research group called Epoch shipped FrontierMath version two — and FrontierMath is basically the hardest math test we throw at an AI. Brutal, research-level problems. And the entire point of version two was to pull out and fix flawed questions that were in version one.

Alex: Which sounds like pure housekeeping. Cleaning up a test. But sit with the implication for a second, because it's bigger than it looks. Think about why they had to harden the test at all. It's because the models are starting to top it.

Sam: Ah — so the test got too easy because the AI got too good.

Alex: Right. And the second a frontier system starts acing a benchmark, the measurement itself becomes the contested ground. The fight quietly moves from "can a model pass this test?" to "is this test even telling us anything real anymore?" The ruler becomes the battleground.

Sam: So how we measure progress is now its own research problem. That's wild.

Alex: It has to be. Here's the analogy I'd use. Imagine the only thermometer everyone trusts starts reading the same number no matter how hot it gets. At that point the urgent scientific question isn't the temperature anymore — it's the thermometer. And every claim the AI industry makes about getting smarter is only as trustworthy as the ruler it's measuring itself with. If the ruler is saturated, the progress is just a rumour with a press release attached.

Sam: That's a genuinely unsettling thought. We might be flying partly on broken instruments.

Alex: And we don't fully know which ones. That's the honest state of it. And it connects right back to the Amodei essay, actually — because if you can't fully trust the measurements, then "just test the models and you'll know if they're safe" gets a lot shakier. The thing he's asking governments to do — third-party testing — depends on the tests still meaning something.

Sam: Right, the fence is only as good as the ruler you build it with.

Alex: Exactly. Couple of quick also-this-month ones before we drop down to the labs — there were two signs this whole "control" worry went properly mainstream in June. The phrase people started reaching for is "agency over raw intelligence."

Sam: Unpack that one — agency over intelligence.

Alex: It means the thing to be scared of stopped being a model that's smarter than us, and became a model that's more autonomous — one that's harder to switch off and predict. Not "it'll outthink you," but "it'll act on its own and you can't easily stop it." And the Council on Foreign Relations — which is about as buttoned-up, establishment foreign-policy as a building gets — published a piece literally titled "AI Is Facing a Crisis of Control, and the Industry Knows It." When that crowd is writing your headline, a worry that used to live on the fringe has officially arrived in the mainstream.

Sam: Okay. So we've got the big idea — the field is now fighting about control, and about who pays, not just about who's the smartest in the room. Let's drop down a level, to the labs themselves. What were they actually shipping and doing in June?

Alex: And the standout here is almost theatrical. Google's research base cracked, in public. Two of its most valuable people in the entire company walked out the door in roughly a single day — and the stock market actually noticed. The model race got cheaper and stranger at the same time, but the human story is the one that moved a share price.

Sam: Give me the names. Because I gather these are not two ordinary engineers handing in their notice.

Alex: They are very much not. Over about twenty-four hours, on June 18th and 19th, Google DeepMind lost Noam Shazeer and John Jumper. Shazeer is a co-author of "Attention Is All You Need."

Sam: And that paper is —

Alex: — that's the paper that introduced the Transformer. In 2017.

Sam: And the Transformer is basically — okay, help me say this right. It's the underlying design that every modern AI is built on. ChatGPT, Gemini, Claude — all of it, same foundation.

Alex: All of it. Same foundation. He's one of the people who invented the thing the entire industry stands on. He's also a Gemini co-lead at Google. And he announced he's leaving — for OpenAI. And on basically the same day, John Jumper — who shared the 2024 Nobel Prize in Chemistry, for AlphaFold, the protein-folding breakthrough — said he's leaving Google for Anthropic.

Sam: So a Transformer co-author and a Nobel laureate. In one day. Both walking out of the same company. To two different rivals.

Alex: And Alphabet's stock fell roughly five to six percent on the news. Analysts tied that drop directly to a fear that Google simply can't retain the talent that built its lead in the first place.

Sam: That's the part that genuinely gets me. A stock — a trillion-dollar company's stock — moves measurably because two human beings quit. So what does that actually tell us about where we are?

Alex: It tells you that in 2026, a research roster is a balance-sheet asset. These people are line items now. A Nobel laureate isn't just a brilliant scientist — he's something the market literally prices. And then look at who gained: OpenAI and Anthropic. Which are — and we'll get to this, it's not a coincidence — the exact two labs that were sprinting toward the public markets in the very same month.

Sam: Ah. So talent flows toward the strongest money story.

Alex: And that's the mechanism that should worry Google most. It compounds. The lab with the best capital story attracts the best people, who make it more valuable, which makes its capital story even better, which pulls in the next round of best people. It's a flywheel. And right now it's spinning away from Google.

Sam: Once it starts turning, it's really hard to stop.

Alex: Very hard. Momentum in this market is self-reinforcing in a way it just wasn't a couple of years ago. And think about what it means for Google specifically. They invented the Transformer — Shazeer's paper is the whole reason any of this exists. And they just watched the co-author leave for a rival. It's a little like the company that invented the thing now struggling to keep the people who can build the next version of it.

Sam: They wrote the recipe and lost the chef.

Alex: That's the fear the market just priced in, in a single afternoon.

Sam: Second lab story. And this one I think is genuinely a big deal, even though it got a lot less noise than two people quitting.

Alex: The OpenAI cyber launch. On June 22nd they shipped the full version of a model called GPT-5.5-Cyber, plus an open-source push called "Patch the Planet," which they co-founded with a respected security firm called Trail of Bits.

Sam: And the numbers on it are kind of wild, right?

Alex: They are. So the model scores 85.6% on a security benchmark called CyberGym — up from about 81.8% for the standard version of the model. But honestly the score is not the story. The story is the surrounding effort. It processed more than thirty million code commits, across thirty thousand-plus codebases. And human reviewers confirmed over seventy thousand genuine security findings actually fixed. Including — this is my favourite detail — a twenty-three-year-old flaw sitting in the OpenBSD kernel.

Sam: Twenty-three years. Something nobody caught for over two decades. So the leap here isn't that the AI is better at finding the holes —

Alex: — it's the verb. That's exactly it. AI security crossed from discovery to verified remediation. From "I found a problem" to "I wrote the fix, and I checked that it works." Finding bugs, machines have done for a while. Writing a validated patch, at this scale, and having a human confirm it landed — that's the new thing.

Sam: And I'm guessing the so-what here cuts both ways. Sharply.

Alex: Very sharply. Here's the uncomfortable symmetry. The exact same capability that lets trusted defenders go patch the open-source backbone of the entire internet — is a weapon, the instant you point it the other direction. The thing that finds and fixes a flaw is, run in reverse, the thing that finds and exploits it. Quick aside — if that "a machine that serves us so well we hand it the keys" worry is your thing, we pulled it apart in number eighteen, "The scariest robot stories were never about evil machines," from a couple of weeks back. It's the same control fear, dressed as science fiction.

Sam: So the tool that secures everything is the tool that could break everything.

Alex: Same tool. Which is precisely why it only ships through a tightly curated partner program — vetted defenders only, no open download. And it's a big part of why governments spent the rest of June thinking very, very hard about who is actually allowed to hold these models. Which — hold that thought, because it's the thread we pick up at the very end, in the governance lens.

Sam: Before we leave the labs — there was a whole wave of model releases too. But you told me the interesting thing isn't any one of them, it's the direction they all moved in.

Alex: Right. The release pace is still absolutely brutal — you blink, there's a new model. But the centre of gravity shifted. It moved from "biggest" to "cheapest good-enough." A few examples. Anthropic's Claude Fable 5 landed right up near the reasoning ceiling — top-tier. A model called MiniMax M2.5 offered near-frontier quality at a fraction of the cost. And Google put out something called Diffusion Gemma.

Sam: And that last one's actually a different kind of thing, isn't it? Not just another model.

Alex: It is, genuinely. So most text AI generates words one at a time, left to right, like writing a sentence by hand. Diffusion Gemma generates them in parallel chunks — more like developing a Polaroid, where the whole image resolves at once instead of being drawn line by line. The payoff is it's materially faster.

Sam: Okay, so everything's getting cheaper, and now there are three credible frontier options. Doesn't that make it easier to just pick one provider and lock in? Less to keep track of?

Alex: You'd absolutely think so — and it's the opposite. Single-provider lock-in got more expensive this month, precisely because three different labs each now have a credible top-tier model. Why would you marry one of them when the price floor is collapsing underneath all three? The smart money keeps its options open and lets them undercut each other.

Sam: A couple of quick "also this month" ones before we go down to the machines?

Alex: Yeah, three fast ones. Google's Gemini 3.5 Pro slipped its month — they'd publicly promised it and then didn't ship it, with leaks hinting at weaker reasoning versus rivals, which only sharpens that retention worry behind the Shazeer and Jumper exits. OpenAI bought a company called Ona, to give its coding agents persistent memory between runs — basically agents that remember what they were doing. And ByteDance launched Seedance 2.0, a video generator, straight into the buzzsaw of US scrutiny over TikTok — Chinese frontier capability shipping into a very hostile regulatory climate.

Sam: Right. So we've covered the big idea, and the labs. Now the layer underneath all of it — the actual compute. The physical buildings, the chips, the power. And this is where the money story really, properly starts.

Alex: This is the engine room of the whole month. And the headline is almost absurd when you say it out loud: a rocket company became a top-tier AI compute provider, from a standing start. Over eighty billion dollars committed, in a matter of weeks. Without ever having run a cloud business in its entire life.

Sam: Okay, walk me through how that's even possible. Eighty billion, from nothing, in weeks.

Alex: So on June 22nd, an open-weight AI startup called Reflection AI signed a compute lease with SpaceX — worth up to 6.3 billion dollars. They're paying 150 million dollars a month, starting July 1st, for Nvidia GB300 chips — the top-end stuff — at a data centre called Colossus 2, near Memphis.

Sam: And that single lease is what tips them over eighty billion?

Alex: That single lease pushes SpaceX's committed compute revenue past eighty billion dollars through 2029. And crucially, that's on top of deals already on the books. Anthropic is already paying about 1.25 billion a month for Colossus 1. Google's already paying around 920 million a month. So a rocket company — with literally zero prior cloud business — is now a genuine peer to Amazon and Microsoft as an AI landlord.

Sam: Okay. "AI landlord." Unpack that phrase for me, because I think it might be the whole frame for the month.

Alex: It is the frame. Compute has become the economy's new rent. Think about it like this. In a gold rush, you can go dig for gold and maybe strike it, maybe not — that's the AI labs, betting on the cleverest model. Or you can sell the picks and the shovels and the land to every single miner, and get paid no matter who strikes gold. SpaceX just became the company that owns the land and rents out the shovels. They don't have to win the AI race. They own the road everyone has to drive down to run it.

Sam: And here's the connection you keep teasing — this is the same story as that record IPO.

Alex: It's the exact same story, one chapter down. Same Colossus, same compute, same rent. The rockets are the costume. The compute is the business. Hold that, because in the money chapter it's going to click all the way into place.

Sam: One thing I want to pin down, though — Reflection is described as "open-weight." Why does it matter who the renter is?

Alex: Good instinct. It matters because it tells you who's renting. Open-weight means they give away the model itself but they still need somewhere enormous to train and run it — so even the companies whose whole pitch is "we're open and free" are paying a landlord 150 million a month for the privilege. The openness is at the model layer. The rent is at the compute layer. And the rent is where the money is.

Sam: So no matter your philosophy, you still have to pay the landlord.

Alex: Everybody pays the landlord. That's the entire point of being the landlord.

Sam: There was a detail buried inside that Reflection deal that you said is the real tell. The thing most people skip past.

Alex: It's quietly the most revealing thing in this entire chapter. Nvidia invested 800 million dollars into Reflection AI — as a venture backer. And Nvidia also supplies the GB300 chips that SpaceX bought, specifically to serve Reflection.

Sam: Hang on. Let me make sure I've got this. So Nvidia is on both sides of the same trade. It put money into the customer, and it sold the gear that serves that customer.

Alex: Both sides. It funded the renter and it sold the landlord the building materials. And once you see that one move, you start to see the whole shape of the thing — a compute economy where a small handful of giant players finance each other, supply each other, and rent to each other. The money moves in a circle, and the circle is getting tighter. Quick aside — if the economics of that circle grab you, who actually makes money on every token and why Google quietly has the edge, that's a full episode: number twenty, "AI's Hidden 70x Subsidy," from just last week. It's the layer underneath this one.

Sam: And the so-what of that tightening circle?

Alex: This is the bit I find genuinely important, and it's the bridge to our last chapter. That concentration — money and compute pooling into a few firms, a few buildings — is exactly what makes June's government moves possible. When the whole industry's compute is concentrated this tightly, a government has very few doors it actually has to knock on to control the entire thing. Concentration, on the company side, is leverage on the government side. Remember that, because in the governance lens it pays off hard.

Sam: Quick mentions to close out the engine room?

Alex: Two. First — and this one's important — the real ceiling here isn't chips, it's power. Colossus runs at gigawatt scale. Energy, not GPUs, is now the binding constraint on the next wave of compute, which is exactly why where you physically site a data centre has suddenly become a geopolitical question — you go where the power is. And second, Qualcomm opened talks to buy a chip startup called Tenstorrent, for eight to ten billion dollars — a clear sign that the scramble to escape depending on Nvidia is now pushing everyone to just buy their way into making their own AI silicon.

Sam: Okay. We have now teased this thing about four separate times. The money chapter. The IPO. Let's finally do it.

Alex: The window didn't just open this month. It produced the single largest stock-market listing in history, and the world's first trillionaire, in one go. Let's do it.

Sam: Give it to me straight. The numbers.

Alex: June 12th. SpaceX — with xAI, Elon Musk's AI company, folded inside it — went public. It sold 555.6 million shares, at 135 dollars each, raising about seventy-five billion dollars. Then it closed up 19% on its very first day, for a first-day valuation near two trillion dollars.

Sam: And that's a record. By how much?

Alex: It's not close. It eclipsed Saudi Aramco's 25.6 billion-dollar record from 2019. We're talking roughly three times the previous biggest IPO ever. And it minted the first-ever trillionaire — Elon Musk, on paper, in a single day.

Sam: But here's the thing you've kept insisting on, and I finally get it now. The headline everyone read said "rockets."

Alex: The headline said rockets. The actual investment case is the Colossus compute empire we just spent a whole chapter on. Investors didn't buy a space company. They bought an AI landlord that happens to be wearing a space company's logo. And for scale — the entire US IPO market raised something like a few tens of billions across all of 2025. This one single deal dwarfed an entire year of the whole market. Quick aside — if Musk himself is the rabbit hole, the man, the contradiction, what xAI inside this thing actually means, that's its own episode: number twenty-one, "Elon Musk Became the Danger He Warned Us About," dropped right at the end of the month.

Sam: So the so-what is — public money, real public-market money, is now flooding toward whoever owns AI's physical layer. The buildings and the power.

Alex: And it does two things at the same time, which is what makes it such a big moment. It validates the infrastructure bet — the market is voting that the picks-and-shovels really are where the value lives. And, simultaneously, it concentrates that bet even further, into a tinier and tinier handful of names. The rich layer gets richer and narrower.

Sam: And the window didn't just close after SpaceX walked through it. The labs themselves moved.

Alex: They did. And this is the part that would've been almost unthinkable a year ago. Anthropic filed a draft S-1 — that's the formal paperwork you file to go public — on June 1st. This was after a sixty-five-billion-dollar funding round, at a valuation around 965 billion dollars. With run-rate revenue near forty-four billion annualised. And a projected first-ever operating profit — about 559 million dollars — in the second quarter.

Sam: Wait, stop — Anthropic's about to actually be profitable? I thought the whole thing about these labs was that they just set investor money on fire.

Alex: That was the whole thing about them. And that's the structural shift hiding in this filing. They're crossing over. And it wasn't just Anthropic — OpenAI filed confidentially on June 9th, at a roughly 900-billion-dollar valuation, though they signalled an actual public listing might still be a while off.

Sam: So if you add up the three of them — SpaceX, Anthropic, OpenAI — all reaching for the public markets at once —

Alex: — together they could demand north of two hundred billion dollars from public investors. And the so-what here is the big structural one. The frontier labs are crossing a line — from cash-incinerating research bets into public companies, with real revenue and, in Anthropic's case, real profit. Which hands them exactly the capital they need to keep winning the talent war we talked about two chapters ago. It all loops back.

Sam: The flywheel again. Money buys the talent, the talent builds the value, the value raises the money.

Alex: Round and round, and accelerating. A couple of quick ones to round out the markets chapter. A Chinese lab called Moonshot AI went looking for a thirty-billion-dollar valuation — which is a marker that the Chinese frontier is now raising at full Western scale, not playing catch-up on a shoestring anymore. And investors have started treating this whole AI-IPO cohort — SpaceX, Anthropic, OpenAI — as one single tracked thing, a "trillion-dollar wave," which only concentrates market attention onto an even smaller set of names.

Sam: Okay. Last lens. We've done the ideas, the labs, the machines, and the money. Now the big one for the rest of us — how are governments, and society, actually reacting to all of this?

Alex: And this is where June got genuinely tense. Governments reached for the off-switch. Literally. And the United States tried to deregulate AI and to hard-control AI in the very same month — and the contradiction showed, in public.

Sam: The off-switch. This is the Anthropic story.

Alex: This is the single rawest demonstration of the whole month. In mid-June, a US export-control directive ordered Anthropic to bar all foreign-national access to its two most powerful models — Mythos 5 and Fable 5. And rather than try to build some partial, leaky, who-gets-in block, Anthropic just disabled both models. Entirely. For every customer, worldwide. To be absolutely sure it complied. That order landed around June 14th.

Sam: So the government didn't ask nicely, and it didn't negotiate. It reached into a live, running product, and the product went dark. For everyone on Earth.

Alex: Globally, in one move. And here's the thing it laid bare, the thing worth really sitting with. Where does control over a frontier AI actually sit? Not with the lab that built the model. With the government that can compel it offline. An "off-switch" for a frontier model stopped being a thought experiment you'd debate in a seminar — and became an operational fact, demonstrated in a single week. Quick aside — this is the exact story we took all the way down a couple of weeks back, in number seventeen, "America Built a Kill Switch for AI — Then Flipped It On Its Own Allies." If the question of who can switch your AI off from another capital is the one that grabs you, that's the whole episode.

Sam: And I'm guessing that lands hardest on exactly the companies we just spent a chapter saying are going public.

Alex: That's the cruel irony, and it's the link back to the money chapter. The labs whose entire valuation assumes uninterrupted, global, always-on reach — just got shown, live, that their reach can be switched off from a single capital. From Washington. That's a brand-new kind of risk, and the markets are going to have to start pricing it.

Sam: Sovereign risk. Right inside the valuation.

Alex: Right inside it. The thing the IPO assumed — that you can serve the whole world — turns out to have a kill switch held by a government. And remember the concentration point from the compute chapter? This is where it pays off. The reason Washington could do this at all is that the frontier is so concentrated. There are only a handful of labs that matter, running on a handful of compute clusters. A government doesn't have to police a thousand companies. It has to make one phone call.

Sam: So the thing that made these companies so valuable — being a tiny, dominant club — is the exact thing that makes them controllable.

Alex: That's the whole irony of the month in one sentence. Concentration is what made the money. And concentration is what hands the government the off-switch.

Sam: But — and here's what I genuinely don't get — wasn't the same government also pulling back on AI rules this month? That's the bit that doesn't fit.

Alex: That's the contradiction, and it's the sharpest point in the whole chapter. So, on June 2nd, the White House issued an order called "Promoting Advanced Artificial Intelligence Innovation and Security." It keeps a frontier-model review — but it makes it voluntary. A pre-release look, no licence required, and crucially, no power to actually block a launch. It drops the old reporting requirements. And it removes the previous administration's civil-rights and labour provisions.

Sam: Okay, so that's hands-off. That's loosen the rules, let the models ship.

Alex: That's hands-off on release. And then, days later, the export-control order on Anthropic was about as hands-on as a government can possibly get. Same government. Same month. Deregulating how models get released — and hard-controlling who's allowed to use them.

Sam: So it's not actually a coherent policy. It's two completely different reflexes fighting each other.

Alex: That's the read, exactly. A deregulatory, get-out-of-the-way philosophy, colliding head-on with a national-security reflex. And the inconsistency itself became a risk — one that even close US allies flagged out loud, because if you can't predict which way a government will jump, you can't plan around it. The phrase for the month is control without consistency.

Sam: You can't deregulate with one hand and grab the kill switch with the other and call it a strategy.

Alex: And yet that's June.

Sam: And then the part that actually touches everyone listening — not just the labs and the governments. The jobs.

Alex: The labour story stopped being a forecast this month and became a count. A real number. Tracked US tech layoffs that explicitly cite AI as the reason mounted past 150,000 for the year. And it's spreading well beyond tech — into finance, logistics, consulting, retail. Oracle alone disclosed about 21,000 cuts — that's roughly 13% of its entire staff — and tied them directly to AI adoption.

Sam: 150,000 people. Citing AI by name. That is not a trend piece anymore. That's happening.

Alex: It's a count, not a projection. And here's the part that makes it so hard to think about — at the very same time, AI adoption surged the other way. This is genuinely the same single wave, seen from two sides. The displacement and the deployment are one phenomenon. And the number I'd really hold from this: ChatGPT's share of the global AI-assistant market slipped below half, for the first time ever. 46.4%. As Gemini climbed to 27.7% and Claude to 10.3%.

Sam: So even the consumer crown is contested now. But — connect that for me. Why does ChatGPT's market share matter for the jobs story specifically?

Alex: Because it's the reason the labour shock is almost impossible to govern. It's a genuinely competitive market now. No single company can be slowed down — by a regulator, by a conscience, by anything — without instantly ceding ground to a rival. If you tap the brakes on one player, the other two just take the share, and the layoffs happen anyway, somewhere else. Nobody can afford to be the one who goes first and goes slow.

Sam: So even if a company wanted to be careful with people's jobs, the market punishes it for it.

Alex: That's the trap. The competition itself removes the brakes. A last couple of mentions to close the chapter. That ChatGPT-below-50% figure comes from Sensor Tower's State of AI report — the headline being the assistant market is officially no longer a one-app monopoly. And Samsung reportedly rolled out ChatGPT Enterprise to around 125,000 seats — which is one of the largest single deployments yet, and it marks enterprise AI moving from little pilot projects to the org-wide default. The deployment side of that exact same wave.

Sam: Okay. That's the news. That's June — all five lenses, the entire month, in one go.

Alex: And now, exactly like we promised at the top — fast, upbeat, because you've earned it — here's what we actually went deep on this month. Every one of these is a full episode that's already live, a proper rabbit hole. We're just handing you the one thing you'll walk away with from each, so you know which one is worth your next forty minutes.

Sam: Rapid fire. Go.

Alex: Right at the start of June we dropped a whole run of portraits — the people steering this thing. Demis Hassabis, on the scientific path to AGI, that's number eleven. Dario Amodei and the scaling thesis, twelve. Ilya Sutskever on safe superintelligence, thirteen. Yoshua Bengio on securing the path, fourteen. And Amanda Askell, the architect behind Constitutional AI, fifteen. If you want the actual minds behind every headline we just ran through — start there.

Sam: Then it got weird, in the best way. Number sixteen — "Will the first true AGI actually be an extraterrestrial entity?" The takeaway: the strangest serious argument in AI is that the first genuine alien intelligence we meet might be one we build ourselves.

Alex: Number seventeen — "America Built a Kill Switch for AI, Then Flipped It On Its Own Allies." And this is the one that just paid off in our governance lens — sovereign AI was never about whose soil the data centre sits on; it's about whether another country can switch your access off. Which, this month, the US proved it can.

Sam: Number eighteen — "The Scariest Robot Stories Were Never About Evil Machines." The takeaway: every great AI nightmare, decoded, is the same quiet fear — a machine that serves us so well we hand over the future — and the safety math now agrees.

Alex: Number nineteen — "The Tech Giants Agreed to Build the Agent Internet Together, So They Could Fight Over Your Front Door." The takeaway: they'll happily share the plumbing of the coming agent internet, as long as each of them gets to own the one piece that matters — the door into your life.

Sam: Number twenty — "AI's Hidden 70x Subsidy, and Why Google Quietly Wins." The takeaway: the AI meter isn't the giveaway — your flat monthly plan is the real subsidy, and who survives the price war gets decided one layer down, at the silicon, where Google owns its chips and everyone else rents Nvidia's.

Alex: And right at the end of the month, number twenty-one — "Elon Musk Became the Danger He Warned Us About." The takeaway: the man who spent a decade warning that AI could end humanity quietly made himself the single person who most concentrates that exact power.

Sam: So that's the month's rabbit holes — all live, all waiting for you. And lining them up like that, you can feel June's through-line running straight through every one of them: power, money, and who gets to hold the off-switch.

Alex: If any one of those is the rabbit hole you've been looking for, go fall into it. That is the whole point of this show.

Sam: So that's June. Let me try to land the plane — the one thing to actually walk away with.

Alex: Do it.

Sam: June was the month AI became a capital game. The biggest IPO in history was really a bet on compute. Both leading labs filed to go public. The talent war moved a stock price. And the moat stopped being the cleverest model, and became who can raise, build, and rent the machines.

Alex: And the counter-current — the thing pushing the other way — was governments grabbing for leverage they'd let slip. They reached in and switched a frontier model off, worldwide, while deregulating with the other hand. And underneath all of it, the human cost kept climbing — past 150,000 jobs, on a wave nobody can slow down without losing the race.

Sam: So the two things to hold from the whole month. The money decisively became the product. And the off-switch turned out to live in a capital — not in a lab.

Alex: And that is it for this month's roundup. Thank you, genuinely, for spending the time with us. I hope you come away seeing a little more clearly where all of this is heading — because it's a fast, complicated, genuinely once-in-a-lifetime picture, with a brutally short shelf life on what's even true, and that is exactly what makes it worth following this closely.

Sam: Quick note, for full transparency: this show is AI-generated. Dan builds a custom stack of AI tools to chase down the questions about all this he can't stop thinking about — it started out made with NotebookLM, and it's now produced with his own engine — mainly so he can learn it himself, and then he publishes it for anyone who'd like to follow along.

Alex: And before you go, one genuinely useful thing you can do. Follow the show. Whatever app you're listening in right now, there's a follow button, or a little plus button — it's one tap, it's completely free, and it does two real things. You'll get next month's roundup, and every deep dive, the moment they land. And honestly — for a small, independent show like this one, a follow is the single biggest lever there is for helping it reach other people who are trying to make sense of all this. So if the month was worth your time, go ahead and hit follow.

Sam: And one real ask, from Dan. Tell us what to dig into next. What landed this month, what didn't, which of those five lenses you want more of, which deep dive you want us to push even further on — because that, genuinely, is the stuff that decides which questions we chase next. So if anything in here sparked a thought, or a disagreement, or a "wait, hang on, but what about…", we would honestly love to hear it. Drop a line to podcast@connectiveshift.com. Every single message gets read.

Alex: Follow the show, so next month's roundup just lands automatically — and if one of those deep dives is calling your name, go answer it. We'll see you next month.

Sam: See you next month.