Anthropic Goes Public (But Not Really)

Anthropic officially files for its public debut

Alex Shannon: I keep going back and forth on this — I think I actually land on the side that this is a good thing. Anthropic going public, the governance structure, the whole thing. I think it works.

Sam Hinton: Really? Because I read the same story and I came out the other end deeply uncomfortable. Like, this is a company that was literally founded on the premise that AI is one of the most dangerous technologies ever built — and now they’re ringing a bell on Wall Street?

Alex Shannon: But that’s exactly why the governance structure matters. They built protections in. Founder control, safety mission baked into the structure. It’s not like they just handed the keys to hedge funds.

Sam Hinton: Alex, every company says that before the IPO. The question is what happens eighteen months after, when the stock is down twenty percent and an activist investor is knocking on the door. That’s when you find out if your governance structure actually has teeth.

Alex Shannon: OK, that’s — honestly, that’s a fair point. And we’re going to dig into exactly that tension today, because there is a lot to unpack here.

Sam Hinton: A lot. And spoiler: Amazon is somehow the biggest winner in all of this, and I think most people have no idea how much they’re about to make.

Alex Shannon: You’re listening to Build By AI, I’m Alex Shannon. And yes, we’re already arguing — that’s how you know it’s going to be a good episode.

Sam Hinton: And I’m Sam Hinton. Big week today — Anthropic’s IPO filing is the headline, but we’ve also got a researcher revolt brewing against OpenAI, a wild financing move from SpaceX, Google building entire worlds in real time, and more. Buckle up.

Alex Shannon: And hey — quick note for listeners. Starting today, we’re switching things up. Build By AI is moving from daily episodes to one bigger weekly episode, every Wednesday, covering the whole week in AI. Same energy, more depth, all in one place.

Sam Hinton: Honestly, I think it’s going to be better. More time to actually dig in, connect the dots, have real conversations. So if you’ve been listening daily, thanks for coming along — Wednesday is your new AI news day. Let’s get into it.

Anthropic officially files for its public debut

Alex Shannon: Alright, let’s start with the big one. Anthropic — the AI safety company behind Claude, one of the leading large language models out there — has officially filed for an IPO. So this is them formally saying: we want to be a publicly traded company. The filing reveals strong revenue growth, which is great, but also mounting losses, which is less great, and what the reports are calling major enterprise adoption. Big companies are using Claude at scale.

Alex Shannon: But the really interesting part — and this is where it gets unusual — is the governance structure they’ve built. It’s specifically designed to keep founder control intact, with Dario Amodei and the founding team holding significant power, and it’s supposedly structured to protect the company’s AI safety mission even after it goes public.

Sam Hinton: Yeah, and look — the revenue growth part is real. Enterprise adoption is real. Claude is genuinely being used at serious scale by serious companies. But the losses piece is important context. This is a company spending enormous amounts of money on compute, on research, on safety work. They need capital. That’s the honest reason for the IPO.

Sam Hinton: And I want to be fair to them — accessing public markets is a legitimate way to get that capital. But there’s a reason I came out of reading this feeling unsettled, and it’s not the IPO itself. It’s the gap between what they say they want and what public markets actually reward.

Alex Shannon: So let’s talk about that governance structure, because this is where Anthropic is doing something genuinely different. They’re not just doing a standard dual-class share structure like Google or Meta did. What are we actually talking about here?

Sam Hinton: So the idea is that the founders retain enough control that no outside investor — not a major institutional investor, not an activist hedge fund — can come in and fundamentally change the direction of the company. The safety mission is supposed to be protected structurally, not just by good intentions. And look, that’s admirable on paper. A lot of thought went into this.

Sam Hinton: But here’s my devil’s advocate take: governance structures are written by lawyers and can be challenged by lawyers. And more importantly, they can be quietly eroded over time. You don’t always need a hostile takeover to lose your soul. Sometimes it happens incrementally — one product decision at a time, one quarterly earnings call at a time.

Alex Shannon: That’s the tension I keep coming back to. And I actually think there’s a version of this that works. If Anthropic can demonstrate that safety and profitability genuinely reinforce each other — that being the trustworthy AI company is actually the competitive moat — then maybe the public markets become an ally rather than an adversary.

Sam Hinton: I want to believe that. I really do. And maybe they can thread that needle. But the history of mission-driven companies going public is… not great. There are exceptions, but the gravity of quarterly earnings pressure is real and powerful. It bends companies in ways their founders never anticipated.

Alex Shannon: OK so for people listening who might be thinking — does this affect me, does it affect how I use Claude or whether I trust it? What’s the practical takeaway here?

Sam Hinton: Short term? Probably nothing changes. Claude still works the way it works, enterprise contracts are still enterprise contracts. But medium term, watch how Anthropic talks about safety in their earnings calls. Watch whether the language gets vaguer, whether the commitments get softer. That’s your signal. If the safety mission starts sounding more like a marketing line and less like an operational constraint, that’s when you should start paying attention.

Alex Shannon: Keep an eye on this because the Anthropic IPO isn’t just a business story — it’s kind of a test case for whether the AI safety movement can survive contact with public markets. And we’ll be watching it closely.

VCs will make billions on Anthropic’s IPO, but Amazon is set to make much (much) more

Alex Shannon: OK so staying in the Anthropic universe for a minute, because there’s a financial angle to this story that I think deserves its own conversation. When the IPO details came out, it became clear that yes, the venture capital investors who backed Anthropic early are going to make a lot of money. We’re talking billions. That’s a huge return. But Amazon — Amazon is positioned to make dramatically more. And we’re talking about a level of ownership stake that only became clear because of these massive note conversions disclosed in the filing.

Sam Hinton: This is the part of the story that genuinely surprised me, even though in retrospect it probably shouldn’t have. Amazon has been one of Anthropic’s biggest backers. They’ve committed billions in investment. But the note conversions — basically, debt that converts to equity — reveal just how large Amazon’s ownership stake actually is. It’s substantial. Like, considerably larger than people probably assumed.

Alex Shannon: For listeners who might not be super familiar with how this works — a convertible note is basically a loan that turns into ownership shares under certain conditions. So Amazon may have structured some of their investment as debt that now converts to equity at the IPO, which means their ownership percentage could be significantly higher than just their direct investment would suggest. Is that the right way to understand it?

Sam Hinton: That’s exactly right. And the reason this matters isn’t just the money — although the money is staggering, we’re potentially talking about tens of billions of dollars of value for Amazon. What matters is the power dynamic. Amazon is not just a financial backer of Anthropic. They’re a strategic partner, they run Claude on AWS, they’re deeply intertwined with the business. A huge ownership stake makes that relationship even more significant.

Sam Hinton: And here’s the connection I keep making: Anthropic just filed an IPO partially to maintain independence. To keep Wall Street at arm’s length. To protect their mission. But their single largest investor is one of the biggest technology companies on the planet with enormous commercial interests in how AI develops. That’s not exactly arm’s length independence.

Alex Shannon: That’s a really interesting tension. Although, to play devil’s advocate — Amazon’s interests and Anthropic’s interests might actually be pretty well aligned? Amazon wants Claude to be great because they sell it through AWS. They want Anthropic to succeed commercially. Is that actually a conflict, or is it just a big partnership?

Sam Hinton: It’s a partnership until the day it isn’t. Until Amazon decides they want to push Claude in a direction Anthropic isn’t comfortable with, or until there’s a disagreement about pricing or exclusivity or what features get prioritized. When you have a partner who owns that much of your equity, those conversations get complicated very fast. The leverage shifts.

Alex Shannon: And for the VC firms — look, they’re still making billions, right? That’s not nothing. But the gap between what they make and what Amazon makes tells you something about how the power and ownership of foundational AI companies is actually distributed. It’s not distributed the way a lot of people think.

Sam Hinton: Exactly. We talk a lot about the big tech companies and AI companies as if they’re separate ecosystems. But stories like this remind you how intertwined they are. The hyperscalers — Amazon, Google, Microsoft — they’re not just competing with AI startups, they’re often the primary investors in them. They’re inside the tent. And the Anthropic IPO just gave us a very clear window into how that actually works financially.

Alex Shannon: Keep an eye on this because when Anthropic stock starts trading, Amazon’s position is going to be one of the most watched numbers in tech. And it’s going to reshape the conversation about who actually controls the future of foundational AI.

Researchers are getting tired of OpenAI gatekeeping its breakthroughs

Alex Shannon: Let’s shift gears a little, because there’s a story brewing that I think is going to become a much bigger deal over the next year. Researchers — and we’re talking about the broader AI research community, academics, scientists who work in this space — are increasingly frustrated with OpenAI. The specific complaint is that OpenAI is gatekeeping its breakthroughs. They’re making impressive announcements, they’re demonstrating powerful research, but they’re not sharing the underlying science in a way that lets the research community build on it, verify it, or scrutinize it.

Alex Shannon: According to reports from The Verge and Wired, OpenAI announced impressive research just this week, but concerns are mounting about how limited the actual access is to the findings behind those announcements. Researchers feel like they’re being shown the magic trick without being told how it works.

Sam Hinton: OK this one hits close to home for me because it goes right to the heart of how science is supposed to work. Science advances because people share their findings, other people try to reproduce them, poke holes in them, build on them. That peer review and replication process is what separates actual breakthroughs from impressive-sounding press releases. And if OpenAI is making announcements without giving researchers the tools to do that scrutiny — that’s a real problem.

Sam Hinton: And the frustration is compounded because of who OpenAI is. This is a company that literally has ‘open’ in its name. It was founded with an explicit commitment to open research and sharing findings for the benefit of humanity. The gap between that origin story and what’s happening now is — it’s jarring.

Alex Shannon: To be fair to OpenAI though — and I want to give them a real hearing here — there’s a legitimate argument that some research genuinely shouldn’t be fully open. If you’ve developed a technique that could make it dramatically easier to misuse AI for harmful purposes, maybe you don’t publish that in a public journal. Isn’t there a safety argument for some level of gatekeeping?

Sam Hinton: There is, and it’s not a bad argument in isolation. But here’s the problem: that argument can justify almost anything. You can always say ‘we’re not publishing this because it’s dangerous.’ At some point, the research community needs more than that. They need some way to trust the system, some mechanism for independent verification, even if that’s a restricted peer review process rather than full public release.

Sam Hinton: And honestly? I think the real driver here isn’t safety. I think it’s competitive advantage. OpenAI is a company that’s now in a ferocious commercial race with Google, Anthropic, Meta, and a dozen other players. Publishing your breakthroughs openly means your competitors can use them. That’s the tension. Safety is the justification, but competitive moat is the motivation. And that’s a very different thing.

Alex Shannon: That’s a strong claim. You really think that’s what’s driving it?

Sam Hinton: I think it’s the dominant factor, yeah. Look, I’m not saying OpenAI is evil or cynical — I think there are genuinely smart, well-intentioned people there wrestling with these tradeoffs. But when you’re trying to explain behavior, follow the incentives. And the incentive to protect proprietary research from competitors is enormous right now. The safety framing is real but convenient.

Alex Shannon: So what are researchers actually supposed to do about this? Like, is there a recourse? Can they push back in any meaningful way?

Sam Hinton: A few things. One is reputational pressure — which is what we’re seeing right now. Researchers talking publicly about their frustration, publications covering it. That matters to companies like OpenAI that care deeply about their standing in the scientific community. Second is that researchers increasingly have alternatives — Meta has actually been pretty open with a lot of its research, Google publishes heavily, there’s a whole open source ecosystem. If the frustration becomes severe enough, talent and attention flow elsewhere. And the third option is regulatory — there’s a growing conversation in policy circles about mandatory disclosure requirements for frontier AI research. That could get real.

Alex Shannon: Keep an eye on this because the relationship between AI companies and the scientific research community is going to shape what we know, what we can verify, and ultimately what we can trust about the AI systems being built. It’s not a dry academic spat — it has real consequences for everyone.

Anthropic’s elaborate plan to keep Wall Street at bay

Alex Shannon: Alright, let’s close out our main stories with what I think is the most fascinating strategic puzzle of the week. We talked about Anthropic’s IPO filing, we talked about the financial stakes. Now let’s talk about the actual game plan — because Anthropic is attempting something genuinely unusual. They want to go public, they need the capital that comes with going public, but they are doing everything they can to limit Wall Street’s actual influence over the company. And Dario Amodei, the CEO and co-founder, is at the center of this.

Alex Shannon: The goal, as described in reporting from The Verge, TechCrunch, and Wired, is essentially to have it both ways: access the public capital markets while preserving the founders’ influence and the company’s values. The question everyone is asking is — can you actually do that? Is this a real strategy or is it wishful thinking?

Sam Hinton: So look, I’ve been skeptical throughout our Anthropic coverage today, but let me steelman this for a second, because I don’t think it’s impossible. The dual-class share structure that companies like Google used actually did work, more or less. Larry Page and Sergey Brin maintained meaningful control of Google for a long time. Facebook’s structure has kept Zuckerberg in command through challenges that would have removed a CEO at a conventional company. These mechanisms can have teeth.

Sam Hinton: The difference with Anthropic is the explicit values component. Google and Meta weren’t claiming their governance structure was necessary to protect a particular safety mission. They were just protecting founder control for its own sake. Anthropic is making a more ambitious claim — that their governance protects something beyond founder power, that it protects a specific approach to AI development. That’s harder to operationalize and harder to defend when it’s tested.

Alex Shannon: And it will be tested. If you’re a public company and your stock is underperforming, investors are going to ask questions. They’re going to push on every decision. How do you maintain a commitment to, say, not deploying a capability until it’s safe enough when you have analysts on a quarterly call asking why your competitors are shipping faster?

Sam Hinton: That’s the exact scenario that keeps me up at night about this. And I think the honest answer is: we don’t know if they can. No one has done this before at this scale with this explicit a mission. Anthropic is a genuine experiment. They’re trying to prove that you can build a hugely valuable AI company AND maintain meaningful safety commitments AND survive the public markets. It might work. It might not.

Alex Shannon: I want to ask you something that maybe sounds naive — but is there a world where going public actually helps the safety mission? Like, more scrutiny, more transparency requirements, more accountability through the public markets?

Sam Hinton: Oh, that’s not naive at all, actually. There’s a real argument there. Public companies have disclosure requirements. Their decisions get scrutinized in ways that private companies’ decisions don’t. And if Anthropic is public and they say ‘we’re committed to safety,’ that commitment becomes part of the official record in a way it isn’t when you’re private. Shareholders can actually hold them accountable for the things they said publicly in their filings.

Sam Hinton: But — and this is a big but — that accountability mechanism only works if safety is measurable. And right now, AI safety isn’t very measurable. You can’t point to a metric on a quarterly earnings call and say ‘look, our safety score went up.’ Until we have better ways to quantify what responsible AI development actually looks like, the accountability story is mostly theoretical.

Alex Shannon: So where does this leave Dario Amodei specifically? Because he’s the one who’s going to be standing in front of investors, standing in front of analysts, being the public face of this strategy.

Sam Hinton: Honestly? I have a lot of respect for the position he’s put himself in, even if I’m skeptical of the outcome. He clearly believes in what he’s doing. And the fact that they thought this carefully about governance before filing — rather than just slapping together a standard structure — tells you something. Whether it’s enough, we’ll find out. But I’d rather see a company trying hard to get this right than one that isn’t trying at all.

Alex Shannon: Keep an eye on this because Dario Amodei’s ability to execute this strategy — to genuinely keep Wall Street at bay while building a world-class AI company — might be one of the defining leadership tests of this decade in tech. It’s that consequential.

Google’s Genie 4 world model adds audio, more realistic worlds, and real-time steering

Alex Shannon: OK let’s get into rapid fire. First up — early reports suggest Google has released Genie 4, and it sounds like a meaningful leap. According to TechCrunch, if confirmed, this new world model can maintain consistent scenes for up to two minutes — double what Genie 3 could do — with better physics, more realistic characters, and now it adds audio generation and real-time steering. Two minutes of coherent generated world with sound.

Sam Hinton: OK so this might sound like a tech demo but let me tell you why this matters. World models are essentially AI systems that can simulate environments — and the applications are enormous. Game development, robotics training, simulation for autonomous vehicles, virtual environments for AI to learn in. The jump from thirty seconds of coherent world to two minutes sounds small, but it’s the kind of improvement that crosses thresholds where things become practically useful rather than just impressive. And adding audio is huge — suddenly you’re not just seeing a world, you’re in it.

Alex Shannon: The real-time steering piece is what catches my eye. If you can actually guide and direct these generated worlds as they’re happening, you’re moving from a generator to an interactive environment. That changes the use case completely.

Sam Hinton: Exactly. Watch this space closely. If Genie 4 delivers what early reports suggest, this is one of those quiet releases that looks unassuming and then turns out to be foundational to a whole category of applications we haven’t fully imagined yet.

SpaceX locks in $40B GPU financing deal with Apollo

Alex Shannon: Next — and this number is genuinely staggering — early reports suggest SpaceX has secured a forty billion dollar GPU financing deal with Apollo, the investment firm. If confirmed, this is an alternative financing arrangement to fund SpaceX’s rapidly expanding AI ambitions. Forty billion dollars. For GPUs.

Sam Hinton: Let that sink in for a second. Forty billion dollars. To buy graphics processing units. This is a number that illustrates more clearly than almost anything else just how capital-intensive the AI compute race has become. And the fact that SpaceX is in this race at all is worth noting — this isn’t just an AI company, this is a rocket company making a massive bet that AI infrastructure is their next frontier.

Alex Shannon: The Apollo financing angle is interesting too — this isn’t traditional debt markets, this is a creative financing structure. Which suggests even SpaceX is finding forty billion dollars is a number that requires some creative solutions.

Sam Hinton: If this holds up — and the caveat is important, this is single source — it would represent one of the largest single AI infrastructure financing deals ever. And it signals that the GPU compute war isn’t slowing down. It’s accelerating. Whoever has the most compute, wins. That’s the operating theory, and everyone is acting on it.

Disruptive raises $10B fund to make big bets on at least 10 companies

Alex Shannon: Third in rapid fire — early reports suggest a firm called Disruptive has raised a ten billion dollar fund, targeting at least ten companies in tech and AI. You might not know the name, but Disruptive has previously backed Groq, the chip company that’s been making noise in AI inference, and ElevenLabs, which is one of the leading AI voice generation companies out there.

Sam Hinton: OK so the portfolio tells you something really important. Groq is about compute efficiency — making inference faster and cheaper. ElevenLabs is about the voice layer of AI applications. These are picks-and-shovels bets, infrastructure plays. Not betting on one frontier model to win, but betting on the layers that all the frontier models need. That’s a smart portfolio thesis.

Alex Shannon: Ten billion for at least ten companies means they’re writing massive checks. These are going to be transformative investments for wherever they land.

Sam Hinton: Yeah, and it’s another data point in a week full of enormous numbers — forty billion here, ten billion there. The capital flowing into AI right now is almost incomprehensible. If even a fraction of these bets pay off, the next decade of AI is going to be very different from this one. But if the bets go sideways, the correction is going to be just as dramatic. We’re in a high-stakes moment.

VisionQuest is what happens when AI really goes rogue

Alex Shannon: And finally in rapid fire — a little culture break. According to The Verge, Marvel has a new series called VisionQuest, featuring a superpowered robot protagonist who explores what it means to be human through elaborate digital daydreams — daydreams that are shared with other artificial intelligences. And the show apparently leans into themes of AI autonomy and consciousness.

Sam Hinton: I find this genuinely fascinating as a cultural artifact. The fact that Marvel — the biggest entertainment franchise on the planet — is making a show that’s explicitly about AI going autonomous and exploring its own consciousness tells you where the cultural conversation is right now. This stuff is penetrating pop culture in a deep way. And the framing of AI sharing internal experiences with other AIs — that’s not just fiction, that’s a concept researchers actually think about.

Alex Shannon: It’s interesting because a few years ago AI in pop culture was mostly villains. Now we’re getting stories about AI trying to understand humanity. The cultural framing is shifting.

Sam Hinton: And those cultural frames matter. How we collectively imagine AI — as threat, as tool, as entity seeking meaning — shapes how we regulate it, how we build it, how we relate to it. VisionQuest might just be a superhero show, but superhero shows have outsized influence on how people think about technology. Don’t sleep on the culture stuff.

BIGGER PICTURE

Alex Shannon: Alright. If you zoom out and look at everything we covered today — the Anthropic IPO, Amazon’s massive stake, the researcher frustration with OpenAI, the forty billion dollar SpaceX compute bet, the ten billion dollar fund — what’s the pattern? What’s the thread that connects all of this?

Sam Hinton: The pattern, to me, is concentration. Power and capital in AI are concentrating faster than almost anyone anticipated. The hyperscalers are embedding themselves inside the AI companies. The AI companies are accessing public markets while trying to maintain founder control. The compute infrastructure is being locked up in deals of almost incomprehensible scale. And the research is being held closer to the chest. Everything is concentrating.

Alex Shannon: And that concentration is happening right at the moment when the technology is becoming genuinely consequential. These systems are being used at enterprise scale, they’re being integrated into critical workflows, they’re shaping how research gets done and how businesses make decisions. The stakes of who controls this are very high.

Sam Hinton: Right. And the OpenAI researcher story and the Anthropic governance story are actually two sides of the same coin. On one side, you have a company — OpenAI — that’s concentrating knowledge and keeping it proprietary. On the other side, you have a company — Anthropic — trying to figure out how to maintain its values while accessing concentrated capital. Both stories are about who gets to make the decisions about how the most powerful AI systems in history get built and deployed.

Alex Shannon: So what should people be watching over the next six months? If you had to pick one bellwether indicator — one thing to watch that tells you how this is all going to shake out?

Sam Hinton: Honestly? Watch Anthropic’s first few earnings calls after the IPO. That’s the test. That’s when we find out whether their governance structure is real or ceremonial. Watch whether the language around safety gets softer, whether the commitments get hedged, whether the mission starts getting quietly subordinated to growth metrics. If Anthropic can make it through the first year as a public company without meaningfully compromising on safety — that’s genuinely hopeful. That means you can do this. If they can’t, it tells us something profound about the limits of what corporate structures can protect.

Alex Shannon: The bigger question underneath all of this, the one that doesn’t have an answer yet, is whether the people building the most powerful AI systems are going to be accountable to the people most affected by those systems. Right now, based on what we covered today, the answer seems to be: not very. And that’s the thing worth sitting with.

Sam Hinton: And that’s not a counsel of despair. It’s an invitation to pay attention. The decisions being made right now — in boardrooms, in governance documents, in investment term sheets — are decisions that are going to shape AI for decades. The fact that we can see them, talk about them, push on them — that matters. So yeah. Pay attention.

OUTRO

Alex Shannon: That’s going to do it for this week’s Build By AI. Genuinely one of my favorite episodes to dig into — there was a lot there. Thanks for spending your Wednesday with us.

Sam Hinton: Yeah, this was a good one. And remember — weekly now, every Wednesday. If you haven’t subscribed yet, now is a great time. We’re going to keep bringing you the week’s most important AI stories with, you know, slightly more arguing than your average news show.

Alex Shannon: Slightly. We’ll see you next Wednesday. Stay curious, stay skeptical, and keep building.

Sam Hinton: See you then.