Narrator: If you keep an eye on tech news, there is 100% chance that you saw something about Anthropic in the last few weeks. Andre Karpathy joins Anthropic. The Andre Carpathy, OpenAI co-founder, former Tesla AI director together with Carpathy, Instagram's CTO. The Vatican of all places has Anthropics co-founder stand beside the Pope. Project Glass Swing, [music] KPMG contract, the most powerful model on the planet that was only released to a handful of users, $65 billion [music] series H, the S1 filing, the stakes are rising. Doesn't this remind you of something? Recruiting the world's leading scientists, government backing and conflict, industrial mobilization, rival programs, the race to the big moment, which looks awfully similar to the Manhattan project, the top secret US government program during World War II that produced the world's first [music] atomic bomb with one important distinction. Anthropic didn't give the bomb to the government, but instead is building its own private, moral, and institutional authority outside of the US with the brightest brains of tech, the Vatican, [music] and the full grasp of the enterprise market. But why are they doing this? Why now? What are they getting ready for? The real question is why does a startup need the world's best researchers, the Vatican, enterprise dominance, and its own moral framework unless it's preparing for something much bigger than an IPO. In today's episode of the business of AI, let's dive in.

Narrator: On February 27, 2026, President Trump ordered all federal agencies to seize using Anthropic Stack immediately and designated Enthropic a supply chain risk to national security. This label was previously reserved exclusively for foreign organizations, [music] specifically Chinese companies, suspected of espionage. This followed a standoff in which the Pentagon demanded that Anthropic remove two restrictions from its government contract. The restriction that banned the government from using Claude for fully autonomous weapons without human oversight and using Claude for mass [music] domestic surveillance of American citizens. The negotiations broke down over Pentagon's insisting on keeping the ambiguous language such as as appropriate. And we all know what as appropriate means. It means we'll do whatever we decide to do and later say that after careful consideration, it was deemed appropriate. In fact, after 9/11, Congress passed broad authorizations for surveillance and counterterrorism as [music] necessary to protect national security. But in practice, those words were used to justify collection of Americans phone records, email metadata, and internet traffic for years. Anthropic said no. And as soon as they did that, within the next few hours, OpenAI steps in and takes the contract from the Pentagon. What is important here is that Anthropic was not expelled. [music] They exited the deal willingly. They knew the consequences. [music] They were the only frontier AI lab with access to Department of Defense data. And giving that up was not easy. Anthropics executives knew that it could slash their revenue by billions of dollars. Every other AI lab that was offered a government contract either cooperated with the Pentagon or avoided the confrontation. And yet [music] they did it anyway. The consequence of this action was the revenue risk, which is a big problem when you're preparing for IPO, and being labeled a national security threat. And this is how Anthropic knew that they needed to find legitimacy beyond the US government. They had to jack up enterprise revenue to offset the loss of the federal contract and they had to recruit the best talent before competitors could exploit their vulnerability. And so the Vatican partnership, KPMG, Carpathy Hire, and the European expansion were planned.

Narrator: If you're not using Claude to ship your products faster, you're missing out. One of the biggest trends in tech right now is that engineering is no longer the bottleneck. A feature can be shipped in a week. Development cycles are getting shorter and shorter, and perhaps the most important factor for tech products today is time to market. But marketing teams were never trained for this kind of speed. [music] For every launch, you need launch assets, product videos, campaign creatives, sales collateral, and visuals for social media. And the speed is becoming a real problem in the AI first world. To solve [music] this, Hicksfield built an MCP integration with Claude to speed up your marketing and creative teams so they can ship as fast as engineering. To set it up, all you got to do is create an MCP connection and Claude and get to work. Imagine your team is launching a new security product. You have the positioning, the messaging, and now it's time to create campaign [music] concepts. Through Hicksfield MCP, you can generate all kinds of visual assets to support the product launch. Social media videos for [music] various platforms in any size or format. Images of any kind with highquality assets, mood [music] boards to inspire your team or your customers, ads with different scripts and filming angles, all kinds of product demos, and the full suite of creative production without leaving claude. [music] With this tool, you can really compress the time between building something and getting it in front of your customers, which gives you a very real advantage among your competitors. If you want to see how Hexfield's MCP integration works with Claude, check out the link below. And now back to Anthropic.

Narrator: On May 19th, 2026, Andre Carpathy joins Anthropic's pre-training division. First of all, Carpathy had no obvious reason to return to a frontier lab. He was a co-founder of OpenAI and left in 2024. [music] He is a founder of Eureka Labs. So an entrepreneur with academic career. He is a lecturer, a speaker, a researcher. Why go back to the lab? The significance of Carpathy's presence anthropic is in the function that he was hired to lead. [music] Pre-training is the most expensive and compute inensive stage of building a new model. [music] But instead of having a best-in-class researcher to lead the pre-training division, they hired him to build a team that will use Claude to do research on what can make pre-training faster. This is the key. Harpathy's team is going to be researching how a model can create its own successor. [music] If you draw the parallel here, just like the Manhattan project built the pipeline for rapid design and mass production of weapons, because the Manhattan project was not just about building one bomb, it was about laying the bricks for the factory that would build every bomb after it. Carpathy was hired to do the same for AI. In addition to Carpathy, they hire Mike Kger, co-founder and CTO of Instagram. OpenAI co-founder John Schulman, OpenAI safety [music] chief Yan Leica, former CTO of Workday and VP of Google Cloud Peter Bis. Doesn't this remind you of something? The US pulling in the best physicist, the best engineers, and the best researchers into one lab in Los Alamos. [music] Anthropic is doing the same, but for AI. They're pulling in the openheimers of this generation under one roof. And of all these people, Carpathy stands out because he did not have a direct interest in going back to the lab. So why him? Because [music] Carpathy already proved the idea that Anthropic wants to build. He proved it on a small model back in March. Anthropic [music] is now taking that same research loop and cranking it up to industrial scale with their most powerful unreleased model, Mythos. And why are they doing that? Because in a frontier race, a faster research that doesn't require more compute is [music] the competitive edge. Open AAI has more compute capacity and a larger consumer base. Google DeepMind has TBU chips and search monopoly. Nvidia goes without saying, Amazon has cloud and chips. So if compute capacity, B2C market, hardware, and search engine are monopolized, what's left [music] is the quality of research and specifically finding a way to squeeze as much as you possibly can out [music] of the existing amount of compute. Carpathy's mandate using Claude to do AI research on itself is Enthropic's answer to the compute gap. [music] His own research will be validated in real time because Enthropic uses Claude to write 90% of its code [music] and Carpathy's team will push this even further. AI doing AI research.

Narrator: Carpathy joined exactly one week before anthropic series H [music] announcement and 13 days before the S1 filing. What a coincidence. Why now? Because institutional investors evaluating an AI lab at $1 trillion need a solid proof why this lab will remain competitive when it doesn't have the leading compute capacity or leading chips. And this is why Carpathy's [music] hire is the signal that Enthropic found the next big vector of development in AI. What comes next is Project Glass Swing. April 2026, a coalition of 50 and then 150 companies receives [music] gated access to Claude Mythos preview. If we use the Manhattan Project analogy, Claude Mythos is the secret weapon of Anthropic. They call it by far the most powerful model it has ever developed. In a leaked blog post, the company warned that Mythos is currently far ahead of any AI model in cyber security capabilities. This is the first time a major AI company issued a warning that it has a model too dangerous to release. The reason they released it to a coalition is because Anthropic itself acknowledges that the model changes the threat level. They're giving a beast of a model to companies that control global AI infrastructure. They integrate Mythos in the cyber security stacks of every major cloud provider. They build relationships with companies and countries and position themselves as the only frontier lab that has both the capability and the ethical framework to manage this kind of power. The day of project glasswing announcement, [music] cyber security stocks dropped around 10%. Which yes is a very normal thing after an announcement of this kind. But just think about it. Crowd strike is simultaneously a glasswing partner and a company whose core value proposition is being disrupted by the technology it's using. [music] And here is the most important thing. The significance of project glasswing for the S1 is that Mythos is a [music] product that is not yet released. Investors are pricing a company at almost a trillion dollars based on the revenue from Claude and that is while they still have an ace [music] up their sleeve. Mythos is their IPO asset, the most powerful model in the world that is not even released yet. [music]

Narrator: Now, if you think that this is the first time that a private company sat on something too scary to unleash, [music] it isn't. In 1974, Kodak, yes, the camera company, installed a nuclear device [music] in the basement of their Rochester headquarters, a small research reactor about the size of a fridge with [music] weaponsgrade uranium. No, they weren't trying to blow anyone up. [music] They just needed an extremely precise neutron source. so they could spot impurities in the film. But the city didn't know, the police didn't know, [music] the fire department didn't know. It was, of course, a lot less dangerous than a real reactor, but it wasn't nothing. The thing was running there for 33 years until an ex employee leaked it to the local newspaper. The difference is [music] Kodak never told anyone. Anthropic, on the other hand, published a blog post about it. That distinction is the whole point of Project Glass Wing. They are letting the public know, but they're doing it in beta.

Narrator: May 19th, 2026, the same day as the Carpathy hire announcement, Entropic announced a global alliance with KPMG. As part of this deal, they embed Cloudco and agents into KPMG's platform for tax clients and private equity firms. Let's dissect this one. What does it mean? This is about the distribution. A strong grasp of the distribution is the bread and butter of tech. Anthropic [music] chooses to distribute to professional services. This is important. KPMG is not a regular client or an enterprise contract. KPMG becomes [music] its own distribution channel. Remember how in earlier videos we analyzed the difference between the business models of OpenAI and Anthropic and the importance of the B2B game for Anthropic's Rise? This is exactly why they're doing it. APMG has 138 countries worth of client relationships in tax audit advisory and private equity. And [music] what is tax audit and private equity? The money managers. This is the layer that brings financial reports to the sea level execs and recommends how to use the money. Through KPMG, Anthropy gains indirect distribution to an enormous slice of the private economy. A single KPMG client that deploys clot for tax compliance will generate recurring revenue with much lower churn risk than regular subscriptions from individuals like us. Selling to individual consumers is volatile by nature, but enterprise contracts are sticky, low churn, high margin, and predictable. Using the Manhattan project analogy, Antropic already has the Loss Alamos, where technology is built. And KPMG is what happens when you take the Los Alamos and turn it into a franchise business and you sell that franchise to a shop that sells plug-and-play kits to top tier tax lawyers, auditors, CFOs, boards, investment committees who have access to massive cash flows. In other words, Anthropic is selling to the people who decide which companies live, which companies die, and who gets the funding.

Narrator: Good lord, this one gave me a mad chuckle when I saw it on LinkedIn. On May 25th, 2026, [music] Pope Leo releases Magnifica Humanitus. In a nutshell, it's a historical moment for the [music] church when the head of the Catholic Church releases a document about AI. [music] Now, what's interesting here is that the document was presented with Christopher Ola, who was standing beside the pope. No other AI lab was at the podium. Amazon, Meta, and Google's representatives met with the Vatican officials before the ceremony, but Anthropics co-founder was the one who stood at the presentation. Here's what's peculiar. Christopher Ola is not the most recognizable figure atropic. That would be Daario Amade. Ola is the co-founder and the scientist who studies what is actually happening inside a neural network. The Vatican's choice of Ola over Amadeay was because they needed a figure who had scientific credibility. In other words, they did not need the business person on stage. They needed the science person for better persuadability. Ola is one of the few people in the world who can credibly argue from first principles why AI development requires external oversight. And he did say during the presentation that companies like his operate under strong commercial, geopolitical and personal pressures that can be at odds with the broader interests of [music] society. So you look at the situation, you see the anthropics co-founder on stage at the Vatican and you may wonder why the Vatican of all places because it does look like a weird move. First of all, the US is mostly Protestant. Anthropic is an American company. But the thing is this partnership isn't about faith. The Vatican is a [music] single institution that can talk to Brazil, Poland, the Philippines, and Brussels in one move. The Vatican has diplomatic relations with 183 countries. When the church makes its first official statement about AI, and you're standing on that stage, you're tapping into legitimacy [music] that is perpendicular to the US federal politics. It cannot be revoked by the US administration. This is especially significant in the context of the federal ban that they got in February 2026. Think about it. The same week Trump called Anthropic left-wing nut jobs, the Pope chose Anthropic's co-founder to stand beside him in Rome. Anthropic now occupies the position of being banned by the US government and endorsed by the head of the Catholic Church. This is called diversification, ladies and gentlemen. But jokes aside, why would an AI lab have a connection with church? Because of the ESG mandates. That's why. Hear me out on this one. Institutional investors, [music] pension funds, sovereign wealth funds, university endowments [music] have ESG mandates. A near trillion dollar AI company IPO that carries both a federal ban designation [music] and Vatican moral endorsement is a very unusual duality. The Vatican partnership is arguably the most credible counternarrative to the dangerous company banned by the US government. They used the reputational risk and turned it into a differentiation story. There is nothing I wouldn't give to know who came up with this whole plan. But if this came from Anthropic's marketing [music] department, I take off my hat and bend my knee for you.

Narrator: And finally, May 28th, Enthropic closes a $65 billion series H round with post money valuation of $965 [music] billion, the largest venture capital round in history that catapulted Enthropic past OpenAI to become the most valuable private AI lab in the US and in the [music] world. And on June 1st, Anthropic filed a confidential S1. The revenue trajectory that Anthropic is on [music] justifies the valuation. $9 billion in December 2025, 14 billion by February, 19 billion by the end of February, 30 billion in April, 47 billion in May. Dario Amade told investors that the company had planned for 10x growth in 2026, but instead reached a growth rate that could make it 80 times larger. There's only one problem with the valuation in the IPO story for Anthropic, the compute bill. The Space X S1 filed the same week and they reveals that Enthropic has committed to pay $1.25 billion per month for GPU compute through May 2029, which is approximately $15 billion per year in raw compute costs. That's without counting additional GPU commitments with AWS or Google. At $47 billion revenue with a projected 559 million quarterly operating profit, Anthropic's unit economics are real. But the margin is [music] thin relative to the infrastructure dependency. But even with these numbers, Anthropic is perhaps the only frontier lab, the only American lab where the math is mathing. [music] The unit economics paradox for Frontier Labs is a stumbling block. They monetize quickly and they reach operating break even. And at the same time, they're locking themselves into enormous multi-year commitments that make fullear profitability very difficult without continuous rapid revenue growth. Which is why [music] Karpathy's research using AI to do AI research faster [music] is the biggest leverage in the IPO game because if AI assisted pre-training can reduce the required compute capacity, it will expand the margins. Every single move in the last 90 days, the federal ban, the hiring spree, the Vatican, KPMG, Glass Wing, the S1, all point in the same direction, and it's much bigger than an IPO. Anthropic is building the institution that governs what comes after the IPO. The question is whether they will succeed. We hope you found this interesting and enlightening. We would appreciate your feedback and we'll see you in the next episode of the business of AI.