In the Loop AI

Anthropic IPO targets $1.5T valuation with $90B raise amid $1T CapEx challenge and rapid revenue growth

42 min Listen to the original Episode page
IPOArtificial IntelligenceVenture CapitalEnterprise SoftwareCybersecurityBiotechCapExMarket Valuation

Summary

Speakers detail Anthropic's IPO mechanics, noting an 825 million share offering priced at $70-$78, aiming to raise $90 billion total, which would surpass SpaceX's record. The company reports explosive revenue growth from $300 million to $17 billion year-over-year, projecting $300 billion by 2028, while facing a $1 trillion five-year CapEx bill. The bull case highlights 900% net revenue retention and enterprise adoption, whereas the bear case cites the massive capital expenditure requirement and potential slowdown in data center build-outs. Speakers also discuss strategic bets on Recursive Self-Improvement, autonomous labs, and new consumer applications like 'Clawed Office'.

Key takeaways

  1. 1
    trajectory that Anthropic has been on. Remember, in the first two years of Anthropic's life, it didn't make any money. When you look at their revenue in Q Q1 one of twenty 2025, twenty five, which is about a year and a half ago, they were only making around three hundred $300 million. Fast forward to Q3, and they made $17 billion. And they're set to hit something insane like $300 billion of

    Unidentified speaker (speaker_0) · 5:02

  2. 2
    listen, it makes a modest $41 million in 2023. It then makes around $380 million in 2024. And then in 2025, they make around, I think, $1.4 billion. So, you know, a very steep increase. But it's at the end of 2025 where they obviously hit their golden product, Claude clawed Code.

    Unidentified speaker (speaker_0) · 13:19

Episode sections

  1. 0:00 – 3:07

    The Biggest IPO Ever

    Speaker_0 introduces Anthropic's IPO as the biggest ever, with a ticker of ANT and a valuation targeting $1.4-1.5 trillion. The IPO is described as 6-10x oversubscribed, aiming to raise over $500 billion, surpassing SpaceX's record. The speaker highlights Anthropic's rapid revenue growth to $100 billion ARR and its dominant position with the Claude product, while noting the massive $1 trillion CapEx challenge and competition from OpenAI, Google, and Meta.

  2. 3:07 – 5:01

    The IPO by the Numbers

    The speaker details the IPO mechanics: 825 million shares offered, with a price range of $70-$78 per share. The total raise is estimated at $90 billion ($65 billion IPO + $26 billion private placement), beating SpaceX's $85 billion. Brad Gerstner of UltimateX is cited as placing a $5 billion order. Google and Amazon hold a combined 28% stake and are contributing to the private placement.

  3. 5:01 – 6:12

    From $376 Million to $17 Billion

    Anthropic's revenue trajectory is highlighted, growing from $300 million in Q1 2025 to $17 billion in Q3 2025, with a projection of $300 billion in revenue by 2028. The speaker notes that despite this growth, the company faces a $1 trillion CapEx bill over five years, with only half committed. The cash balance post-IPO is estimated at $212-$220 billion, which is more than most Mag 7 companies.

  4. 6:12 – 8:27

    The Self-Improving AI Bet

    The speaker discusses two key innovations: increased compute efficiency (more money made per dollar of compute) and Recursive Self-Improvement (RSI). RSI is described as AI designing its own successors. The speaker references an AI 2027 report predicting white-collar work replacement by 2027. OpenAI, Google, and Anthropic are all reportedly close to achieving RSI, with Anthropic focusing on biology and medicine.

  5. 8:27 – 16:45

    The $1 Trillion Check

    Anthropic faces a $1 trillion CapEx commitment over five years, with $212-$220 billion in cash post-IPO covering only ~20%. The speaker questions how the remaining $800 billion will be funded through revenue growth or new sectors. The company's history is traced back to seven OpenAI defectors, including Dario Amodei and Tom Brown, who co-authored the 'Scaling Laws' paper. The first funding round in April 2022 was led by Sam Bankman-Fried.

  6. 16:45 – 18:02

    Seven Founders, Still Together

    All seven co-founders remain at Anthropic, unlike OpenAI where most have left. The founders are billionaires pledging 80-90% of their wealth to public benefit corporations, with Anthropic committing 10% of the company to public benefit over the next decade. The founders are selling approximately $1.5 billion in shares at the IPO.

  7. 18:02 – 23:42

    The Bear Case

    The bear case includes the $1 trillion CapEx bill, NVIDIA's leadership crisis after Jensen Huang's resignation and $2 trillion market cap loss, and slowing data center build-outs due to US midterms and regulatory pressure. SemiAnalysis predicts a 45% slowdown in data center capacity. Tight credit markets and risk-off sentiment from investors like Stanley Druckenmiller further complicate funding. NVIDIA's Q3 revenue of $151 billion was met with stock declines.

  8. 23:42 – 33:14

    The Bull Case

    The bull case rests on Anthropic being the fastest-growing company ever (1121% growth in 2025) and targeting the enterprise market. Net revenue retention is 900%, up 3x in five months. Customers spend less than 0.1% of their annual budget on Claude, suggesting massive upside. The speaker praises Claude's 'taste' and usability compared to GPT-6.2 Astro and Gemini. Compute efficiency is improving, with $1.72-$2 returned for every $1 spent on compute.

  9. 33:14 – 34:45

    Cybersecurity and Knowledge Work Opportunities

    The speaker argues that AI usage will grow beyond search queries as consumer apps like 'Clawed Office' replace traditional software. He identifies cybersecurity as a major growth area, noting increased hack frequencies and the need for AI-driven defense models.

  10. 34:45 – 38:10

    Medicine and Drug Discovery

    Anthropic and OpenAI have spent over $10 billion acquiring scientific data to train models for drug discovery. The speaker highlights efforts to automate science, citing Google's Isomorphic Labs and the potential for AI to design personalized cancer treatments.

  11. 38:10 – 39:40

    Robotics and Autonomous Labs

    The speaker discusses the shift toward robotics, noting OpenAI's GPT-6.1 excels in robotics evaluations. He mentions Anthropic is building an autonomous wet lab to test AI-designed drug formulas, aiming to solve data constraints in pharmaceutical research.

  12. 39:40 – 42:02

    RSI and Final IPO Details

    The speaker briefly mentions Recursive Self-Improvement (RSI) as a future possibility where AI surpasses human experts. He concludes by detailing the IPO structure: 825 million shares, a cliff unlock schedule, and the potential to create many centi-millionaires among the 5,000 employees.

Assets discussed

AssetKindStanceContext
Anthropicprivate companyneutralcap would probably be a quarter Anthropic, which is just insane to say. Now, fast forward to today, and all seven co-founders are still there at Anthropic, which is impressive beyond means because just generally in any startup, if you have multiple co founders, it is a very high attrition rate. You're likely to
OpenAIprivate companyneutralhave at least one leave. If you look at OpenAI, I think the majority of the co founders have left and it's just Sam and maybe one other. And then when it comes to anthropic, the fact that they were able to cultivate the right sort of mission, culture. and I guess camaraderie and keep all their co-founders there is a high
SpaceXprivate companyneutralSpaceX pivoted recently. Um they're still doing the space stuff, and they recently caught in a very impressive rocket, which I made an episode about go check the that out. But they've also become the number one AI cloud provider. And Elon Musk's take was we were only able to pull this off because we have investment grade financing. In other words, we're able to borrow as much money as

“Stance” is how the speakers talked about the asset, not a recommendation.

Claims worth verifying

Anthropic's IPO is the biggest ever, with over $500 billion in demand and 6-10x oversubscription.

Unidentified speaker (speaker_0) ·0:20

How to verify: SEC S-1 filing for Anthropic (ANT) and press releases from IPO underwriters. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Anthropic aims to raise $90 billion total ($65 billion IPO + $26 billion private placement), beating SpaceX's $85 billion raise.

Unidentified speaker (speaker_0) ·3:43

How to verify: SEC S-1 filing for Anthropic (ANT) and company press releases. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Brad Gerstner's UltimateX is placing a $5 billion order in the IPO.

Unidentified speaker (speaker_0) ·4:22

How to verify: SEC Form 4 filings for UltimateX and SEC S-1 for Anthropic (ANT). For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Google and Amazon hold a combined 28% stake in Anthropic and are contributing to the $26 billion private placement.

Unidentified speaker (speaker_0) ·4:42

How to verify: SEC S-1 filing for Anthropic (ANT) and investor presentations. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Anthropic is projected to hit $300 billion in revenue by 2028.

Unidentified speaker (speaker_0) ·5:22

How to verify: Anthropic's S-1 filing and company guidance. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Post-IPO, Anthropic will have $212-$220 billion in cash, covering ~20% of its $1 trillion CapEx bill.

Unidentified speaker (speaker_0) ·8:40

How to verify: SEC S-1 filing for Anthropic (ANT) and company financial projections. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

NVIDIA's Q3 revenue was $151 billion, the most ever in a quarter.

Unidentified speaker (speaker_0) ·22:55

How to verify: NVIDIA's Q3 2026 earnings report and SEC filings. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Net revenue retention for Anthropic is 900%, up 3x in five months.

Unidentified speaker (speaker_0) ·24:35

How to verify: Anthropic's S-1 filing and investor presentations. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Customers spend less than 0.1% of their annual budget on Claude.

Unidentified speaker (speaker_0) ·25:54

How to verify: Anthropic's S-1 filing and customer surveys/reports. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

The percentage of the US population that uses AI as a Google search query is around 5-10%.

Unidentified speaker (speaker_0) ·33:14

How to verify: Pew Research Center or similar US demographic surveys on AI usage trends; check for reports citing 5-10% penetration for non-search AI tasks. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Anthropic released 'Clawed Office' to replace Excel, PowerPoint, and Docs.

Unidentified speaker (speaker_0) ·33:34

How to verify: Anthropic's official press releases or product launch announcements; search for 'Clawed Office' or 'Claude Office' announcements. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

OpenAI has a cyber defense model called 'Daybreak'.

Unidentified speaker (speaker_0) ·35:51

How to verify: OpenAI's official website or product documentation; search for 'OpenAI Daybreak model' or 'OpenAI cyber defense'. Note: This may be a mishearing of 'Codex' or a specific internal tool; verify against public OpenAI releases. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

OpenAI has an internal model two to three times more capable of cyber defense than any other model.

Unidentified speaker (speaker_0) ·36:11

How to verify: OpenAI whitepapers or third-party cybersecurity benchmarks (e.g., MITRE ATT&CK evaluations) comparing OpenAI models to competitors. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

OpenAI and Anthropic spent upwards of $10 billion to acquire scientific data since the end of last year.

Unidentified speaker (speaker_0) ·36:51

How to verify: SEC filings (if public) or company press releases detailing data acquisition costs; check for reports on Anthropic/OpenAI data spending. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Demis Hassabis stepped down as Google CEO to focus on Isomorphic Labs.

Unidentified speaker (speaker_0) ·37:30

How to verify: Google/Alphabet press releases or Demis Hassabis's LinkedIn profile; verify his current title and role at Isomorphic Labs. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

OpenAI's GPT-6.1 exceeds all evaluations in robotics.

Unidentified speaker (speaker_0) ·38:30

How to verify: OpenAI's technical reports or third-party robotics benchmarks (e.g., DROID, RT-X) comparing GPT-6.1 performance. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Anthropic is building an autonomous wet lab to test AI-designed drug formulas.

Unidentified speaker (speaker_0) ·39:09

How to verify: Anthropic press releases or news articles about 'Anthropic autonomous lab' or 'Claude wet lab'. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

The Anthropic IPO involves 825 million shares.

Unidentified speaker (speaker_0) ·40:24

How to verify: S-1 filing with the SEC for Anthropic's IPO; check the 'Number of Shares Offered' section. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Anthropic has 5,000 employees.

Unidentified speaker (speaker_0) ·40:44

How to verify: Anthropic's S-1 filing under 'Management’s Discussion and Analysis' or 'Human Capital' section; verify employee count as of the filing date. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Do your own research

  • Review the SEC S-1 filing for Anthropic (ANT) to verify the $90 billion raise structure, $70-$78 price range, and 825 million share count.
  • Check SEC Form 4 filings for UltimateX and other institutional investors to confirm Brad Gerstner's $5 billion order and Google/Amazon's 28% stake.
  • Analyze Anthropic's audited financial statements in the S-1 to validate the $17 billion Q3 2025 revenue and $1 trillion CapEx commitment.
  • Verify the 'Scaling Laws' paper authored by Dario Amodei and Tom Brown to understand the technical foundation of their model scaling strategy.
  • Research the 'AI 2027 report' cited regarding white-collar work replacement timelines and RSI feasibility.
  • Compare Anthropic's enterprise net revenue retention (900%) against competitors like OpenAI and Google using public earnings reports or third-party surveys.
  • Investigate the regulatory and political risks mentioned, specifically US midterm impacts on data center build-outs and SemiAnalysis's 45% slowdown prediction.
  • Confirm the details of the 'Clawed Office' product launch and its competitive positioning against Microsoft Office and Google Workspace.
Full transcriptexpand

0:00Unidentified speaker (speaker_0): Ladies and gentlemen, the time has officially come. Today, a company that didn't exist six years ago is about to make history with the biggest IPO ever. That company's name is Anthropic. Its ticker is Ant, which is very ironic given that it's probably going to become one of the biggest companies in the world. And it is officially the most in-demand IPO. Over $500 billion is set for the IPO. That's half a trillion dollars and around six to ten times what Anthropic actually has available. And this means they'll raise more money than any other company in history, beating the previous record holder SpaceX from just a few months ago. And this is off the back of the fastest growing revenue of any company ever, set to hit around $100 billion ARR by the end of this year. And it makes sense. They have some of the leading models in the world, their hit product clawed code. absolutely took over the entire world and Fortune 500 companies. And of course, the question on everyone's mind is should I buy or sell this IPO? And that's where we have to get to the other side of the story, which is they have a trillion dollar check that they need to pay for AI CapEx over the next couple of years. They have a growing amount of competition from the likes of OpenAI, Google, and Mes. Meta. The stock market has been cratering over the last eight weeks, with of course Jensen Huang

1:19leaving NVIDIA and NVIDIA losing around $2 trillion market cap. The reason why that's important is because NVIDIA has been the bedrock for a lot of the CapEx and AI cycle so far. And then there's finally the question of where does AI go from here? Is it just coding? Is it just general chatbots? Or can we actually enter the age of miracles? as Dario puts it, and solve cancer, cure a bunch of different ailments, fix robotics or build robotics in the first place, and effectively create an artificial intelligence for any living and breathing bit of human work out there? It's an ambitious but also very daunting and may I say slightly scary future that we have to look into. The main thing being that a lot of it is unscathed. uncertain. So on today's episode, I want to walk you through the entire IPO, the numbers, the amount they're raising, share on locks, etc., but also what this means for Anthropic, for how far they've come, the journey that they've built over the last six years, and most importantly, where they're going next.

2:23Unidentified speaker (UNKNOWN): You see,

2:23Unidentified speaker (speaker_0): in order for this to be a successful IPO, they need to break the pattern and trend that we've seen over the last five to ten major tech IPOs, which is they've popped at the start, and then since then they've created. We can see that with SpaceX right now. So the question is: is this going to be the same repeat pattern, or will Anthropic finally break that trend? And at this point, Anthropic has a huge grip on the entire market. They're one of the largest consumers of compute. They are the one, maybe number two AI lab in the entire world. Everyone around the world that uses AI uses Claude in some way, shape, or form. So a lot of the world in general is dependent on this company. And if that becomes increasingly so, what does that mean for Anthropic's valuation going forwards? Now, first let's talk about the actual IPO. They're issuing around 825 million shares. A chunk of that is coming from the seven co-founders themselves. who are selling around, I think, $360 million at the upper end of the price range. The price range, speaking of, is going to be between $70 to $78 per share. But I hear rumors right now at the time of recording that they're probably gonna push for an extra hit there, which would value at the company at around 1.4 to 1.5 trillion. But the demand itself has been overwhelming. There's six to 10x oversubscribed. The available capital that they wanted to raise was

3:43around $65 billion with an additional private placement of $26 billion. So the total would be around $90 billion, which beats SpaceX's raise of $85 billion. So the long story short of this is that there's an insane amount of demand. So if you were looking at the events over the last couple of weeks with NVIDIA and Jensen Wong specifically, and NVIDIA shedding around what, $2 trillion since the start of October, you'd think, hmm, maybe there's not that much insatiable demand for a major IPO, specifically in AI. Turns out that's not the case at all. And in fact, a lot of that capital might actually be going to anthropic Brad Gerstner, who is one of the heads, founder, CEO of UltimateX. Ultimate Capital, actually said that his firm alone, and they are early investors in Anthropic, are putting in a $5 billion order at the IPO bell. And it's not just Anthropic's investors that are excited about this, it's also their suppliers. Google and Amazon already have major stakes in Anthropic. I think combined they have around 28% twenty eight percent of Anthropic at IPO. And they're putting in an additional, I think, 10 ten to maybe fifteen billion dollars of a private placement, twenty six billion dollars. So the idea here is there is a lot of doubling down going on on Anthropic right now. And it makes sense when you consider the insane

5:02trajectory that Anthropic has been on. Remember, in the first two years of Anthropic's life, it didn't make any money. When you look at their revenue in Q Q1 one of twenty 2025, twenty five, which is about a year and a half ago, they were only making around three hundred $300 million. Fast forward to Q3, and they made $17 billion. And they're set to hit something insane like $300 billion of revenue in 2028, the fastest growing that we've seen at any company level. But listen, let's put the numbers aside for a second and actually look at the company and the innovations that they're making. Now, if you look at every single Claude model that's dropped over the last 12 months. you'll notice one particular thing, which is weirdly, it costs less compute or it uses less compute to train. So what you end up seeing is kind of the opposite of what you would expect to see, which is the more compute you spend, the more money you end up making. The margins are increasing, which is the compute efficiency, I guess, of what Anthropic is observing with their AI models. and assuming Google and OpenAI are seeing the same thing. So, what this means is Anthropic can spend less compute and make way more money. And that's the directory for every single model release going forwards. The second innovation is this entire goal of RSI or recursive self improvement. You might have heard this term quite a bit. It basically describes an AI

6:21model that is so smart that it can design, build, and research the next version. of itself. So an AI that is smarter than the smartest AI researcher, yeah, the exact ones that are being paid hundreds of millions of dollars to work at these AI labs in the first place. They're building their own replacement. But the part that excites investors isn't the self-improving part, it's what the self-improving means, which is you can kind of point that model at anything, any profession, any vertical, any hard task. And theoretically, it should be able to figure it out. In fact, I made an episode uh, I don't know, like a few days ago, maybe last week, of uh an AI 2027 report, which are guys that created this prediction, I think back in 2025, or maybe 2024, and they had one of the most accurate predictions so far. And one of the cases was when we'll get this self self-improving improving technology. but also when white-collar work is going to be replaced by AI. And now they've moved up their dates. They were about 10 months late. And they predict this to happen as early as the end of next year, 2027. Now, in that world where you have an AI model that can automate 99% of white collar work, which makes up trillions of dollars worth of value in the economy. And let's say that leading model that

7:40does that is an anthropic clawed model. The question then becomes what do you value the company that was the creator of such a model? And I know you might hear me say this and say, you're just selling into propaganda for Anthropic, but the truth is, I've looked into the research and the data, and it's not just Anthropic that's proving it, it's also OpenAI and Google. In fact, all three of these companies have made public statements saying that they're very close to achieving RSI, which means that their AI models can start to be pointed at really hard problems. And we've seen it with OpenAI solving a bunch of really complex, hard math and physics problems. We're seeing it with Google solving a lot of complex science problems and advancing their own scientific experiments internally at the company. And then you have Anthropic that's focused on biology and medicine and cancer specifically. Now, of course. all of this assumes that Anthropic is able to overcome a number of challenges that still lay ahead of it. They have to somehow pluck $1 trillion of Capex to spend over the next five years. They have about half of that already owed and committed. The amount that they're raising, I think it'll end up with them having around $212 to $220 billion dollars in. their cash balance account, which is also just insane. I think that's more money than the majority of the Mag 7 companies that exist today have, or even

9:00less because they've been spending it all on CapEx Anthropic will have more money than them. And let's see, that'll cover maybe around 20% of what they owe. They still need to figure out where the other $800 billion dollars comes from. And that's relying on, you know, growing their revenue. But I would assume at some point their existing products will hit a limit and you're counting on anthropic to be able to keep innovating and breaking into new sectors and that's really where the uneasiness comes from it's like can they break into this future market that's a big bet to make an evaluation of 1.5 trillion dollars is a lot of money but before I go any further I want to recount kind of how Anthropic got here in the first place because that starts to answer the question. And I will, I promise you, get to it later as to whether this IPO is a buy or sell, give you the bull and bear case. But in order to understand that, you need to understand how this company was formed, who the founders are, where the talent has come from, why Anthropic has been kind of like the number one place to work at if you are an AI researcher, why we've seen some of them defect recently to go to competitors such as OpenAI, which might tell us a different story. And most importantly, where their resources are being focused. You see, for the last couple of years, I would say Anthropic's main focus has been to, well, aside from building one of the best AI models in the world, acquire as much

10:19compute as they can. And for all intents and purposes, with this IPO raise, they're gonna going to have solved a large chunk of that. They have a huge amount of cash, they can spend it. Then the question becomes where are they allocating those resources? Is it going to be acquiring endless amounts of compute, or are they going to acquire as much talent as they can to, you know, solve some of these other sectors that they're going for? As Dario has mentioned, he wants to solve a lot of major diseases within the next decade. If he wants to do that, he needs to acquire a lot of scientific data, he needs to acquire some of the best researchers from universities in the world. And then he needs to kind of build this new AI framework around all of these people. That's not going to be an easy job. But anyway, let's rewind six years ago where seven then OpenAI employees left to start Anthropic. Now these seven employees were very tenured. They had been working in AI research and machine learning before it was sexy, and they worked at OpenAI for a considerable amount of time. And Dario

11:19Amode specifically was the VP of research. Tom Brown, who was another employee that left to found Anthropic, was the lead author and researcher on GPT 3, which, if you remember back then, was an absolutely mind blowing model that made the entire thing go goes very viral. Then, of course, Daniela Amode is Dario's sister. All of these guys were in. some way an author or contributed very significantly to this paper called Scaling Laws. And scaling laws became the fundamental trick or proprietary formula to build modern day LLMs. Now, what this basically means in a very distilled form is you take as much compute as you can, you combine it with as much data as you can. and out pops the best model. And the bet that that paper was making was if you acquire the most compute and you have, you know, a bunch of good researchers that can like mold the taste of a model, you'll end up with the best model. And that was Dario's entire bet. Now back then OpenAI wasn't really as aggressive when it came to compute or they didn't really believe that, you know, if you kept shoving compute at the thing, both in pre-training and post-training, you you'd be able to build the best model. But it ended up being the case. And that's what Dario and his co-founders wanted to prove in the initial years of Anthropic's founding. Now, for the first couple of years, not much actually happens until April

12:39of I think 2022, when Anthropic raises their first round. And it's a big round. It's around a $580 million round, I believe. And it's led by none other than Sam Bankman Friedman. Fried, the infamous founder of FTX. Now, fun fact if he had kept his stake at the time, you know, if none of this fraud stuff went down, if he wasn't caught or whatever, or if he just, you know, was a sane normal person, his stake would be worth over a hundred billion dollars today. Then in November of twenty twenty two, Chat GPT launches and it becomes a viral sensation immediately. But it's not until twenty twenty. 2023 that Anthropic drops their first model, Claude. And listen, it makes a modest $41 million in 2023. It then makes around $380 million in 2024. And then in 2025, they make around, I think, $1.4 billion. So, you know, a very steep increase. But it's at the end of 2025 where they obviously hit their golden product, Claude clawed Code. I remember it coming out at the time, and my software engineering friends were saying, hmm, good attempt, but AI will never replace coding. It is one of the hardest art forms and skills. I went to college. I've been studying this for well over a decade. I have the experience that most people won't be able to pick out, and there's no way that AI will be able to automate it. And indeed, that was the feedback

13:59for I want to say eight weeks, first eight weeks of the product. launch, because of course the product improved massively. And software engineers, starting with the engineers within Anthropic and then AI adjacent, and then technology in general, started to realize, oh crap, we can now automate a lot of the level of thinking and work that we would pay engineers uh are annually around $200,000 a year. And that's really when the company started to really gain momentum. The irony of that being that was nine months ago, right? And so at the start of the year, anthropic clawed code is flying. And most importantly, they're working on a secret internal model that is meant to be breaking edge, the most frontier powerful model that they've seen so Safari, far. In fact, so powerful that it has cyber capabilities that we wouldn't want to put in the hands of the general. public, citing that it could be used for dangerous causes. And this is where the emotion and sentiment of being scared of AI really started. It's like, ugh, what if this AI doesn't end up creating really good benefits for human society? What if it could be used against us? What if it could create a bunch of hacks? There was a lot of fear mongering at the time. And it wasn't helped by anthropic getting involved with the government and war. specifically. Their models were being used by the Pentagon and they ended up getting blacklisted.

15:19And it was a whole drama, which I'm not going to bother with right now. But the idea here is Anthropic went from zero momentum, year one, still zero momentum, year two, and again zero momentum, and then just a straight spike up fueled by Claude clawed Code. Now it wasn't just companies that were sitting around and watching this. You had Amazon that was trying to train their own AI models. And remember, Amazon is Amazon. They have billions of dollars worth of money that they can spend to do this. Google did the same thing. And of course, OpenAI is a startup. But I'm talking about the larger companies here. They realized, oh crap, we might be a little bit too late. But don't worry, it's not a problem that money can't solve. And so they started throwing billions of dollars meta infamously doing the same thing. And so they thought, okay. if we spend enough amount of money and compute, we'll be able to train our own models. But there were also hedges being made. And that was around the time that Amazon and Google took major stakes in Anthropic itself. I think Amazon's first investment was in 2023, and Google, I don't know when, but collectively they have around 28%

16:19of the entire company, which is weird. I guess you could kind of think of Google and Amazon as a proxy investment hold for. Anthropic. And maybe these stocks get re-rated because of that. Like, let's assume that Anthropic hits a $2 trillion market cap. Maybe it's already hit it by the time of this recording. Amazon's entire market cap would probably be a quarter Anthropic, which is just insane to say. Now, fast forward to today, and all seven co-founders are still there at Anthropic, which is impressive beyond means because just generally in any startup, if you have multiple co founders, it is a very high attrition rate. You're likely to have at least one leave. If you look at OpenAI, I think the majority of the co founders have left and it's just Sam and maybe one other. And then when it comes to anthropic, the fact that they were able to cultivate the right sort of mission, culture. and I guess camaraderie and keep all their co-founders there is a high testament to what they're building or the team nature that they were able to build. And now they're all billionaires and they are pledging their wealth to public benefit corporations, saying, hey, we're gonna give away 80 to 90 percent of our wealth because we don't need it. That's an unofficial Official statement from all of them. But you know, if they do end up committing, that would be crazy. In fact, Anthropic as a company was or a public. as

17:39benefit corporation has already committed around, I think, collectively 10% of the entire company over the next decade to go to the public benefit. The uh, you know, the disclaimer, of course, being that they decide where that money goes to, but it's still a noble cause. And I think the founders are collectively selling around $1.5 billion at this IPO open. Which brings me to should you buy? or sell this IPO? First, let's look at the reasons not to. One, the $1 trillion check that they need to pay off over the next couple of years is an insane amount of money, the most amount of money that any singular company is committed to. Even if you compare it to like Amazon and Google, who have actual separate business lines that generate tens to hundreds of billions of dollars. every year. Even they have committed roughly around 200 to maybe 500 billion dollars. Anthropic is almost twice that. And then you have NVIDIA. Jensen Huang resigned in a crazy scandal in Coupe, which we've covered in previous episodes. But the idea here is NVIDIA has shed around $1.76, let's scall call it $2 trillion

18:48over the last eight weeks. And the reason why this is a important is, hey, you know, a lot of that compute that Anthropic needs to buy, a lot of that $900 billion, it needs to be spent on majority NVIDIA GPUs. I think around 50% of that budget is allocated for GPUs. So the question now becomes in the absence of Jensen Huang, will NVIDIA still be able to function as the company that it was under his leadership? Will they be able to deliver on some of their future GPUs? Will they be able to fulfill demand? and supply? If you look at recent comments from Colette Gress, who is the interim CEO and acting CFO of NVIDIA, she said on the quarterly earnings report most recently that one, listen, we've had a record quarter, and two, most importantly, we're going to deliver on anything. The momentum is with us, and we feel confident we have the talent and research to be able to do so. So if you believe that, then you can trust that Anthropic will. have GPUs to buy that they can then feed into their AI models and build smarter model, achieving RSI and doing all the white-collar stuff. But at the same time, if NVIDIA isn't able to deliver on those things, you're going to see a slowing down of AI capex. And that's not the only reason why you would see slowing down. You have the recent elections in the midterms and the US, which was an overall win by the Dems, and they heavily support. slowing down data center build out. If you look at a recent semi-analysis report,

20:07which is funny because they're the semiconductor research guys and they've been bullish in general about AI, but recently they just went bearish and they had a whole in-depth take, which I'll get into in a second. But their whole take there was we're probably going to see around 45% of next year's data center capacity be slowed down, paused, or in the worst case, stopped. I don't think it'll be. stopped, but I think it'll be slowed down. So what this means is you're gonna have slower delivery of GPUs that come online that are able to train brand new models, large part of which is going to be called ananthropics models. So in that event, you're probably gonna see a slowing down of revenue, which means that they're gonna make less money to pay off the CapEx bill. And so they're probably gonna need to raise more money or loan money or whatever that. might be. Which then brings me to my final point. The credit market, the lending market, is getting a little tight these days. It was originally called out about a month ago by Elon Musk himself, who was making the comment not towards Anthropic, but other neo-cloud companies. You see,

21:07SpaceX pivoted recently. Um they're still doing the space stuff, and they recently caught in a very impressive rocket, which I made an episode about go check the that out. But they've also become the number one AI cloud provider. And Elon Musk's take was we were only able to pull this off because we have investment grade financing. In other words, we're able to borrow as much money as we can at a cheap enough rate that justifies building out these CapEx data centers and earning the money from that. Now, when you compare that to smaller neo cloud providers, they do. don't have the perks of being able to do that because they don't have an established reputation. In some cases, they can. They can go by, you know, saying, hey, Microsoft promised me uh 10 to 100 billion dollars worth of revenue. And so Elon's argument was smaller neo clouds wouldn't be able to pull off this level of loans. And he's not wrong. And what you see on the loan market is that people are less willing to kind of give their money. So when I bring that back to anthropic and the amount of money that they're going to need in this kind of recursive flywheel or this capex bubble, as a lot of bears like to call it, you can see where some things might become problematic. And to kind of top it off, you have some of the largest fund managers in the world. I think Stanley Truckenmiller Trockenmiller most recently said that he's gone completely risk-off, that he doesn't own.

22:27any AI stocks. And if anything, he would want to go short. And his main reasons, he said this at the Tomorrow X summit, which is actually where I'm at right now. I'm recording this and heading straight back there. That's why I'm in the hotel room and not in my studio. His main reasoning for that was he's just uncertain of where the market is going. And in fact, if you just look at NVIDIA as a prime example, right? They had blowout earnings. their biggest ever. They made $151 billion in a single quarter in three months. That's unprecedented. It's the most money made by any company ever in that allotted period of time. They crushed it on revenue margins, they crushed it on everything else, and their revenue backlog grew. So you can expect even larger growth for the next quarter, and their stock still declined. So good news is resulting in stocks declining, probably instills a bunch of fear. And so, you know, you have esteemed investors like Sandy Druckenmiller and a bunch of others that are kind of going risk off. They're not going one way or the other. And yet, despite all of this, despite all of it, Anthropic's IPO is still six to

23:3510x over demand, $500 billion, half a trillion. dollars is sitting on the sidelines. So it begs the question: what is the bull case? Well, number one, they are the fastest growing company ever. I think just in this year, they've grown 1121% compared to last year, and they're making bucket loads of money as a result of that. Now, the second thing is where that money is coming from. Anthropic has made no secret of this. They went after the enterprise market, which I think is a very smart move because enterprises are where a lot of the wealth is concentrated. And Claude Code as a winning product in enterprises has earned them the most money. Now, kind of putting that aside for a second, if you look at some of their top customers, so we're talking about customers that pay minimum hundreds of thousands of dollars a year to up to 10 million dollars a year, this catches category of customers has not only grown, but they've doubled down. See, there's this uh statistic, or rather, this data point called net revenue retention, which basically estimates how much a customer spends at the start of the year versus how much they spend at the end of the year. That figure for Anthropic is 900%. That has gone up 3x since they measured this about. five months ago. If you are spending $100,000 in January

24:551st, you are likely to spend $900,000, almost a million dollars by the end of the year because you love the product so much and you're obviously reaping so much value from it. Now, the counterpoint to that is the companies that are spending the most on anthropics products are other AI adjacent companies, or they are startups that are incredibly VC backed. So in other words, they're being subsidized to an extent. I don't know how much of that is necessarily true. There's an Apollo Global Research Report which kind of digs into it and says listen, outside of the top seven companies in the Fortune 500, the other 493 aren't actually spending much or aren't actually seeing the ROI. They're kind of subsidizing, they're testing it out, but they're not really seeing the ROI just yet. Maybe it's not with Clawed models, maybe it's with other open models. And the fact is, we just don't see where a lot of this revenue is going to come from. But I guess you know that'll be answered over the next six months. And you can actually argue that the market is so much larger for these companies because if you look at the total budget spent that each of these anthropic customers are spending on Claude, it is less than 0.1%

26:00of their entire annual expenses. Let me repeat that. Not 1%, less than 0.1% of the entire annual budget is being spent on Claude. So even if that 10x is and it goes to 1% or heck, even if it should be at around 5%, which is honestly where I think most companies will be, at least by the end of next year, especially at the rate at which these models are improving, that is going to be a humongous revenue. driver for anthropic and clawed models alone. Now, let's not forget about the actual models themselves. Putting all the data aside, I personally think Clawed is one of the best AI models out there. Doesn't sound like AI Slop. It speaks to me formally, but it doesn't speak too lengthily, certainly with their most recent Opus 5.5 model. It speaks to me like a normal human being, not like an esteemed a professional. professor that needs to give me 10 pages worth of results for a simple question. And most importantly, Anthropic has been able to figure out taste and quality of service. When I compare it to the likes of, let's say, ChatGPT, because I've been playing around with 6.2 Astra recently, and I'm gonna give you my review of that pretty soon. And I've been playing around with Gemini Carbon and their new model, which is rumored to be dropping today. The idea here. is that Anthropic has been able to balance smarts and

27:19intelligence with actual usability. And that's what I'm seeing a lot of the market kind of react to. Whenever, and by the way, this is across any profession. If you are a writer, or if you're working in design, or if you're working as a software engineer, every experience is pretty much the same. Every response is the same. It's I prefer Claude over some of these other. models. Now, it's not to say that Clawed is necessarily hard to beat or that so or that they're so far ahead, but it's rather the other models haven't been able to figure out the kinks quite well. In fact, GPT 6.2 Astro, which is their most recent model, which they launched just this week to try and kind of dampen the prospects for Anthropics IPO, kind of fell a little bit flat on its face because the response has been. it's too nerfed as a model, it's got too many safeguards up. And when I ask it to do actual work, it ends up doing a really terrible job. So the whole art or the balancing act of creating a model, giving it the right taste, giving it the right personality, you know, and doing so in a very efficient way is something that Anthropic has nailed. Now, speaking of efficiency, that's the other or final. part of this entire bull case. It doesn't matter if you have the number one most desirable model. It doesn't matter if you're spending all this compute to increase its intelligence, if you're not making money on it. And most importantly, if you aren't

28:39proving a trend that these AI models are going to get continually cheaper and more accessible whilst intelligence increases, you can't have a kind of growing thesis to justify the crazy kind. of revenue projections that you have. Fortunately for Anthropic, they have proven both. Number one, that the amount of compute that they're spending is now yielding more money per dollar spent. So uh there's this stat or bit of data that I saw in the S1 where for every dollar that they've spent on compute, they're now making around a $1.72 dollar seventy-two or maybe even two $2 dollars most recently.

29:14Unidentified speaker (UNKNOWN): And so

29:14Unidentified speaker (speaker_0): that's almost. you know, double what they're putting in. But also for every successive model that they train, it is requiring less compute. Or rather, for the compute that they spend on training that model, you end up with such a larger multiple of intelligence. And that's kind of like this whole graph that METR and the likes of AI 2027 research highlighted, which basically pointed out anthropic is on. a winning streak right now, and they might have already run away with it. And if you assume that, um, that would justify then Dario's takes earlier this year, where he said, and I think he said it on the IPO Road show as well, which is Anthropic's coming after everything. We are a company that is building AI models, but AI models are going to consume most of the stuff around us, and we want Anthropic to be a competitive player in every one of these. sectors. And he wasn't kind of minting his words there. Now I do want to point out the obvious Achilles heel to this entire ambition that Anthropic has, which is chips, AI chips, ones that are being built from Nvidia, but also from other companies like AMD, Intel, and Google. The idea here is if you want to be able to build these AI models, you're going gonna to need the chips. And if you look at Semi-Analysis's semi-analysis' report, they're expected expecting a huge oversupply in 2027 and 2028. The reason that is justifying that are a few. One,

30:34slowing AI lab revenue growth. Now, there are two bits of data to kind of support this. Number one, if you look at Q3 revenue increase for anthropic, it was 45%. The previous quarter was 100%. So you could be like, oh my God, what if this halves again and it becomes 20%? What does that mean for the future of cash? CapExpend? And then on the other side, if the data sender built-out slows down, you'll end up having a lot of GPUs just collecting dust. In fact, that's already happening right now, not because the data sender build-out was slowing down, but because there's not enough power or electricity to even turn these things on. So there are a lot of factors that sit outside of the AI lab layer. of the stack. It sits in like the kind of fundamental chip layer, the substrate layer, getting enough, you know, amount or capacity of memory to be able to build the chips in the first place. Memory prices have absolutely skyrocketed, but then it's also getting the power to these data centers, getting the human capital to actually build these data centers. All of these are constrained to an extremely high magnitude right now.

31:34And actually the company that semi-analysis estimates to be hit the worst is Nvidia. because, well, that's no surprise they have the largest stake in the game. They make up, you know, the majority of the GPU market, around 80 to 90%. And then there's neo clouds. Remember, a lot of neo clouds exist today. It's become a burgeoning market of right 200 plus major neo clouds all over the world, spinning up data centers and compute to be able to serve these labs. They have relied a lot on external financing. And some of this financing is quite risky debt. I mentioned it earlier, Elon Musk called it out. But the idea here is a lot of these neoclouds have borrowed money at pretty extreme or crazy rates that fluctuate quite a lot and might significantly eat into their margins. And so if you end up having neo clouds that aren't able to kind of fund their CapEx buildup because they're slowed down by the same thing, you know, chip shortages or anthropic and open air buying all of it. or the lack of power, electricity, or access because, you know, let's face it, hyperscalers that are as large and prominent and reputable as Meta, Microsoft, Google, and Amazon are going to get first dips and all of this, then these neo clouds will end up dying. But the problem is you have anthropic and OpenAI and other major AI labs committing tens to hundreds of billions of dollars to some of these smaller neoclouds. neo clouds. So I think NScale is the one example, or uh and they've signed up

32:54with multiple others. But the idea here is if any of those deals fall through, that will also slow down Anthropic's ascent to creating AGI. But all this being said, I still don't see it slowing down too much for the simple reason, and you can call me low IQ for this. I think that AI is going to solve a lot more than exists today. And I can even clearly make that statement, even if AI progress stalled right now. And the number one indicator of that is how many people actually use AI beyond just a general search query. In fact, the percentage of the US population that uses AI as a Google search query is around, I think, five to 10%, which is a significant chunk. But the value of using AI models as they exist today actually far exceeds that. And so I think if we have some time to properly build out, you know, front-end facing consumer apps, of which Anthropic is now making a very major move towards doing that with their recent release of Clawed Office, I like to call Clawed Microsoft Office, where they're replacing uh Excel,

33:54they're replacing PowerPoint, they're really they're replacing Docs, Google, G Suite, all of that, and creating their own version, you'll end up seeing a much more increased use. usage or consumption of clawed models in general, which of course will end up resulting in more money in Dario and Anthropic's pockets. And then there are some of the larger bets that Anthropic is making in the likes of medicine, science, research, and robotics, as well as a host of other sectors, which gives me confidence that the vision that Anthropic and other major AI labs are building out is gonna have a decent probability of. solving some other major thing. And if they are able to do so, if they're able to pull off another clawed code-esque feat, that'll result in another couple hundred billion dollars of ARR

34:36Unidentified speaker (UNKNOWN): and

34:37Unidentified speaker (speaker_0): they'll be able to pay off all their capex. So let's see what some of those opportunities look like right now.

34:45Unidentified speaker (UNKNOWN): The

34:45Unidentified speaker (speaker_0): first one is obviously knowledge work. If they are able to create an AI that can do 99% of software-based human tasks, then that is a huge amount of the economy or value that is generated within the economy. That's number one. Number two is a trend that we're now starting to see become very popular, which is cybersecurity. Over the last couple of weeks, heck, over the last month, we have seen some of the worst cyber attacks ever. Some are directly related to AI, some are indirectly related to AI. But one thing is certain and true: the frequency of hacks is increased and will continue to increase as AI models get into the hands of more people. So the question then becomes: how do you defend your business? How do you defend yourself? How do you defend your bank accounts, flight bookings, etc., et cetera, from getting attacked by someone armed with an AI model? Well, the answer is simple. You use AI to defend against that. And there are a few ways of doing this. You can use AI models to proactively plug any holes in your code base or kind of look over all your accounts and be kind of like a secure defender. Or

35:51you could get AI to kind of solve hacks as they're being dealt with in real time. Either way, you're going to rely heavily on some of these products. And you can kind of see OpenAI pivoting slowly towards this. They have their cyber models that can be used for defense called Daybreak. And rumor has it that OpenAI is working kind of closely with the government on a lot of these recent wave of attacks with their own internal models to do so. In fact, I think OpenAI has an internal model that sits currently two to three X more capable of cyber defense than any other model out there. So the point being is I think that's going to be another greenfield category that a lot of these companies include. including Anthropic, is going to focus on because it makes sense there's a lot of wealth concentration there that they could potentially earn from. The third one is one I'm most excited for, which is medicine or solving diseases in general. Now, over I think since the end of last year, OpenAI and Anthropic have spent upwards of 10 billion dollars to acquire Require a lot of scientific data. I'm talking things about genes, things about chemical reactions.

36:51laboratory data, you know, understanding how laboratories work, what kind of tests they do, what were the results of those tests. And the reason is they are basically feeding all that data into different versions of internal models of Claude. And each company, Anthropic especially, has set up their own Claude for Science or Claude Research for Science departments. And they're kind of like a ragtag team of some of the smartest researchers, scientists in the world trying to kind of automate as much as they can of science and have breakthroughs similar to the ones that we've been seeing in mathematics recently. The idea being if they are able to make significant progress, then maybe we can have AI designer drugs. It's not just Anthropic focusing on this. In fact, I'll say their major competitor right now, ironically, is Google or the former CEO of Google. DeepMind, Demis Hassibis, who stepped down to focus on his own company called Isomorphic Labs, which is basically Google's AI designer drug lab. The idea being is, I think we're reaching a point, not yet, but very soon, where you're gonna have these AI models solving Involving cancer individually for patients. And what that means is once we've created the right concoction that should be administered to this individual, how do we then create that formula? Which brings me to the third or fourth point. I don't know. I'm losing track now

38:10at this point. Robotics. We need to be able to automate a bunch of the physical world. AI is really good at doing all the software-based tasks, you know, using a screen, moving my browser around, blah, blah, blah. But as soon as I shut down my laptop, as soon as I walk outside the door, it can't help me with a bunch of physical tasks. Robotics is a very important sector that we're seeing, again, Anthropic and OpenAI, the two leading labs focus on pretty massively. In fact, if you look at GPT 6.2 Astro, heck, if you look at GPT 6.1 that came from OpenAI, it exceeds all the evals when it comes to robotics. That's a very intentional use. Remember, these labs can pick what they train their models for or how they train it specifically. The most recent example being OpenAI wanted to solve a bunch of math problems, so they made sure that the model was really good at solving maths. Now it's doing the same for robotics. Anthropic is doing the same thing. In fact, they are in the early stages of building an autonomous wet lab, a biological lab which can function on its own and kind of take the AI. designer drug formulas that you know Claude is cooking up and actually test them 24 7. Now, you have to understand one of the biggest constraints that are holding back, you know, pharmaceuticals and science labs today,

39:29putting regulation aside is being able to test and get the data for all these novel experimental drugs that they can then give regulators and say, hey, listen, we've proven

39:40Unidentified speaker (UNKNOWN): that

39:40Unidentified speaker (speaker_0): we can pass your bar of quality. Let's create this drug. If anthropic and open air and whoever are able to automate this sector, then it'll end up unlocking billions of dollars worth of value for them. And then finally, of course, is RSI, where you have an AGI, an AI model that is smarter than the top expert. in any human field,

40:05Unidentified speaker (UNKNOWN): uh,

40:05Unidentified speaker (speaker_0): which means that it can basically do anything and everything. And at that point, I don't know why I would be sitting in front of this camera. It'll probably be an AI version of myself uh spouting some kind of script that Claude generates. I hope it never reaches that point. I hope that I can continually make mistakes and stutter on this show. But yeah, that is the anthropic IPO. 825 million shares going live. It's around six to maybe eight percent of the entire float. An unlock schedule, which looks very similar to other unlock schedules. I think it's like a year, cliff, and then you know, every subsequent kind of couple of months you'll see like major unlocks. Uh concentration is kind of like well spread across all the employees. But remember, the company is only 5,000 employees. So this IPO is going to mint so many centimillionaires, centi millionaires, $100 hundred million. dollar plus. I believe that's the right term. If I'm not, feel free to correct me. But yeah, it's going to be one of the biggest IPOs ever. I am personally super excited about this company. I um, you know, have loved Claude models, and this is coming from someone that just enjoys the product. That's kind of like where I do a lot of my gut or intuitive investing. And my gut tells me right now that AI is going to solve a lot more complex problems in the future that are going to be insanely valuable. to the economy, to other stocks and companies in general, but also to humanity, most importantly. So yeah, I hope you enjoyed this episode. I hope you've got enough information to

41:24kind of make a somewhat informed decision on what you're going to do with the Anthropic IPO. And yeah, I guess we'll see how it all plays out. Because remember, macro is still a very large factor in all of this, and there are so many moving parts. So as always, we never know which way this is going to go. But I'll be here. keeping track, being terminally online, trying to figure out what the hell is going on. But thank you so much for listening to this episode. If you listen to this on YouTube, Spotify, wherever you are, please give us a rating, follow us, subscribe to us, turn on notifications, it helps us out massively, and also share this with a friend. And yeah, I'll see you on the next one. Bye.

This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights