Anthropic IPO Analysis: Governance Risks, Financial Scale, and Short-Term Trading Strategy
Summary
Speakers analyze Anthropic's upcoming IPO, highlighting its massive Q3 2026 revenue of $16.9 billion and projected full-year revenue of $60-70 billion. They detail the company's unique governance structure, where founders retain control via Class T shares, raising concerns similar to Meta or Tesla. The discussion notes that the reported $80.7 billion net loss is primarily a non-cash accounting artifact from revaluing convertible notes. Finally, the speakers advise treating the IPO as a short-term trade rather than a long-term investment due to expected post-IPO price action and high valuation.
Episode sections
- 0:00 – 1:00
Intro and Sponsor Read
Speakers introduce Anthropic as a potential top IPO of the decade and promote the Semi Insider subscription service at chipstockinvestor.com, noting a price increase on December 1st.
- 1:00 – 3:10
AI Supply Chain and Five-Layer Cake
The speaker describes the semiconductor flow chain and Jensen Huang's 'five-layer cake' model, noting that AI labs like Anthropic, OpenAI, and Google DeepMind are a new point of entry where data usage feeds back into the infrastructure layer.
- 3:10 – 5:40
Share Structure and Voting Control Risks
The transcript details Anthropic's Public Benefit Corporation (PBC) status and Long-Term Benefit Trust (LTBT). Class F shares hold 50.1% of voting power, while Class T shares allow founders Dario and Daniela Amodei to control the majority of the board, raising governance concerns similar to Meta or Tesla.
- 5:40 – 7:15
Q3 Revenue and Financial Scale
Anthropic reported $16.9 billion in Q3 2026 revenue, up 46% year-over-year, surpassing its entire H1 2026 revenue of $16.3 billion. The company has nearly 12,000 enterprise customers spending over $100,000 annually, with full-year 2026 revenue projected between $60-70 billion.
- 7:15 – 9:25
Net Loss and Non-Cash Expenses
The $80.7 billion net loss is primarily driven by non-cash accounting changes in the fair value of convertible notes and derivative liabilities due to rapid growth and impending IPO valuation, rather than operational cash burn.
- 9:25 – 11:55
Revenue Model: Tokens vs Subscriptions
Anthropic generates revenue through subscription fees and consumption-based 'pay for intelligence' (tokens). Consumption is now the vast majority of revenue, with subscription revenue at $3.2 billion, representing roughly a quarter of the total.
- 11:55 – 19:35
Compute Efficiency and Cloud Provider Tax
Anthropic earns $2.08 in revenue per dollar of compute spend. However, this efficiency metric may be inflated because hyperscalers like Alphabet, AWS, and Microsoft build the infrastructure, allowing Anthropic to smooth expenses. The speakers note that cloud providers also take a 1.5-3% cut if Anthropic distributes via their marketplaces.
- 19:35 – 25:01
Balance Sheet, Commitments, and IPO Strategy
Anthropic holds over $131 billion in cash/assets against $109 billion in liabilities, with $266 billion in purchase commitments extending into the 2040s. The speakers view the IPO as a likely short-term trade due to a projected $1.5-2 trillion valuation and potential for massive post-IPO price action, advising against long-term holding at debut.
Assets discussed
| Asset | Kind | Stance | Context |
|---|---|---|---|
| AlphabetGOOGL | equity | neutral | Question on this, because I think maybe some would have questions. So if we say cursor. which is an AI company, Alphabet has Google DeepMind, Amazon isn't necessarily an AI lab. Microsoft is not an AI lab, but it does have AI components like Copilot. Why are all these companies very large |
| AmazonAMZN | equity | neutral | Question on this, because I think maybe some would have questions. So if we say cursor. which is an AI company, Alphabet has Google DeepMind, Amazon isn't necessarily an AI lab. Microsoft is not an AI lab, but it does have AI components like Copilot. Why are all these companies very large |
| MicrosoftMSFT | equity | neutral | Question on this, because I think maybe some would have questions. So if we say cursor. which is an AI company, Alphabet has Google DeepMind, Amazon isn't necessarily an AI lab. Microsoft is not an AI lab, but it does have AI components like Copilot. Why are all these companies very large |
| NVIDIANVDA | equity | neutral | directly themselves. They're going to be buying from NVIDIA and AMD and from Intel and from Broadcom and everybody. But they're definitely going to have the assets at their disposal to be able to do it. So there is that. And they're going to need it though. |
| AppleAAPL | equity | neutral | 70 billion in revenue for full year 2026. I don't think the market's going to let it go for that. So by the time it actually starts trading on the market, I think there's a probability, not a huge probability, but this thing immediately is up there with Apple and NVIDIA. I think the market will push it to some insane |
“Stance” is how the speakers talked about the asset, not a recommendation.
Claims worth verifying
Anthropic is one of the hottest IPOs of the decade.
Unidentified speaker (speaker_1) ·0:05
How to verify: Market sentiment analysis and IPO allocation demand data upon Anthropic's S-1 filing and IPO launch. For recent events, verify 2026-08-05 through 2026-11-05. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic is a Public Benefit Corporation (PBC) with a Long-Term Benefit Trust (LTBT).
Unidentified speaker (speaker_0) ·3:24
How to verify: Review Anthropic's S-1 filing on SEC EDGAR under the 'Corporate Structure' or 'Voting Power' sections. For recent events, verify 2026-08-05 through 2026-11-05. Use another period only when the speaker explicitly names it.
Primary sourceClass F shares control 50.1% of outstanding voting power.
Unidentified speaker (speaker_0) ·4:53
How to verify: Review Anthropic's S-1 filing on SEC EDGAR for the 'Capitalization' and 'Voting Power' tables. For recent events, verify 2026-08-05 through 2026-11-05. Use another period only when the speaker explicitly names it.
Primary sourceClass T shares allow Dario Amodei and Daniela Amodei to control the majority of the board of directors.
Unidentified speaker (speaker_1) ·4:43
How to verify: Review Anthropic's S-1 filing on SEC EDGAR under 'Board of Directors' and 'Voting Power' sections. For recent events, verify 2026-08-05 through 2026-11-05. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic had $16.3 billion in revenue in the first half of 2026.
Unidentified speaker (speaker_0) ·6:47
How to verify: Review Anthropic's S-1 filing on SEC EDGAR for H1 2026 financial results. For recent events, verify 2026-08-05 through 2026-11-05. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic had nearly 12,000 enterprise customers spending over $100,000 a year.
Unidentified speaker (speaker_0) ·6:27
How to verify: Review Anthropic's S-1 filing on SEC EDGAR under 'Business' or 'Management's Discussion and Analysis' sections. For recent events, verify 2026-08-05 through 2026-11-05. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic's net loss was $80.7 billion, primarily due to non-cash changes in fair value of convertible notes.
Unidentified speaker (speaker_1) ·7:29
How to verify: Review Anthropic's S-1 filing on SEC EDGAR for the 'Consolidated Statements of Operations' and notes on convertible debt. For recent events, verify 2026-08-05 through 2026-11-05. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic earns $2.08 in revenue for every dollar spent on compute.
Unidentified speaker (speaker_0) ·12:03
How to verify: Review Anthropic's S-1 filing on SEC EDGAR for 'Revenue' and 'Compute Costs' or 'Cost of Revenue' breakdowns. For recent events, verify 2026-08-05 through 2026-11-05. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic has $266 billion in purchase commitments extending into the 2040s.
Unidentified speaker (speaker_0) ·20:53
How to verify: Review Anthropic's S-1 filing on SEC EDGAR under 'Liquidity and Capital Resources' or 'Commitments and Contingencies' sections. For recent events, verify 2026-08-05 through 2026-11-05. Use another period only when the speaker explicitly names it.
Primary sourceDo your own research
- Review Anthropic's S-1 filing for detailed governance structure, specifically Class F and Class T share voting rights and board composition.
- Verify Q3 2026 revenue figures and customer concentration risks against the company's official financial statements.
- Check the composition of the $80.7 billion net loss to distinguish between cash burn and non-cash accounting adjustments for convertible notes.
- Analyze the 'revenue per dollar of compute' metric by comparing Anthropic's reported efficiency against actual infrastructure costs borne by hyperscalers.
- Assess the 'one-year IPO rule' by reviewing historical post-IPO performance of similar high-growth tech companies to evaluate short-term trading risks.
- Identify major customers accounting for significant revenue percentages, such as the noted 11% from 'Customer C' (likely Cursor/SpaceX).
- Examine the $266 billion in purchase commitments extending into the 2040s to understand long-term capital expenditure obligations.
- Review the impact of stock-based compensation (SBC) on operating margins, noting the $2.19 billion in SBC excluded from adjusted operating margins.
Full transcriptexpand
- 0:00
- Unidentified speaker (UNKNOWN): One
- 0:00
- Unidentified speaker (speaker_0): of the hottest IPOs this year is definitely going to be anthropic.
- 0:05
- Unidentified speaker (speaker_1): Possibly one of the hottest IPOs of the decade.
- 0:08
- Unidentified speaker (speaker_0): Maybe of all time.
- 0:10
- Unidentified speaker (speaker_1): Whoa, easy.
- 0:12
- Unidentified speaker (UNKNOWN): We
- 0:12
- Unidentified speaker (speaker_0): talked about the upcoming IPO with our semi-insiders last week. And what you're about to see is an excerpt from that live event. Now, if you want to join us on our live events, ask us pretty much anything. make sure you go over and check out semi-insider at chipstockinvestor.com.
- 0:33
- Unidentified speaker (speaker_0): That'll get you access to not just that, but everything we have to offer at CSI.
- 0:39
- Unidentified speaker (speaker_1): Yeah, all of our research articles, a look at the companies that are in our portfolio, investing tools and investing data, and more. There's a lot of stuff on there, and we have it all organized. in such a way that it's how we use it every day in managing our investments, managing our portfolio. So that's what you get at Semi
- 0:59
- Unidentified speaker (speaker_1): Insider.
- 1:00
- Unidentified speaker (speaker_0): Make sure you go and check it out, tipstockinvestor.com. Price is going up December 1st. So if you want to trial it, there's a free demo available as well. Try before you buy. Make sure you hit that subscribe button. Really helps us out here. And let's get into it. Anthropic. Let's start with this. We'll start from our most basic level:
- 1:20
- Unidentified speaker (speaker_0): the semiconductor flow chain, the industry flow. We have said from the very beginning that all of this work that's being done in the very first step gets fed back into itself at the very end, which is exactly what's happening with anthropic and these other mega. AI labs. All of
- 1:40
- Unidentified speaker (speaker_0): that data, all of that usage is now getting fed back into the top of the funnel. And that's why it's such an important part of this part of the supply chain, why it matters to the whole ecosystem. And Anthropic's a big one, obviously.
- 1:55
- Unidentified speaker (speaker_1): Yeah. Anthropic, OpenAI, those two, Google DeepMind as well. This is the new point of entry.
- 2:03
- Unidentified speaker (speaker_1): from the software industry, where it re-enters all the other layers of not just the semiconductor industry, but the entire economy. The five-layer cake. I remember we talked about this throughout the year after it got introduced by Jensen. Usually this is an inverted pyramid. The base of the pyramid, those
- 2:23
- Unidentified speaker (speaker_1): industries are smaller than the top end markets.
- 2:27
- Unidentified speaker (speaker_1): products and services that are driven by consumption, either business consumption or consumers. And it's inverted right now. Energy needs to be built and secured. The infrastructure needs to be built out and filled up with chips and servers and racks. But we're starting to see early stages of you know it actually looking like a more
- 2:47
- Unidentified speaker (speaker_1): normal consumption-driven industry. if we can call it that. The digital economy, where, okay, these AI models are now going to start getting really, really big. And the word that Anthropic used throughout its S1 was diffusion. Diffusion of its models. They want the AI
- 3:07
- Unidentified speaker (speaker_1): to proliferate throughout the entire economy into applications and devices and market services. That's starting to happen now. Okay, we're going to get to that, but we need to talk about the biggest risk, I think, right up front. I think this needs to be acknowledged right now.
- 3:24
- Unidentified speaker (speaker_0): Yeah. Let's talk about the share structure and voting power and what type of company that Anthropic is. It's filed under that PBC or Public Benefit Corporation, which I find this kind of funny and a little ironic that it's a corporation that matters to the good of
- 3:44
- Unidentified speaker (speaker_0): humanity. That's what that means. You have to weigh the good of humanity with not just the shareholders and the founders making money. And then there's also this LTBT long-term benefit trust that needs to be addressed. So those two factors, let's look at the share structure a little
- 4:04
- Unidentified speaker (speaker_0): bit.
- 4:04
- Unidentified speaker (speaker_1): Oh my God. Opening up a prospectus, seeing this on the first few pages, for me, it just almost immediately kills any enthusiasm.
- 4:14
- Unidentified speaker (speaker_0): But we can boil this down.
- 4:16
- Unidentified speaker (speaker_1): What you would be investing in when the stock becomes publicly traded is class A. Class N
- 4:23
- Unidentified speaker (speaker_1): is for companies agree to buy anthropic. stock as part of a deal, like some of the chipmakers, for example, or chip designers. That's class N. Class T is for that LTBT long-term benefit trust. It's basically like a super vote that allows Dario Amodei and I think
- 4:43
- Unidentified speaker (speaker_1): his sister Daniela, maybe as well, to be able to maintain control over the board of directors.
- 4:49
- Unidentified speaker (speaker_0): And they also have class. F shares.
- 4:51
- Unidentified speaker (speaker_1): Which is a super vote as well.
- 4:53
- Unidentified speaker (speaker_0): And so class F controls 50.1% of the outstanding voting power. The rest makes up the 49%. However, that class T, the way I understand this, ultimately they have more say in the board of directors.
- 5:09
- Unidentified speaker (speaker_1): Yeah, they would get to basically vote the majority of the board of directors. What
- 5:13
- Unidentified speaker (speaker_0): about class E? Those are
- 5:14
- Unidentified speaker (speaker_1): for employees. The point being is you have to believe that Dario and Daniela Omoday will be able to effectively manage the balance between being a public benefit corporation, making sure the business is benefiting you, the shareholder, because they're going to be
- 5:34
- Unidentified speaker (speaker_1): ultimately the only ones calling the shots at the end of the day. And that might be a deal breaker for you. It's like having to get over Mark Zuckerberg being in control of Meta or Elon Musk of Tesla or SpaceX. It's that type of situation. Okay. Enough said about that.
- 5:53
- Unidentified speaker (UNKNOWN): Let's
- 5:53
- Unidentified speaker (speaker_0): talk about financials.
- 5:55
- Unidentified speaker (speaker_1): Okay.
- 5:55
- Unidentified speaker (speaker_0): Back to the exciting stuff.
- 5:57
- Unidentified speaker (speaker_1): Q three revenue.
- 5:58
- Unidentified speaker (speaker_0): Okay. So they released the Q Q3 three revenue revenue after after they they released the first S1. 16.9 billion in Q3 2026. That's an important number to remember because we're going to be talking about the first half of each year in just a moment. So 86X on the Q three revenue in 2026
- 6:18
- Unidentified speaker (speaker_0): over Q4 2024, which was disclosed. 16.9 billion, 46%
- 6:25
- Unidentified speaker (speaker_0): up on the year.
- 6:27
- Unidentified speaker (speaker_0): And that number there on the very right, 11.9,000 enterprise customers that spend over 100,000 a year. Also important note. So here's the revenue breakdown for the first half of 2026 is all on the right side. 16.3 billion in revenue. Now we
- 6:47
- Unidentified speaker (speaker_0): just said the entire quarter of Q3 2026 is. 16.9 billion. So first half of the year, 16.3 billion, and in just the most recent quarter, 16.9. So we can assume it'll be much higher in Q4 2026.
- 7:04
- Unidentified speaker (speaker_1): Let's say full year 2026, somewhere between 60 and 70 billion in revenue. That would be our baseline expectation for Q4, which we'll probably have an average actual company report on at the beginning of 2027, assuming the IPO actually happens.
- 7:21
- Unidentified speaker (UNKNOWN): And
- 7:21
- Unidentified speaker (speaker_0): then the operating loss, $693 million in the first half of 2026. We'll talk about that here in just a moment.
- 7:29
- Unidentified speaker (speaker_1): The unit economics here are actually pretty good. But yeah, okay. Net loss, 80.7 billion. The vast majority of that, though, is non-consuming. non-cash expense. If you look at the first line item in red, change in fair value of convertible notes, derivative liabilities, and safes, which is a
- 7:49
- Unidentified speaker (speaker_1): venture capital early stage venture vehicle where basically the investor gives cash to a startup and the investor and the startup don't agree on a valuation yet. They agree on a valuation later on. And so especially the convertible notes like Amazon's convertible notes that it owns in Anthropic, all of
- 8:09
- Unidentified speaker (speaker_1): those just got valued much, much higher than they were before because the company is growing so fast and because the IPO is getting ready to happen. Expectation is this is a one and a half to two trillion Million dollar IPO valuation. When it debuts on the public market, it's probably going to be much higher than that. So pretty crazy valuation.
- 8:29
- Unidentified speaker (speaker_1): But that's that where the non cash loss is coming in. It's not actually a loss because it just goes on the liabilities of the balance sheet.
- 8:37
- Unidentified speaker (speaker_0): That's the same case as with this slide here. We're showing nearly $40 net loss per share in the first half of the year. But again, that's not cash.
- 8:46
- Unidentified speaker (speaker_1): This is one of the reasons why investing in a recent IPO within the first year is so messy.
- 8:51
- Unidentified speaker (speaker_1): because of all of these different things that get resolved by the IPO itself. But during that first year, it's just very difficult to read what's actually happening with the business. I would say at this point, though, the most important metric is that operating loss shrinking and almost reaching break-even the first half of the year. And in fact, I think we have the
- 9:11
- Unidentified speaker (speaker_1): quarterly breakdown here later on that shows they actually had an operating profit in just Q2 2026. Yeah,
- 9:18
- Unidentified speaker (speaker_0): we'll get to that in just a moment. And here's a look at those enterprise customers. You can see the huge ramp up in customers that spend over 100,000, now nearly 12,000 companies. And this is all part of the diffusion plan. More enterprise companies using Anthropic and getting the
- 9:38
- Unidentified speaker (speaker_0): word out. Okay, let's talk about how they make their money. This is one of the most important things to understand about a business. How do they make money? Probably all of you have used Claude at this point. They have model classes, which you can see there: Mythos, Opus, Sonnet, and Haiku. You can use them based on
- 9:58
- Unidentified speaker (speaker_0): their capabilities, and the cost is commensurate with those capabilities. And so a customer can use any of those models. They get paid based on pay for intelligence or pay for tokens or subscription pricing. And we'll show you the breakdown here in just a moment of what that looks like in dollars and cents terms.
- 10:18
- Unidentified speaker (speaker_0): But customers that have subscription pricing, if you have to buy extra credits or extra tokens, that goes towards that pay for intelligence. And so you're going to see that number. It's actually really high. And it's because of those nearly 12,000 enterprise customers, especially.
- 10:35
- Unidentified speaker (speaker_1): Yeah. And we can see that here in the next slide in the revenue breakdown. I think the US versus international revenue is an interesting metric. I like seeing that the international revenue is catching up to the US revenue a bit. the last year or so. I was surprised to see it more diverse.
- 10:55
- Unidentified speaker (speaker_1): I was expecting a really, really big, mostly concentrated in the US. We'll see how that plays out over time. Especially the bottom part, I think that's the more important, like Casey was just saying here. Consumption based, that's the pay for intelligence. The amount of tokens you get pre-packaged with your monthly subscription isn't enough. Your You're coding and
- 11:15
- Unidentified speaker (speaker_1): you run out of tokens and you buy more. That's the consumption-based model. See, that's actually the vast majority of the revenue now. The subscription-based revenue at 3.2 billion, a much smaller percentage of the total. Let's say roughly about a quarter. If the consumption pricing and the consumption model continues to
- 11:35
- Unidentified speaker (speaker_1): scale up, I think that will also favor the US. just because Silicon Valley is consuming so many tokens right now doing stuff. But we'll see how that pans out over time and how that relationship between these two revenue performance indicators kind of plays out.
- 11:51
- Unidentified speaker (UNKNOWN): And
- 11:51
- Unidentified speaker (speaker_0): then here's another look at what NVIDIA has been saying all along buyer product, you'll make more money ultimately.
- 11:59
- Unidentified speaker (speaker_1): Yeah. Intelligence compute equals revenue.
- 12:03
- Unidentified speaker (speaker_0): Yeah. And so this is revenue per dollar of compute. So every dollar that Anthropic is spending towards compute is now worth $2.08 worth of revenue.
- 12:15
- Unidentified speaker (speaker_1): Quite the scale up. This is actually showing signs of a very efficient business. Now, the one caveat to this is Anthropic could actually be spending a lot more money on the computer. compute infrastructure. The reason it's not showing up on their expense list is because they're getting other companies to
- 12:35
- Unidentified speaker (speaker_1): build it for them. They're getting Alphabet to build it. They're getting Amazon, AWS to build it, they're getting Microsoft now as well. All of hyperscalers Scalers and NeoClouds, SpaceX in on this, are building this infrastructure and encouraging. incurring cost. Anthropic is smoothing all of that expense out
- 12:55
- Unidentified speaker (speaker_1): and spending it as it comes online and as they use it. So there's a couple of different ways you could look at this. The way you could look at it from a positive angle is wow, what an incredibly efficient business scaling up. The other way of looking at this a little more pessimistically is well, yeah, but there's a lot more expense being realized.
- 13:15
- Unidentified speaker (speaker_1): It's just not anthropics expense right now at the moment, but it will be later on. Out the gate, this is a pretty good looking number. We'll see if this is IPO window dressing or if it stays a very, very lean, efficient business model long term.
- 13:30
- Unidentified speaker (speaker_0): Okay, let's look at their operating margin. Now this is on an adjusted basis. So this backs out like stock-based stock based comp, but on an adjusted basis, they've made a ton of progress here and actually positive 19% operating margin.
- 13:45
- Unidentified speaker (speaker_1): Yeah, Q2.
- 13:46
- Unidentified speaker (speaker_0): Let's go to these charts.
- 13:48
- Unidentified speaker (speaker_1): SBC stock based comp 2.19 billion in the first half of 2026. That's virtually the entire difference between the small gap loss of 4%, negative 4% operating margin. and the adjusted positive operating margin of 10% for the first half of the
- 14:08
- Unidentified speaker (speaker_1): year, or 19% in Q2 2026. That's the adjusted net loss reconciliation again, most of that being the convertible debt, revaluation, and related equity instruments. We already said the peers. Anthropic named specifically OpenAI and Google
- 14:27
- Unidentified speaker (speaker_1): DeepMind as its primary competitors and peers, customers, which are also infrastructure providers, suppliers. Strange relationships here. Who are these though? Customer A, B, C, and D. The good news is through the first half of 2026, only one customer accounted for more than 10%
- 14:47
- Unidentified speaker (speaker_1): of revenue. Customer C at 11%. eleven percent. We can use that one as a starting point. That's probably cursor, which is now part of SpaceX.
- 14:57
- Unidentified speaker (UNKNOWN): That
- 14:57
- Unidentified speaker (speaker_0): seems reasonable.
- 14:58
- Unidentified speaker (UNKNOWN): It's
- 14:58
- Unidentified speaker (speaker_1): probably who that was. A, B, and D.
- 15:01
- Unidentified speaker (speaker_0): Yeah. Alpha Bat Amazon.
- 15:03
- Unidentified speaker (speaker_1): Meta would probably be the other in this particular customer breakdown. But yeah. Okay.
- 15:09
- Unidentified speaker (speaker_0): Question on this, because I think maybe some would have questions. So if we say cursor. which is an AI company, Alphabet has Google DeepMind, Amazon isn't necessarily an AI lab. Microsoft is not an AI lab, but it does have AI components like Copilot. Why are all these companies very large
- 15:29
- Unidentified speaker (speaker_0): customers of Anthropic?
- 15:32
- Unidentified speaker (speaker_1): Why are they customers? Okay, so this is again AI diffusion happening. Let's just use Cursor as an example. SpaceX, they were using cloud models to run their AI-powered coding assistant and now AI agents for coding. Many of these companies are doing something like that, where they're using these
- 15:52
- Unidentified speaker (speaker_1): AI models to power their own products. That's why they would be considered customers. Like Microsoft uses OpenAI OpenAI for a lot of its AI products. They're not developing their own model. So this is why those AI labs, if you think again about the five-layer cake, these AI models still at this
- 16:12
- Unidentified speaker (speaker_1): point are not overtly trying to compete with existing incumbents in the economy. That's definitely coming, or they're helping other technology companies displace incumbents encumbants in different parts of the economy.
- 16:27
- Unidentified speaker (speaker_1): But really where they sit on that five-layer cake is between infrastructure and the actual end market, software and services and devices. So it's kind of like this new intermediary layer where the AI labs are almost like a infrastructure software provider. It's
- 16:47
- Unidentified speaker (speaker_1): not an actual end market, but it's awfully close. It's needed.
- 16:51
- Unidentified speaker (speaker_1): by its customers to kind of finish off the product that they want to sell. This is where things get messy for Anthropic again, because some of its own suppliers, compute infrastructure suppliers like Alphabet, are also competitors. Or those suppliers are also making money off of
- 17:10
- Unidentified speaker (speaker_1): Anthropic as well. Now, Yeah, if you think back early in the year. originally with cybersecurity, we talked about the reliance a software company, cloud-based software, has on the cloud service providers, the CSPs. They rent the infrastructure from them on the back end. And then on the front end, when they make a sale through the
- 17:30
- Unidentified speaker (speaker_1): cloud marketplace, they're usually handing over one and a half to three percent of their sales to the CSP. So if Cloud distributes its product. products through a cloud marketplace, let's say upwards of 3% of that sale via Google Cloud or Amazon AWS or Azure or whatever goes back to the
- 17:50
- Unidentified speaker (speaker_1): cloud service provider as well. So the CSPs end up making money two ways on the sale and also on actually running the software itself. This is a problem for Anthropic and Open OpenAI. AI. And the way that they are now addressing this is they have also begun building their own data
- 18:10
- Unidentified speaker (speaker_1): centers, their own infrastructure. So again, if you think back to the slide, we were talking about the efficiency of the revenue generating over $2 in revenue now for every dollar they spend on compute. We may see that figure stop scaling for a while if they ramp. up the construction of their own data centers because they're not
- 18:30
- Unidentified speaker (speaker_1): offloading the cost of the compute to somebody they're going to assume the full cost themselves up front. That's going to be an additional risk factor for Anthropic because they're going to have to raise more money somehow. Even after the IPO, it's very likely they will have to raise more money either via debt or issuance of even more stock.
- 18:49
- Unidentified speaker (speaker_1): It's a TBD. So we'll stay tuned for that.
- 18:53
- Unidentified speaker (UNKNOWN): All
- 18:53
- Unidentified speaker (speaker_0): right. What about leasing chips in facilities instead of what they've been doing?
- 18:58
- Unidentified speaker (UNKNOWN): Yeah,
- 18:58
- Unidentified speaker (speaker_1): that'd be like the Google TPUs. Also Amazon, Amazon is actually the Graviton CPUs and the other ones, tradium chips before that, also heavily used by Anthropic. So yeah, they're going to be leasing some of those chips that actually belong to Google and
- 19:18
- Unidentified speaker (speaker_1): Amazon, but putting them in their own data center facilities.
- 19:23
- Unidentified speaker (speaker_0): So as you said, another risk.
- 19:25
- Unidentified speaker (speaker_1): Yep.
- 19:25
- Unidentified speaker (speaker_0): They're going to have to spend a lot of money, a lot of CapEx is going to need to go into this. And how are they going to fund it?
- 19:30
- Unidentified speaker (speaker_1): Yeah. And Broadcom is going to be a big beneficiary of that. Broadcom was able to get that big deal done with the help of some big banks to get some funding. to Anthropic to buy Broadcom chips, Broadcom infrastructure.
- 19:44
- Unidentified speaker (speaker_0): Want to walk us through balance sheet and valuation? Yeah,
- 19:48
- Unidentified speaker (speaker_1): cash and equivalence and short-term investments, other assets, over $131 billion. Okay. That's incredible.
- 19:56
- Unidentified speaker (speaker_0): Feels like really good balance sheet there.
- 19:58
- Unidentified speaker (speaker_1): Yeah. Liabilities, primarily convertible notes. 108. almost 109 billion. All right. This is going to get skewed a lot. Who knows how much Anthropic will ultimately end up raising from this IPO, but let's say 100 billion, plus or minus, I don't know, 50 billion to the downside and another extra 100 billion to the upside.
- 20:18
- Unidentified speaker (speaker_1): It's going to be a lot of money. Some of these convertible Vertible notes will convert to stock. And so you're going to have this really cash-rich business. They'll be able to borrow. very heavily against this balance sheet as well. So that's another positive, I think, here to be considering and thinking about is okay, they're going to get into the data center game
- 20:38
- Unidentified speaker (speaker_1): directly themselves. They're going to be buying from NVIDIA and AMD and from Intel and from Broadcom and everybody. But they're definitely going to have the assets at their disposal to be able to do it. So there is that. And they're going to need it though.
- 20:52
- Unidentified speaker (UNKNOWN): Oh,
- 20:53
- Unidentified speaker (speaker_0): yeah. Here's their purchase commitments. $266 billion in purchase commitments. And this goes through anywhere from one to 20 years out. They've made a lot of commitments.
- 21:04
- Unidentified speaker (UNKNOWN): Into the
- 21:05
- Unidentified speaker (speaker_1): 2040s, which is insane. But company has big aspirations. And obviously, the IT sector, Silicon Valley, has kind of dragged the rest of the economy this direction. It's happening. Whether you like it or not, this is the future. These AI models, one way or another, are
- 21:24
- Unidentified speaker (speaker_1): going to be embedded into our daily lives at work and at home. And everybody is lining up, anthropic included, and voting with their capital, with their dollars to make sure that that happens. You spread it out. over 20 years. It's not that big of a commitment, but these are large commitments for a long time.
- 21:44
- Unidentified speaker (speaker_1): So the company needs to get all of this right. They need to get their infrastructure built themselves to build in that diversification. They need lots of AI diffusion of their models throughout every other sector of the economy, but it is happening. The risks are high, but the remote rewards are potentially
- 22:04
- Unidentified speaker (speaker_1): very high as well.
- 22:05
- Unidentified speaker (speaker_0): They made the small acquisition of stainless software, 306 million. And then they've made some other smaller acquisitions in the first half of 2026, which totaled 260 million.
- 22:17
- Unidentified speaker (speaker_1): A lot of those going towards it looks like just more infrastructure, software infrastructure to make sure everything is running right and probably in support of their future roadmap.
- 22:28
- Unidentified speaker (UNKNOWN): What
- 22:28
- Unidentified speaker (speaker_1): are you doing? Are you buying at the IPO? Are you going to try to participate in the IPO?
- 22:33
- Unidentified speaker (speaker_0): Maybe.
- 22:34
- Unidentified speaker (speaker_1): Maybe.
- 22:34
- Unidentified speaker (speaker_0): But we usually have a one-year IPO rule. It's going to be a crazy valuation. I think I would be more willing to wait for the desk to settle a bit.
- 22:43
- Unidentified speaker (speaker_1): Yeah, I concur. It's going to be oversubscribed at the IPO. The market is not going to let this thing go for one and a half to two trillion value. valuation because the price to sales ratio, even at the high end of that, $2 trillion, would be less than 30x trailing 12 months, let's say current year sales. 2 trillion divided by let's say
- 23:03
- Unidentified speaker (speaker_1): 70 billion in revenue for full year 2026. I don't think the market's going to let it go for that. So by the time it actually starts trading on the market, I think there's a probability, not a huge probability, but this thing immediately is up there with Apple and NVIDIA. I think the market will push it to some insane
- 23:22
- Unidentified speaker (speaker_1): level of valuation. So if we were able to get shares allocated before that massive pump, I think it's something that we'd be interested in having. But with the understanding that if people lose their minds over this because it's the end of the world as we know it, the only two things you need to own are anthropic stock and
- 23:41
- Unidentified speaker (speaker_0): land
- 23:43
- Unidentified speaker (speaker_1): that you can grow food on, which is also maybe a viable long term outlook for all this stuff. I don't know.
- 23:49
- Unidentified speaker (speaker_0): We may give you our thoughts on that and you may think we're crazy, but but
- 23:53
- Unidentified speaker (speaker_1): maybe not the crazy version, maybe the lucid version of that, which is real. The understanding would be if we got the shares allocated to us, we'd probably take the profit and sell. And then wait for a more reasonable valuation to come back to Earth. Kind of like what happened with SpaceX. Remember Figma summer
- 24:13
- Unidentified speaker (speaker_1): 2025, which seems like forever ago, we ended up just passing on that because everybody kind of lost it on that. And that was just totally crazy. It was good to wait. This seems to be the theme. here with these big blockbuster IPOs, and it seems anthropic would probably fall into that
- 24:33
- Unidentified speaker (speaker_1): category as well. We think it's going to be ultra hot right out the gate. If you can get in early, just have the expectation that it's a short-term trade and exit.
More from Chip Stock Investor Podcast
Nvidia After Jensen: What Investors Can Do Next
Episode discussion: Speakers advise holding NVDA amid leadership uncertainty and warn against media speculation
Speakers’ claims: The hosts describe a volatile week for Nvidia investors, criticizing media outlets for presenting speculation as fact regarding Jensen Huang's potential exit. They emphasize that while charismatic leadership may be lost, operational continuity remains with interim CEO Colette Kress and long-tenured lieutenants. The speakers are holding their NVDA position without adding to it, adhering to a 'panic early' strategy rather than selling after a drop. They advise against buying the dip due to insufficient data and are instead diversifying capital away from semiconductors. Finally, they warn that many chip companies lack the staying power to survive the normalization of AI capex, urging investors to only buy dips on high-quality businesses.
Rocket Lab Is BOOMING -- So Why Aren’t We Buying RKLB?
Episode discussion: Rocket Lab Q3 2026 Earnings: Record Revenue, Neutron Delay, and Iridium Acquisition Analysis
Speakers’ claims: Speakers analyze Rocket Lab's Q3 2026 earnings, highlighting record revenue of $268M and a $3.6B backlog driven by Space Systems. The episode details the company's vertical integration thesis, including the pending Iridium acquisition which adds $900M in revenue but also $1.93B in debt. Key operational updates include the Neutron launch vehicle delay to 2027 and a 25% dilution from stock issuance to fund the deal. Speakers project full-year 2026 adjusted gross margins of ~41% and discuss the path to profitability over a 10-year horizon.