The Morning Filter

Morningstar analyst Malik Ahmed Khan discusses Anthropic's IPO valuation, moat, and financial outlook.

56 min Listen to the original Episode page
Artificial IntelligenceLarge Language ModelsEnterprise SoftwareIPO ValuationCloud ComputingEconomic MoatRevenue ForecastingTech Stocks

Summary

Malik Ahmed Khan joins Dave Sekera to analyze Anthropic's upcoming IPO, defining its position in the frontier AI landscape alongside OpenAI, Google, and Meta. Khan highlights Anthropic's narrow economic moat derived from intangible assets like research know-how and cost advantages from multi-chip orchestration and long-term compute contracts. The discussion covers revenue forecasting based on enterprise budget reallocations and operating leverage that could drive margins above 30% over five to ten years. Valuation perspectives contrast Morningstar's $1.7 trillion base case with the IPO midpoint of $1.33 trillion, attributing the discount to bearish sentiment. Key risks include the expiration of discounted compute deals and the pace of AI adoption in enterprise workflows.

Key takeaways

  1. 1
    Anthropic's cost advantage stems from an orchestration system that routes workloads to the cheapest available hardware (NVIDIA, AMD, Google TPUs, AWS), lowering its cost basis compared to competitors locked into single vendors.

    Unidentified speaker (speaker_2) · 15:26

  2. 2
    Long-term compute agreements with providers like Google and Microsoft allow Anthropic to pay below spot prices, providing a structural cost advantage over new entrants who must buy compute at market rates.

    Unidentified speaker (speaker_2) · 16:37

Episode sections

  1. 0:00 – 2:10

    Introduction and Cybersecurity Context

    Dave Sekera introduces guest Malik Ahmed Khan and praises his prior calls on cybersecurity stocks, noting how AI's role in security has shifted market sentiment from disruption fears to value appreciation.

  2. 2:10 – 4:41

    Anthropic's Business Model and LLM History

    Khan explains that Anthropic builds large language models (LLMs) for enterprise knowledge work, tracing the technology's origins to Google's 2017 'Attention is All You Need' paper and the transformer mechanism.

  3. 4:41 – 6:00

    LLM Use Cases and Evolution

    The discussion identifies coding as the primary current use case for LLMs, with cybersecurity, design, life sciences, and financial/legal research identified as the next major domains for adoption.

  4. 6:00 – 10:05

    AI Terminology: Open vs. Closed vs. Frontier Models

    Khan defines open weight models (public weights), closed weight models (proprietary), and frontier models (leading edge capability/price Pareto frontier), noting that 'open source' is a misnomer because training data is rarely public.

  5. 10:05 – 19:53

    Competitive Landscape and Moat Sources

    Anthropic competes with OpenAI, Google DeepMind, and Meta in the frontier space. Morningstar assigns a narrow economic moat based on intangible assets (research know-how, recursive self-improvement) and cost advantages (multi-chip orchestration, long-term compute contracts).

  6. 19:53 – 26:51

    Enterprise vs. Consumer Strategy

    Khan argues Anthropic's focus on enterprise is superior to consumer models because enterprises have a higher willingness to pay for safety and security, unlike consumers who prefer free, ad-supported services dominated by Meta and Alphabet.

  7. 26:51 – 32:21

    Revenue Forecasting and Adoption Curves

    Khan outlines Morningstar's revenue forecasting method, which relies on tracking Annual Recurring Revenue (ARR) and analyzing budget reallocations from traditional software/IT services to AI, rather than using abstract total addressable market percentages.

  8. 32:21 – 35:25

    Compute Fungibility and Monetization Strategy

    Speaker 2 explains that for model companies like Anthropic, research compute and cost of sales are fungible, allowing a split between training (R&D) and inference (COGS). Currently, 25% of compute is used for inference to generate sales, while 75% is dedicated to training, similar to how Ferrari races to sell cars.

  9. 35:25 – 37:30

    Operating Leverage and Margin Outlook

    The speaker contrasts Anthropic's high revenue per employee against Salesforce, noting that scaling on compute provides significant leverage compared to traditional software. They forecast operating margins north of 30% over five to ten years due to this leverage and decreasing cost per token.

  10. 37:30 – 40:15

    Compute Commitments and Funding Strategy

    Speaker 2 details $902 billion in forward compute commitments, split between $450 billion through 2030 and $400 billion thereafter. They argue that Anthropic can fund this via $130 billion in cash, an expected $90-100 billion post-IPO raise, and revenue growth, despite potential free cash flow pressure in 2027-2029.

  11. 40:15 – 44:29

    Free Cash Flow Dynamics and Sweetheart Deals

    The analyst notes that free cash flow will likely remain negative until 2030 due to ramping compute costs as discounted 'sweetheart deals' with Amazon and Google expire. They identify three key risks: AI adoption pace, model capability improvements, and value capture mechanisms.

  12. 44:29 – 48:01

    Valuation and Market Sentiment

    Morningstar's base case valuation is $1.7 trillion, compared to an IPO midpoint of $1.33 trillion. The speaker attributes the lower IPO pricing to bearish market sentiment following the SpaceX IPO peak, distinguishing between the pricing event of an IPO and the fundamental value of the business.

  13. 48:01 – 51:12

    Bull and Bear Case Scenarios

    The bull case values Anthropic at $3.4 trillion, assuming it captures 15% of software/IT budgets and takes share from competitors like Google and OpenAI. The bear case values it at $500 billion, assuming slow AI diffusion and reluctance from companies to adopt agentic workflows.

  14. 51:12 – 56:24

    Lockup Periods and Share Availability

    Anthropic has staggered lockup periods, with shares becoming available from day one up to day 76, then after fiscal 2026 closes, and finally after one year. The speaker advises investors to focus on long-term fundamentals rather than near-term selling pressure from insiders and early investors.

Claims worth verifying

Anthropic is building large language models designed for knowledge work areas such as cybersecurity and interacting with Microsoft Office.

Unidentified speaker (UNKNOWN) ·2:14

How to verify: Company website and product documentation for Claude; Morningstar research reports on Anthropic. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

The large language model breed was kicked off in 2017 by a Google paper called 'Attention is All You Need' which introduced the transformer mechanism.

Unidentified speaker (speaker_2) ·2:54

How to verify: Search for the paper 'Attention is All You Need' on arXiv.org, published by Google researchers in 2017. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Anthropic does not do video generation or image generation; those capabilities are more consumer-facing.

Unidentified speaker (speaker_2) ·3:50

How to verify: Review Anthropic's product offerings on their official website; compare with OpenAI's Sora or Midjourney. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Anthropic paid a settlement north of $1 billion related to training data issues.

Unidentified speaker (speaker_2) ·8:50

How to verify: Search for 'Anthropic settlement training data' in news archives or legal filings; verify the amount and parties involved. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

There are roughly four frontier research labs: Anthropic, OpenAI, Google DeepMind, and Meta.

Unidentified speaker (UNKNOWN) ·10:05

How to verify: Industry reports from AI research aggregators (e.g., Hugging Face Leaderboards) and market analysis from firms like Gartner or IDC. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Morningstar rates Anthropic with a narrow economic moat.

Unidentified speaker (speaker_3) ·12:09

How to verify: Access Morningstar's equity research report for Anthropic (private company valuation report) or the published article 'Why Morningstar Thinks Anthropic Is Worth $1.7 Trillion'. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

26% of research tasks within Anthropic are led by Claude, up from below 10% a couple of years ago.

Unidentified speaker (speaker_2) ·14:20

How to verify: Verify via Anthropic's internal reports or public statements regarding their 'recursive self-improvement' or RSI loop metrics. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Anthropic has built an orchestration system allowing models to run on NVIDIA GPUs, AMD GPUs, Google TPUs, and AWS chips.

Unidentified speaker (speaker_2) ·15:12

How to verify: Review Anthropic's technical blog posts or press releases detailing their infrastructure partnerships and multi-chip training capabilities. 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 signed long-term compute commitments with Google, Microsoft, Amazon, NVIDIA, AMD, and Broadcom running through 2033.

Unidentified speaker (speaker_2) ·16:37

How to verify: Check SEC filings if Anthropic has filed for IPO (S-1) or press releases from the respective hardware/cloud providers detailing contract durations. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Anthropic's Annual Recurring Revenue (ARR) is $81 billion.

Unidentified speaker (speaker_2) ·28:30

How to verify: Verify via Morningstar's valuation model for Anthropic or the company's own financial disclosures upon IPO or in 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

Anthropic dedicates roughly 25% of its total compute to inference and 75% to training.

Unidentified speaker (UNKNOWN) ·33:40

How to verify: Check Anthropic's amended S-1 filing on SEC EDGAR for details on compute allocation or revenue recognition policies. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

The quarter of compute dedicated to inference carries a $50 billion business in gap sales.

Unidentified speaker (UNKNOWN) ·33:56

How to verify: Verify against Anthropic's revenue figures in the amended S-1 filing on SEC EDGAR. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Morningstar forecasts Anthropic's operating margins to be north of 30% over the next 5-10 years.

Unidentified speaker (speaker_2) ·36:25

How to verify: Review Morningstar's research report 'Why Morningstar Thinks Anthropic Is Worth $1.7 Trillion' linked in the episode description. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Anthropic has $902 billion in forward compute commitments, split into ~$450 billion through 2030 and ~$400 billion thereafter.

Unidentified speaker (UNKNOWN) ·37:36

How to verify: Check the amended S-1 filing on SEC EDGAR for the section on 'Risk Factors' or 'Management's Discussion and Analysis' regarding contractual obligations. 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 $130 billion in cash on its balance sheet and expects to raise $90-100 billion post-IPO.

Unidentified speaker (speaker_2) ·38:41

How to verify: Verify cash balance in the amended S-1 filing on SEC EDGAR and raise amount in the IPO pricing announcement or S-1. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Morningstar's base case valuation for Anthropic is $1.7 trillion.

Unidentified speaker (speaker_2) ·44:32

How to verify: Review Morningstar's research report 'Why Morningstar Thinks Anthropic Is Worth $1.7 Trillion' linked in the episode description. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

The midpoint of Anthropic's IPO range was $1.33 trillion.

Unidentified speaker (UNKNOWN) ·44:39

How to verify: Check the final IPO pricing announcement or the amended S-1 filing on SEC EDGAR. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Morningstar's bull case valuation is $3.4 trillion and bear case is $500 billion.

Unidentified speaker (speaker_2) ·48:01

How to verify: Review Morningstar's research report 'Why Morningstar Thinks Anthropic Is Worth $1.7 Trillion' linked in the episode description. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

In the bull case, Anthropic captures north of 15% of software and IT services budgets.

Unidentified speaker (UNKNOWN) ·49:07

How to verify: Review Morningstar's research report 'Why Morningstar Thinks Anthropic Is Worth $1.7 Trillion' linked in the episode description. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Anthropic's base case CAGR is 47%, bull case is north of 60%, and bear case is around 22% over five years.

Unidentified speaker (speaker_2) ·50:10

How to verify: Review Morningstar's research report 'Why Morningstar Thinks Anthropic Is Worth $1.7 Trillion' linked in the episode description. 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 Anthropic's amended S-1 filing for detailed breakdowns of the $902 billion in forward compute commitments and associated risks.
  • Verify Morningstar's $1.7 trillion base case valuation assumptions against current market multiples for comparable AI infrastructure companies.
  • Analyze the expiration timeline of Anthropic's discounted compute agreements with Amazon and Google to assess future free cash flow pressure.
  • Compare Anthropic's revenue per employee and operating margin forecasts against historical data from Salesforce and other enterprise software leaders.
  • Investigate the legal settlement details regarding Anthropic's training data issues to understand potential ongoing litigation risks.
  • Assess the competitive landscape by reviewing recent product releases and enterprise adoption metrics from OpenAI, Google DeepMind, and Meta.
  • Check lockup period schedules to identify potential selling pressure from insiders and early investors post-IPO.
Full transcriptexpand
0:00
Unidentified speaker (speaker_0): Please stay tuned for important disclosure information at the conclusion of this episode.
0:23
Dave Sekera: Hello, I'm Dave Sakara, Morningstar's Chief U.S. market strategist, and the co-host of the weekly podcast, The Morning Filter. Today I'm joined by Amik Khan, senior equity analyst on our technology team, who covers two of the largest companies in the world by market cap, alphabet, and meta, as well as the cybersecurity industry. Today is Tuesday,
0:43
Dave Sekera: November 17th, and we're conducting a bonus episode of our podcast to conduct a deep dive into our analysis and valuation. of anthropic. Ahmed, thank you very much for joining us today and sharing your expertise with us. But even before we start talking about anthropic, I also have to commend you on some of your prior calls. So we had our podcast or
1:03
Dave Sekera: our bonus episode a number of months ago on Cybersecurity Socks, talked about your view of cybersecurity overall and how you thought AI actually was going to make cybersecurity even more valuable. and more important going forward. Whereas I think it was last fall and even earlier this year, the market was much more concerned that AI would disrupt or displace the
1:23
Dave Sekera: cybersecurity industry. A lot of those stocks sold off, getting well into four and five star territory. And as we've seen the past couple of months, a lot of those have rallied back. And not only are there three-star three star stocks, some of them have moved up enough that they're now even into two-star territory. So from a lot of our viewers' point of view, thank you very much. I think a lot of people have done very well on
1:43
Dave Sekera: those.
1:45
Unidentified speaker (speaker_2): First of all, thank you so much.
1:47
Unidentified speaker (speaker_2): Great to be here, Dave.
1:50
Unidentified speaker (UNKNOWN): All
1:50
Dave Sekera: right. Well let's get down to uh the business here. So Anthropic. So we've got the IPO, it's happening. So just to start off with, I mean, could you just describe an overview of what Anthropic actually does? I think a lot of people may necessarily, you know, have heard of the term of AI, but they may not actually really have a good deep understanding
2:10
Dave Sekera: of what Anthropic is and what their product is.
2:14
Unidentified speaker (UNKNOWN): At
2:14
Unidentified speaker (speaker_2): a very, at a very high level, what Anthropic does is that they build large language models or LLMs. LLMs LLMs that are essentially complicated or complex uh AI models uh that are designed. uh to complete uh complete tasks that are different than than how we've perceived AI uh uh uh in
2:34
Unidentified speaker (speaker_2): the past. So a lot of a lot of AI models that are excelling in knowledge work areas such as cyber, such as um you know interacting with uh uh with with Microsoft Word or PowerPoint and actually helping uh workers or knowledge workers. uh become more productive. Um this
2:54
Unidentified speaker (speaker_2): entire sort of like breed of models was was essentially kicked off in in in 2017 uh by this uh Google paper uh published by Google researchers called Attention is All You Need.
3:07
Unidentified speaker (UNKNOWN): Um,
3:08
Unidentified speaker (speaker_2): and it introduced like this transformer uh uh uh mechanism that has since then developed into the these large language models. And and you've seen the size of these models scale up quite rapidly. And you've seen capabilities scale up quite rapidly, right? So if we sort of like go back, you know, to
3:28
Unidentified speaker (speaker_2): 2022 and 2023,
3:30
Unidentified speaker (UNKNOWN): a
3:30
Unidentified speaker (speaker_2): lot of us perceived large language models to be like cool party tricks. Uh uh, we all know the you know, Will Smith eating spaghetti. videos that went around and how those videos have subsequently sort of improved. And that's a very good proxy at a high level of how the capabilities of these large language models have
3:50
Unidentified speaker (speaker_2): improved. So Anthropic, you know, at its core, builds these large language models and then it sells them primarily to enterprises. So, you know, Anthropic doesn't do video generation, doesn't do image generation.
4:04
Unidentified speaker (speaker_2): Those capabilities, you know, are more consumer facing. They're more focused on selling tools to entreprises in some key domains, such as cyber, that we've previously obviously talked about, but also other facets of knowledge work.
4:22
Dave Sekera: And just to maybe help people understand better, I mean, as they sell these large language models, into different enterprises, what are some of the use cases that you see people using these models for today? And maybe what's you know in the future over the next couple of years and how enterprises will adapt these large language models to their businesses?
4:41
Unidentified speaker (speaker_2): For sure. So uh the first killer use case for large language models has obviously been coding. And all of us have heard about uh developers using plot Claude code. or Google's anti-gravity or OpenAI's codex, all these sort of different applications to you know essentially help with coding workflows. Uh these large language models uh
5:01
Unidentified speaker (speaker_2): have gotten to the point where a lot of large companies are basically there's very little human code being written. It's all being written by these large language models. Um, the next few areas that we think are going to be areas of focus. for these large language model companies or AI model companies are
5:20
Unidentified speaker (speaker_2): cybersecurity, there's design, there's life sciences, there's financial and legal research. We think that these sort of like five domains are
5:31
Unidentified speaker (speaker_2): going to be the next sort of target areas once we we've have sort of like traversed through that trajectory in.
5:39
Unidentified speaker (speaker_2): coding.
5:40
Dave Sekera: Well, if they get too much into financial research, uh, I guess maybe I'm glad I'm closer to being retired now than I was uh in the past. Now, I've also heard you know a lot of other terms out there which may not necessarily be clear to a lot of our viewers. We've heard things called frontier models. We've talked about like or heard about like open weight models, closed weight models. Can you maybe just
6:00
Dave Sekera: briefly describe what those are and maybe is that something that relates to large language models or are those completely different facets of AI?
6:08
Unidentified speaker (UNKNOWN): No,
6:09
Unidentified speaker (speaker_2): for sure. So all these terms are interlinked, and it can be quite a lot to consume all these terms and how they sort of interlink with each other. So I'll try to give uh our viewers like just a high level overview of what these terms mean. First of all, when you when you train a large language model, you can think of it as
6:28
Unidentified speaker (speaker_2): different. So if you think of large language models as essentially next token predictor machines. So if I say the cat is out of the and I fill in the and there's a fill-in-the-blank there, different words will be given different weights.
6:47
Unidentified speaker (speaker_2): So cat is out of the bag will have the highest weight. followed by let's say a house or or hat or or something of that of that of that sort, versus a cat is out of the kennel may may be maybe may may be associated with a lower weight. So what these large language models do is that they
7:06
Unidentified speaker (speaker_2): essentially look at all uh texts Text that could potentially sort of uh follow uh uh that sentence and they assign weights to what the next term uh uh would most likely be. That's why you'll hear terms that uh large language models are inherently probabilistic, not deterministic, because of this sort of like probabilistic
7:26
Unidentified speaker (speaker_2): nature of assigning weights uh uh uh uh to what the next sort of like token uh uh uh would be or next word would be. Now, if you're building this large language model,
7:38
Unidentified speaker (UNKNOWN): what
7:39
Unidentified speaker (speaker_2): you end up with is, and this is gonna surprise a lot of people, you end up with a file that you can store on a USB drive. So it's like, you know, let's call it like 10, 15 gigabytes, and that is your large language model. It's just a series of weights, a collection of all words that could proceed and and come after certain words. And this large language model,
7:59
Unidentified speaker (speaker_2): this combination of weights, if you decide to publish it.
8:03
Unidentified speaker (speaker_2): uh openly to people, for people to view those weights and and see how uh things are set up,
8:10
Unidentified speaker (UNKNOWN): um,
8:10
Unidentified speaker (speaker_2): you have an open weight model. If you choose to keep it proprietary, uh, you end up with a closed weight model. And now the reason why um uh I I sometimes push back on people calling it open source is that open source would also uh uh uh imply that we had access or we would have access to what the
8:30
Unidentified speaker (speaker_2): models were trained on. And uh, and that's like a uh a big no-no in in the in the AI model space, primarily because uh uh I'm I'm I'm I'm I'm fairly confident that all AI model companies trained on stuff that they were not supposed to train on. There have been obviously uh large settlements such as a BART settlement, that philanthropic Anthropic
8:50
Unidentified speaker (speaker_2): uh uh paid I hit, I believe, north billion dollars.
8:54
Unidentified speaker (speaker_2): For
8:55
Unidentified speaker (UNKNOWN): um,
8:56
Unidentified speaker (speaker_2): in terms of your set, your the other question of what is a frontier model? Frontier just you can think of it as a leading edge model, right? So a model that is at the leading edge in terms of capability. There's another sort of interlinked concept, which is the Pareto frontier, which is for any given level of
9:16
Unidentified speaker (speaker_2): intelligence, what is the cheapest model? So you can think of it as the interaction between a capability and price. And you can, if you look at a series of models, you can sort of model the Pareto frontier. And you can see both Anthropic and OpenAI, as well as companies like Google, try to push that Pareto burrito
9:35
Unidentified speaker (speaker_2): frontier uh uh uh forward. So those would be some some. so of the key uh terms that I think would be important to would be important to define when having a conversation about uh AI models.
9:47
Unidentified speaker (speaker_3): Okay. So then I'm also just trying to understand too like, you know, who are the direct competitors, you know, to Anthropic and, you know, how do you classify or think about how Anthropic's models, you know, may maybe be different, maybe similar to some of those competitors. And what really sets Anthropic apart, you know, from their competition today?
10:05
Unidentified speaker (UNKNOWN): For
10:05
Unidentified speaker (speaker_2): sure. So first of all, um, just to name the competitors that
10:09
Unidentified speaker (UNKNOWN): in
10:09
Unidentified speaker (speaker_2): the frontier model space, uh uh we have roughly four frontier research labs right now. That's Anthropic, OpenAI, Google's Deep Mind, so GDM, and uh Meta, so Meta's research lab. Uh people will throw in XAI XAI's as Grok there as well, but if you think about the companies that are
10:29
Unidentified speaker (speaker_2): actually at the leading edge, I would probably say
10:32
Unidentified speaker (UNKNOWN): uh
10:33
Unidentified speaker (speaker_2): OpenAI, Anthropic, and and
10:35
Unidentified speaker (speaker_2): Google are right there at the frontier, with Meta very close uh uh to the frontier.
10:42
Unidentified speaker (UNKNOWN): Um,
10:42
Unidentified speaker (speaker_2): in terms of the overall competitive set, it's a bit broader because it also includes open weight labs that are primarily out of China, companies like uh uh uh Tencent and Alibaba and and and Kimi and a host of other sort of open weight uh uh uh model uh uh research labs. DeepSeek is another one as well.
11:02
Unidentified speaker (speaker_2): The sort of difference between
11:05
Unidentified speaker (UNKNOWN): the
11:05
Unidentified speaker (speaker_2): frontier and then some of the open weight models are that open weight uh uh models tend to be fast followers. Um, there's been a lot of conversation.
11:14
Unidentified speaker (UNKNOWN): that
11:14
Unidentified speaker (speaker_2): I'm sure everyone here has heard of is of distillation of OpenAI or open weight models actually copying the outputs of frontier models and sort of becoming fast followers. We see that we see Frontier Research Labs as cracking down on that, on that distillation in order to sort of like
11:34
Unidentified speaker (speaker_2): prevent the leakage of their secret sauce and allow them to uh out-earn. uh their cost of capital over time. Um but that's the that's the broader sort of uh competitive uh competitive landscape uh facing anthropic right
11:48
Unidentified speaker (UNKNOWN): now. Yeah.
11:49
Unidentified speaker (speaker_3): Well that actually dovetails right into my next question. And you were talking about their ability to earn a long-term return over their weighted average cost of capital. So in this case, you know, Morningstar has rated Anthropic with a narrow economic moat. And I have to imagine that was probably a pretty contentious moat meeting for the other committee on this one.
12:09
Unidentified speaker (speaker_3): But if you could just maybe quickly walk us through what you think the primary moat sources are and why do you think that they provide that long-term durable competitive advantage to Anthropic?
12:21
Unidentified speaker (UNKNOWN): First
12:21
Unidentified speaker (speaker_2): of all, Dave, you're absolutely right. Uh It was a very contentious uh moat mode committee meeting. Uh the two mode sources uh that we think uh Anthropic has uh that that that that um undercurred its its its narrow economic mode are intangible assets. Now, intangible assets, the way we look at it is frontier research know how.
12:41
Unidentified speaker (speaker_2): Like um AI is probably one of the very few areas in broader technology that is
12:50
Unidentified speaker (UNKNOWN): where
12:50
Unidentified speaker (speaker_2): you have a superstar effect, where like a small group of AI researchers can actually drive a ton of progress. And we've seen that actually happen out, happen. Uh uh, I mean, GDM or Google DeepMind is a very good example of this, where you can just like go down and do research. You can build large language models and you can sort of scale those models up. And
13:10
Unidentified speaker (speaker_2): then when you're scaling those models
13:12
Unidentified speaker (UNKNOWN): up, um,
13:12
Unidentified speaker (speaker_2): you have to run a ton of experiments. You have to, in AI parlance, it's called hill climb. So start with smaller problems, solve them as you get to bigger and bigger models. And all the all that learning, it's, it's, it, it accumulates. And that's why we've seen the uh the model landscape really whittle down to, as I mentioned, uh three or
13:32
Unidentified speaker (speaker_2): four frontier For research companies. If you if you go back to 2023 or 2022,
13:38
Unidentified speaker (UNKNOWN): um
13:38
Unidentified speaker (speaker_2): there were more than a dozen. There was Cohere in Canada and Mistral in Europe, and all of them have fallen off the frontier because of this sort of like frontier research know-how and how those learnings accumulate over time. The second intangible asset, which is worth uh keeping in mind, is
13:57
Unidentified speaker (UNKNOWN): this
13:57
Unidentified speaker (speaker_2): idea of closing the loop, which refers to
14:01
Unidentified speaker (speaker_2): AI models. helping train and build the next slate of AI models. And this loop, which is in again in industry language, it's called the recursive self-improve or self-improvement loop or RSI loop. 26% of
14:20
Unidentified speaker (speaker_2): research tasks within Anthropic Enthropic are being led by Claude. That number was below 10%, just a couple of years ago. And we've seen a dramatic uptick. And by the way, this is not just Enthropic. We're also seeing similar trends at Google, seeing similar trends at OpenAI. And those are some of the
14:40
Unidentified speaker (speaker_2): intangible assets that these companies, uh these companies possess.
14:45
Unidentified speaker (speaker_2): The second mode source is a cost advantage.
14:48
Unidentified speaker (UNKNOWN): The
14:49
Unidentified speaker (speaker_2): way we think about a cost advantage is
14:52
Unidentified speaker (speaker_2): when we're looking at a company like Anthropic, you think about okay, what are its primary sort of costs? When you're running these models, training these models, serving these models to users, what are the costs that you incur? And it's primarily hardware costs, right? So you have to run these models on expensive AI accelerators or
15:12
Unidentified speaker (speaker_2): AI chips. What anthropic has done very smartly is that they have built
15:21
Unidentified speaker (speaker_2): an orchestration system where
15:26
Unidentified speaker (speaker_2): its models can be trained and run on NVIDIA GPUs, AMD GPUs, Google TPUs, and AWS chips, or Amazon Web Services chips. And what that allows them to do is that it allows them to route, if they get a workload, they get to route that workload to the cheapest chip
15:46
Unidentified speaker (speaker_2): possible. And that structurally lowers their cost. We've seen a lot of uh commentary on compute multipliers, which is essentially how much
15:57
Unidentified speaker (speaker_2): output or how much like actual like you know uh uh uh tokens you can get out of of a of of off of a hardware uh asset.
16:06
Unidentified speaker (UNKNOWN): And
16:06
Unidentified speaker (speaker_2): we've seen those compute multipliers sort of inflect quite dramatically um for anthropic uh over the last over the last year in particular, but over since the company has sort of like started this journey of training uh large language models a few years back. Uh And then a related sort of concept when it comes to our our our V1 cost advantages is is our
16:26
Unidentified speaker (speaker_2): are the compute agreements that the company has signed with the likes of Google and Microsoft and Amazon and NVIDIA and AMD and Broadcom.
16:37
Unidentified speaker (speaker_2): If Anthropic is going and and signing a deal with Google that is going to run through 2033, they're not going to pay the spot price. They're going to pay a price that's below that because it they give Google that certainty that they can go ahead and build out their infrastructure and then and the and a buyer is going to be there. And these long-term compute
16:57
Unidentified speaker (speaker_2): commitments also, again, structurally lower their cost basis versus a new entrant who may have to go ahead and buy computed spot prices, which we know to be significantly greater than those than those long-term compute commitments that the company has signed.
17:13
Unidentified speaker (UNKNOWN): So
17:13
Unidentified speaker (speaker_2): those are the uh the two. two primary mode sources, two emerging mode sources that we also highlight. We do not have enough confidence in them in terms of we think that they exist right now, but we're seeing sort of positive developments along both fronts or switching costs.
17:29
Unidentified speaker (speaker_2): There's a ton of research out there that if a company, an entrepreneur, tries to rip About out their primary AI provider, it actually leads to a lot of disruption. So we're seeing some of that, you know, the trigger. traditional software stickiness be built up in
17:45
Unidentified speaker (speaker_2): the model uh space, especially when we're thinking about not just APIs, but we're thinking about long-running agents and specifically vertical applications such as Cloud Code or Cloud Design, et cetera. And then finally, network effect, uh, which is uh uh which the way we see it is. that more people use Claude,
18:05
Unidentified speaker (speaker_2): especially the ones that are not protected by an entreprise zero data retention agreement, Claude actually or anthropic, uh, uses their data, figures out what the failure modes are, how to sort of like improve their products. They improve their products, more people use the product, and there's there is this like nice flywheel effect. Uh, but again,
18:25
Unidentified speaker (speaker_2): uh the switching costs and network effect. effects are emerging uh mode sources, something that we'd flag for investors to keep an eye on. Um
18:32
Unidentified speaker (UNKNOWN): but
18:32
Unidentified speaker (speaker_2): intangible assets and and and and and and the and the and the cost advantage are are the two mode sources that we think uh sort of uh the economic mode is built on today.
18:43
Unidentified speaker (UNKNOWN): Well,
18:43
Unidentified speaker (speaker_3): and to be honest, like the first I really heard of artificial intelligence and its use cases was you know when OpenAI chat ChatGPT GPT first came out.
18:51
Unidentified speaker (speaker_3): and everybody was playing with that online. And then it wasn't until, I don't even remember, like maybe 18 months ago, I started hearing about Claude and Anthropic. So of course, you know, technology evolves very quickly. So I guess it it's maybe a little bit of a mixed question here, but yeah, how can you be confident that when technology evolves that quickly and we see changes
19:11
Unidentified speaker (speaker_3): in leadership in such a short amount of time? Yeah, how can you be confident really that they can maintain that leadership, maintain that long-term durable competitive advantage over the next 10 years? And I guess maybe to loop in with that same question, you know, we've gotten other people asking, how many large language model providers do we need? Is this something like the
19:31
Unidentified speaker (speaker_3): search industry where all you really need is Google? They've got a huge amount of the market share, everyone else follows, you know, very, very far. behind, very small market shares. Or is this really a space where you can have, like in this case, you know, three different LLM providers out there, and each of them being able to provide their own differences and why people might pick one
19:51
Unidentified speaker (speaker_3): versus another?
19:53
Unidentified speaker (speaker_2): For sure. So
19:56
Unidentified speaker (UNKNOWN): I
19:56
Unidentified speaker (speaker_2): think the first thing to keep in mind is that
20:00
Unidentified speaker (speaker_2): entreprise and consumer are two distinct. different markets.
20:05
Unidentified speaker (speaker_2): And a very good example of this is actually the Google example that you brought up, where Google was a consumer giant, but in the entreprise space, uh, I think everyone would agree that Microsoft was the most dominant uh player, especially when it comes to knowledge work tools such as Microsoft Office and And and all the associates, associated tools there. Um when it comes to
20:25
Unidentified speaker (speaker_2): AI, I think it is helpful to think about it along those two axes, consumer and uh and and entreprise. I think Anthropic has correctly, in our opinion, made the bet that enterprise is a more lucrative uh place to be.
20:41
Unidentified speaker (UNKNOWN): Uh,
20:41
Unidentified speaker (speaker_2): I can I can sort of walk through uh what the underlying reasons are. Primarily, it's to do with consumers due to companies like Meta and Alphabet. Our default is not to pay for services online. Right. We're we're we're in this default setting where we're okay watching advertising or advert ads, but we're not going to pay subscription fees
21:01
Unidentified speaker (speaker_2): for a product that we can get for free. Um there's a lot of like super interesting research on this. Like uh, for example, the European Union uh tried uh to get uh Meta to offer its Facebook application. Um uh without ads as a paid uh service. And I think less than a percent of European
21:21
Unidentified speaker (speaker_2): users signed up for that, despite that being available. Um and on the other side, entreprises have the exact opposite default setting, where entreprises actually prefer to pay. And the reason for that, and again, it's it's it's all about revealed preference, right? Like open source software has been around. since software has been around, but more
21:41
Unidentified speaker (speaker_2): like the vast majority of entreprises actually prefer to have uh uh an entreprise agreement with a proprietary software company, even if the proprietary software company is building on top of open source software. Elastic is a good example of this. Um
21:56
Unidentified speaker (UNKNOWN): and
21:56
Unidentified speaker (speaker_2): I think the the reason for enterprises uh sort of uh preferring uh entreprise agreements with with companies. is it comes down to uh safety, security, SLAs, uptime requirements, etc. So I think Anthropic correctly sort of understood that entreprises because their default is actually
22:16
Unidentified speaker (speaker_2): to pay, uh, they opted for that, uh for that for that market. OpenAI uh um you know famously went down the consumer route. They have more than 1.2 billion weekly active. users. And they also kind of recognize that enterprise is a more lucrative market because they have also been prioritizing enterprise. They shut down Sora, which was their video
22:37
Unidentified speaker (speaker_2): app, because it's just really expensive to serve users without having a scaled-up advertising business. And as soon as you get into advertising, you're coming up against the two behemoths that I cover, Alphabet and Meta, companies that have really perfected the art of showing the right ad at the right moment to the right user.
22:55
Unidentified speaker (UNKNOWN): Um,
22:56
Unidentified speaker (speaker_2): and I think Athropic just did not want to get into that sort of knife fight uh uh uh dynamic. In terms of your earlier question about what gives us the confidence that these like Athropic will be able to out-earn its cost of capital, is first of all, their core focus on the enterprise.
23:14
Unidentified speaker (UNKNOWN): Number two.
23:15
Unidentified speaker (speaker_2): their core focus within the enterprise as well. Right. So if you think about an entreprise's series of different workflows, they have really focused on areas such as coding, such as cybersecurity, such as legal and financial research, such as design and life sciences. And some of the commonalities between these
23:35
Unidentified speaker (speaker_2): these these areas is that uh users with within these uh the within these um verticals tend to be very highly compensated. So you know attaching a $1,000 per month Claude license on top of um a financial analyst that is making way greater uh than that and making them more
23:55
Unidentified speaker (speaker_2): productive is is it's it's it's it's very sensible or very logical for an entreprise to
24:00
Unidentified speaker (UNKNOWN): do.
24:01
Unidentified speaker (speaker_2): Um they also tend to have very high bars of like what good is. So, you know, like if you're a financial analyst or you're a le you know lawyer, uh if you give a task to the AI to do, you need to have very high confidence that there is not going to be any hallucination, not going to be any sort
24:21
Unidentified speaker (speaker_2): of like, you know, uh uh mistakes.
24:23
Unidentified speaker (UNKNOWN): Um
24:24
Unidentified speaker (speaker_2): and all of these essentially translate to a high willingness. to pay,
24:28
Unidentified speaker (UNKNOWN): uh,
24:28
Unidentified speaker (speaker_2): which the company is really tapping into. And then they're building vertical applications. Clawed code is a very good example of this, where you can actually build a vertical application and scale it
24:39
Unidentified speaker (UNKNOWN): uh
24:39
Unidentified speaker (speaker_2): to billions of dollars. And we're seeing the companies sort of run the same playbook in all these different domains that I mentioned.
24:45
Unidentified speaker (UNKNOWN): And
24:46
Unidentified speaker (speaker_2): that sort of like again uh uh undergirds our confidence or or or or support. supports our confidence in the company's ability to out earn its cost of capital. But at the same time, I think it's also important to sort of like remain intellectually honest and say that, hey, this is it is there is a lot of uncertainty. You know, you talked about technological uncertainty. There's an
25:05
Unidentified speaker (speaker_2): uncertainty of AI adoption. So, you know, Morningstar, for example, we've been we've been adopting AI across various departments.
25:13
Unidentified speaker (UNKNOWN): Uh
25:13
Unidentified speaker (speaker_2): what what is the what is the pace of adoption you know going to be in the next couple of years uh the historical analogy that i like using is is is is the adoption of radios versus the adoption of washing machines radios if you look at when from when they went from 10 penetration to 90 penetration it was a period of 10 years
25:33
Unidentified speaker (UNKNOWN): to
25:33
Unidentified speaker (speaker_2): go for for washing machines to get to the same like 90 it took a multi-decade period so if you're if you're in the camp that you know AI is gonna is gonna is gonna is gonna proliferate across enterprises kind of like radios or or even some of the recent technologies like the internet or mobile, you're probably gonna end up being more bullish on AI. Versus if you
25:53
Unidentified speaker (speaker_2): think this is gonna be a multi-decade period, um, you know, uh perhaps some of the adoption curves uh uh uh do not make uh sense uh uh uh to you. And that would obviously call into question
26:05
Unidentified speaker (UNKNOWN): uh
26:06
Unidentified speaker (speaker_2): anthropics' ability to outpreneur its cost of capital.
26:10
Unidentified speaker (UNKNOWN): So I
26:10
Unidentified speaker (speaker_3): guess that leads into the next question. So how do we actually forecast that adoption? When I'm thinking about with what's going on with AI right now, the AI build-out boom, the number of data centers, the size of data centers that are being built out is you know certainly mind-boggling at this point in time. So, how do we forecast or how do you think about just how much AI
26:30
Unidentified speaker (speaker_3): compute is going to be needed next year? How do you think about how much AI compute power is going to be needed, you know, three years from now, five years from now? And I guess the other part of that is like, how can we be confident, or how do we think about like what are the use
26:44
Dave Sekera: cases that are going to be developed over the next couple of years that really require that amount of compute?
26:51
Unidentified speaker (speaker_2): For sure. First of all, I would just say that
26:56
Unidentified speaker (speaker_2): sometimes people sort of like view the AI trade as one sort of monolith moving together. And certainly the stocks definitely trade like that. But there's two distinct camps. There's a sellers of compute and the buyers of compute. So if we end up in a situation where we have overbuilt, that's actually like net net positive for
27:16
Unidentified speaker (speaker_2): anthropic because they're a buyer of compute. They're not a seller of compute. Now that may be bad for hardware companies or energy companies or any anyone beneath uh Anthropic in that in that AI stack. But I think that's the first distinction I would draw. But still obviously leaves a question is like, okay, Anthropic still has $900 billion dollars of commitments. How are they gonna sort of
27:36
Unidentified speaker (speaker_2): like pay for them? Or how do we get confidence? in how much revenue they're gonna generate in 27 or 28 or beyond? Um I think when we're looking at 27, I think we have a decent understanding of how revenue uh will inflect. Um, and these are you know famous last words because if we were having this conversation in
27:56
Unidentified speaker (speaker_2): 2025 and I showed you Anthropic doing $5 billion in sales, and then I said, hey, Dave. Anthropic's going to do 53 in 26. You would have me, you would have laughed me out of the room, right? And that's just the reality is the the sort of like uncertain or the dispersion of outcomes is so great. But when we're thinking about uh 2027 in particular, we have the
28:16
Unidentified speaker (speaker_2): company's ARR number, uh the annual recurring revenue number, which is inherently forward looking.
28:21
Unidentified speaker (UNKNOWN): Um,
28:22
Unidentified speaker (speaker_2): what you can do is
28:23
Unidentified speaker (UNKNOWN): you
28:24
Unidentified speaker (speaker_2): can sort of have a view on where the company's ARR. will be at year end.
28:29
Unidentified speaker (UNKNOWN): So
28:30
Unidentified speaker (speaker_2): the company's AR right now is $81 billion.
28:34
Unidentified speaker (speaker_2): Let's say they end the year somewhere in the $90 billion range. And then you say, okay, what their year, how much AR are they going to add in 2027? And what does that translate in terms of revenues? So just to for some so so sort of quick mental math. if they ended 2026 with $90 billion
28:54
Unidentified speaker (speaker_2): in sales or ARR, sorry, and they did not add a single dollar of ARR, that would imply, or that and they also did not lose a single dollar of ARR. What that would imply is that 2027 sales, which is gap sales, income statement sales, are gonna be like $90 billion. So when we're thinking about 2026 gap sales being
29:14
Unidentified speaker (speaker_2): somewhere in the range of 50 to 55 billion, and then that growing in our forecast to north of 100, what that is basically what we're saying is that they're going to maintain that ARR of like 90 odd billion and then continue to grow incrementally grow uh that ARR. And that's what gets uh gap sales to be uh to be north of 100 billion. Now
29:34
Unidentified speaker (speaker_2): that's 2027.
29:37
Unidentified speaker (speaker_2): When we're thinking of 2028, 29, 30, and onward, um, we have to have some view on budget reallocations and net new dollars being spent on AI. And I think those are the two sort of correct ways, in my opinion, to think about future growth. Because
29:56
Unidentified speaker (speaker_2): you'll see some market analysis, including by AI companies, that says
30:01
Unidentified speaker (UNKNOWN): our
30:02
Unidentified speaker (speaker_2): total address. addressable market, target addressable market is 40 billion, 40 trillion dollars of all knowledge work that happens. If we capture 1% of that, that implies this many, this much in sales. For me, that sort of analysis is almost nonsensical because you can get to whatever number you want, because that that that capture is
30:22
Unidentified speaker (speaker_2): completely, you know, sort of your input.
30:24
Unidentified speaker (UNKNOWN): Um,
30:25
Unidentified speaker (speaker_2): the way we sort of conduct conducted it. our own analysis on thinking about the target addressable market and what their capture off that market would be is that we started off with total software spend
30:37
Unidentified speaker (UNKNOWN): and
30:37
Unidentified speaker (speaker_2): total IT services spend. So this is spending on companies like Accenture and IBM and Tata and Wii Pro and Infosys and whatnot. And then we have a view on how that spending is going to inflect over the next 10 odd years. And then we think, okay, like how much of those software dollars and IT services dollars actually can be
30:56
Unidentified speaker (speaker_2): reallocated to AI models that are producing the same cognitive work, right? Um, and that's how, and that's how we sort of view, and that is essentially the adoption curve uh uh that that that that we use. We think it's more analytically sound, probably uh gears more uh uh bearish for some
31:16
Unidentified speaker (speaker_2): uh AI investors, especially considering that we have not included like everyday salaries
31:22
Unidentified speaker (UNKNOWN): uh
31:22
Unidentified speaker (speaker_2): that could be potentially part of the target addressable market, especially if you listen to these AI model companies. But we think at least it's it's it's more analytically sound. Um and and we do think that when companies are making the decision to allocate more. dollars to AI, they are thinking along the lines of
31:42
Unidentified speaker (speaker_2): net new dollars and budget reallocations, uh uh, especially when it comes to the enterprise where anthropic is is is the um
31:51
Unidentified speaker (UNKNOWN): is
31:51
Unidentified speaker (speaker_2): the dominant player.
31:54
Unidentified speaker (UNKNOWN): So
31:54
Unidentified speaker (speaker_3): then as I'm thinking about moving down the income statement, you know, you kind of already discussed, you know, how you conceptualize how to drive your revenue forecast for you know not only 2027, but you know, beyond. How do you think about, you know, the operating margin? I mean, can you walk us through like how you think about your assumptions for gross margin, how you'd be able to model out like
32:14
Unidentified speaker (speaker_3): SGNAs or percent of sales, and you know, what are your operating margin forecasts for the next few years?
32:21
Unidentified speaker (speaker_2): For sure. So there there's a few things to unpack here. And and and again, it this this kind of like runs uh face first into how we sort of typically conceptualize software companies or tech companies more broadly.
32:33
Unidentified speaker (UNKNOWN): Uh
32:34
Unidentified speaker (speaker_2): tech companies more broadly, you know, let's say you're a Salesforce. You book a dollar of revenue and then you have cost of sales, which is how much, how many dollars did you need to spend to get that sales? That would be cost of like serving, like compute or or or or the infra infrastructure costs. uh uh uh uh and and uh and and other associated
32:54
Unidentified speaker (speaker_2): charges. When it comes to model companies, you can sort of think of their research compute and cost of sales or cost of goods sold as fungible. And the reason I say that is that if you're if you're anthropic and you have one gigawatt of compute, you know, a stylized
33:14
Unidentified speaker (speaker_2): example, just one gigawatt. of compute, you can dedicate like 90% of that to training, 10% of that to inference. Inference is only time when you actually generate sales.
33:23
Unidentified speaker (UNKNOWN): Uh,
33:23
Unidentified speaker (speaker_2): you can do a difference split, you can dedicate 50% to inference, uh, uh 50% to training, and training costs are going to show up in research and development, uh in the operating sort of like, you know, if you're looking at a at the income statement, and inference costs are gonna show up in cost of goods. sold.
33:40
Unidentified speaker (UNKNOWN): Um,
33:40
Unidentified speaker (speaker_2): that fungibility is is is very interesting to monitor because it actually, like right now, uh we know that roughly a quarter of Anthropic's total compute is being dedicated to inference. So that quarter of total compute is carrying a $50 billion business
33:56
Unidentified speaker (UNKNOWN): in
33:56
Unidentified speaker (speaker_2): in in in in in gap sales. 75% of their compute is just for training newer models. An interesting analogy that I thought of was um was was Ferrari and and and F1, and how uh um typically companies race or or racing teams race in in the
34:16
Unidentified speaker (speaker_2): F1 so they can sell more cars. Uh Ferrari uh sells more cars so they can race in F1. And I think I think it's interesting to think about anthropic from that view, because Cause I think anthropic what their sort of goal is, is to build AGI or ASI or whatever sort of acronym you want, like this powerful intelligence that can go and
34:36
Unidentified speaker (speaker_2): do a ton of tasks. And you know, you can take whatever perspective you want on that, but that's their goal. Inference is just basically the way they can monetize enough to build toward that goal. Um,
34:49
Unidentified speaker (UNKNOWN): so they
34:50
Unidentified speaker (speaker_2): are very like a very sort of research oriented
34:54
Unidentified speaker (speaker_2): team. And when we're modeling the business out,
34:58
Unidentified speaker (speaker_2): research and development remains like a huge cost throughout the business. Even as this revenue scale, we think that that is going to be a big part of their of their overall bill because of because of this, because of this focus on research and training you know, the next model and buying more computer compute and whatnot. If we're thinking about
35:18
Unidentified speaker (speaker_2): the rest of sort of expens the other expenses, um, these would be personnel expenses, et cetera.
35:25
Unidentified speaker (UNKNOWN): Um,
35:25
Unidentified speaker (speaker_2): when you're scaling on compute, you have just you just have way more leverage.
35:30
Unidentified speaker (UNKNOWN): Um a
35:31
Unidentified speaker (speaker_2): good example of this again is is Salesforce. Force And you can sort of look at Salesforce's revenue per employee. And if you do like a 10-year sort of survey. of how that inflected, it basically more than doubled. But it's still like, I think somewhere in the $500,000 per employee range anthropic. That number basically went from where
35:51
Unidentified speaker (speaker_2): Salesforce was to like in the millions. Because again, you when you're when you're when you're scaling on compute
35:59
Unidentified speaker (UNKNOWN): and
35:59
Unidentified speaker (speaker_2): more compute comes online, your actual employee base doesn't need to go. proportionally.
36:04
Unidentified speaker (UNKNOWN): Uh,
36:04
Unidentified speaker (speaker_2): and that also is the advantage of offselling, uh, of having a consumption-based model. As consumption scales, you don't need to scale your sales team, you don't need to scale your engineers at the same rate. And that produces more uh more leverage. So I think in our model, over the next like five to 10 years, we expect their operating margins to be north of 30%,
36:25
Unidentified speaker (speaker_2): which again is a reflection of this leverage. But also on the cost of goods sold side, inference side, they're getting more efficient. They're just like, you know, I mentioned the fungibility of compute and four different hardware chips that they're building on. All of that actually enables them to push their cost of serving uh uh uh uh tokens or AI
36:44
Unidentified speaker (speaker_2): models down, which leads to this natural inflection in both cost of goods sold, but also throughout their throughout. their business.
36:52
Unidentified speaker (speaker_3): Well, and you had mentioned earlier about you know anthropic has contracted towards, and I don't know what that number is anymore, but close to a trillion dollars worth of forward commitments to lock up that compute capacity over the next couple of years. And as I think about like our discounted cash flow model, I mean, And how much clarity do you have as far as you know what those commitments are,
37:12
Unidentified speaker (speaker_3): when the cost of those commitments have to run through the income statement and get expensed off? And I
37:18
Dave Sekera: still think one
37:19
Unidentified speaker (UNKNOWN): of
37:19
Dave Sekera: the
37:19
Unidentified speaker (speaker_3): big questions out there is, you know, how are they going to be able to fund that amount of spend in the next couple of years? And
37:25
Dave Sekera: you
37:25
Unidentified speaker (UNKNOWN): know,
37:25
Unidentified speaker (speaker_3): how
37:25
Unidentified speaker (UNKNOWN): do
37:26
Dave Sekera: you
37:26
Unidentified speaker (UNKNOWN): model
37:26
Unidentified speaker (speaker_3): that versus like your revenue and your cash flow
37:29
Dave Sekera: burn?
37:30
Unidentified speaker (speaker_2): For sure. So uh first of all, in the most recent S1,
37:34
Unidentified speaker (UNKNOWN): uh
37:34
Unidentified speaker (speaker_2): the amended S1, they actually gave
37:36
Unidentified speaker (UNKNOWN): the
37:36
Unidentified speaker (speaker_2): full number, which is 902 billion. north of $900 billion.
37:42
Unidentified speaker (speaker_2): It's split up into essentially two portions. One portion is around $450 billion, that leads all the way to 2030. And then there's an hereafter. So after you know 2030, you have another 400 odd billion. So if you were to take that 450 billion and you were to just straight line it, although we don't think it's a straight line, I think we think it's it more
38:01
Unidentified speaker (speaker_2): backend weighted because like how compute comes online. But even if you were to straight line it, that essentially implies compute costs of slightly north of $100 billion, like $110, $112 billion. Um, which if you think about their revenue uh in your end ARR that I mentioned before, I think they should be able to uh uh support that with even like minimal
38:21
Unidentified speaker (speaker_2): growth in revenue. I think the situation gets much more difficult. for Anthropic if revenue just not just does not grow, it actually declines. I think declines in revenue would be would be would be uh quite quite uh quite difficult for the company to manage. One sort of like buffer that the company is
38:41
Unidentified speaker (speaker_2): built in is that after they're done with their IPO, we're expecting them to raise somewhere in the 90 to 100 billion dollar range.
38:51
Unidentified speaker (UNKNOWN): They
38:51
Unidentified speaker (speaker_2): have $130 billion in cash sitting on their balance sheet right now. They're gonna have another $100 coming up. So that's $230 billion. So depends on how you model it. We actually model, so our model does not call for this like straight line approach. We actually expect them to increase increase their compute commitments. So I think in 2028,
39:10
Unidentified speaker (speaker_2): 2029, our compute costs are not like 110, 120 billion. They're like 160, 170 billion dollars. Because we expect them to continue to add a short-term compute to their bill, which will sort of lead to an increase in their bill. Despite sort of like modeling a more aggressive compute path than what the S1 indicates, we
39:30
Unidentified speaker (speaker_2): still view them being able to sort of like grow their revenues enough where they don't have to do massive capital raises.
39:39
Unidentified speaker (UNKNOWN): Um
39:41
Unidentified speaker (speaker_2): and and and and and be and and and be able to pay their compute using a combination of revenue, but also the two hundred and thirty billion dollars that they've built up on their balance sheet.
39:49
Unidentified speaker (UNKNOWN): Well
39:50
Unidentified speaker (speaker_3): that actually gets right to my next question about like their free cash flow dynamics. Now, taking a look at your model, if I read it correctly, yeah, I don't think they're gonna have a positive operating margin or even like positive free cash flow until 2030. So following the IPO, I mean, are you confident that they have enough after the IPO to be able to fund themselves?
40:10
Unidentified speaker (speaker_3): Or do you think that at some point they may need to come back to the markets for some sort of additional capital?
40:15
Unidentified speaker (UNKNOWN): I
40:15
Unidentified speaker (speaker_2): think first of all, um
40:18
Unidentified speaker (UNKNOWN): if
40:18
Unidentified speaker (speaker_2): you if you if you sort of go through our model, we are modeling for the next few years for the company to be free cash flow negative, as you pointed out. I think that's due to this sort of like increased compute commitments that they have coming online, especially in 27 and 28.
40:32
Unidentified speaker (UNKNOWN): Um,
40:32
Unidentified speaker (speaker_2): these are the deals with Broadcom, the deals with Google, the deals with Amazon. Um, one sort of related concept is, and I get this question from investors, is like, how were they, how was Enthropic able?
40:44
Unidentified speaker (UNKNOWN): You know,
40:45
Unidentified speaker (speaker_2): they raised $140 billion to date and they still have 130 odd out of it. Like, how were they able to sort of build this business out by burning just like eight, nine billion dollars in in total capital? And the reason for that is that they they got like pretty like you know, sweetheart deals from Amazon and Google. It was discounted deals that enabled them uh uh uh to preserve
41:05
Unidentified speaker (speaker_2): that, preserve that cash. Some of those sweetheart deals are gonna go away in 27, 28, and that's gonna that's that's what's basically leading to this ramp up uh in compute costs. which then leads to uh uh a free cash flow burn
41:19
Unidentified speaker (UNKNOWN): or
41:19
Unidentified speaker (speaker_2): negative free cash flow and then operating margin pressure. Um, I think again, like going back, if you're an investor or you're thinking about investing in anthropic, the sort of three key questions that you need to answer
41:32
Unidentified speaker (UNKNOWN): are
41:32
Unidentified speaker (speaker_2): number one, what is going to be the pace of AI adoption? How companies are enterprises are gonna are they gonna spend going to more. and more on AI, or at some point they're gonna get tapped out and they're not gonna see returns? As Enthropic builds these large models, as I said, 75% of their compute is going for training right now. So they're building new AI models. Our capabilities continue
41:52
Unidentified speaker (speaker_2): to they'll will they continue to improve? AKA the returns on training these large language models, are they gonna going to be there in terms of capability improvements for these for these models? And then number three, you can create a beautiful ecosystem and create a ton of value, but have no value capture mechanism. And a very sort of like pertinent example in recent history is the telcos with the
42:12
Unidentified speaker (speaker_2): 4G and 5G built-out. You would argue that they really didn't reap any rewards, but companies like Uber and Lyft and DoorDash and then all these like social media companies really did. So you can create a ton of value and not be able to capture it. Uh, So what the that that value capture mechanism looks like uh for anthropic. Now, our forecast sort of looks at these three questions and
42:32
Unidentified speaker (speaker_2): says entreprise adoption is going to stay strong, which means that yes, you have more compute commitments online, but revenue is able to inflect upward. And and that over time, you know, leads them to have free free cash flows, Those positive free cash flows. Uh, model capability improvements are gonna going to continue. We think that
42:50
Unidentified speaker (UNKNOWN): like if
42:51
Unidentified speaker (speaker_2): anything, they're accelerating. Like recent reports by Meta, which is a third-party research organization, said that Anthropic's internal AI engine is like the progress is inflecting upward. It's accelerating. And 80% of that is being driven by their own models, by Claude.
43:12
Unidentified speaker (speaker_2): And then finally, on the value capture mechanism. we think vertical applications are going to be a big sort of answer to that. So actually embedding first-party applications in key workflows across the domains that I mentioned before is going to be
43:28
Unidentified speaker (speaker_2): the sort of clear value capture mechanism beyond just consumption and API and
43:32
Unidentified speaker (speaker_2): what have you. And all these sort of like things when viewed together,
43:38
Unidentified speaker (speaker_2): point to yes, there is going to
43:40
Unidentified speaker (UNKNOWN): be
43:41
Unidentified speaker (speaker_2): margin pressure and free cash flow pressure
43:43
Unidentified speaker (UNKNOWN): in
43:43
Unidentified speaker (speaker_2): 27, 28, 29. But then they're going to be able to sort of inflect, uh, inflect upward. And and and and the numbers obviously, you know, they're much greater. But like if you think about like how software companies and how technology companies have historically developed, it's this is not a pattern that's like anomalous. Uh, We've seen this pattern play out uh in other in other technology tech companies as they
44:03
Unidentified speaker (speaker_2): drive adoption over
44:04
Unidentified speaker (UNKNOWN): time.
44:06
Unidentified speaker (UNKNOWN): So
44:06
Unidentified speaker (speaker_3): I'm curious. Okay, all is said and done. You've got your base case DCF model, you know, fleshed out. So what is your valuation at the end of the day? And I don't know where the market is today, but like what's the current whisper number for you know the IPO valuation? And I guess the the follow-up there then is: you yeah, know, what are some of the key aspects you think that differentiate what your
44:26
Unidentified speaker (speaker_3): valuation
44:26
Unidentified speaker (UNKNOWN): is?
44:27
Unidentified speaker (speaker_3): versus the whisper number?
44:29
Unidentified speaker (speaker_2): For sure. So
44:32
Unidentified speaker (speaker_2): our our valuation, our base case valuation is one point values of business is 1.7 trillion. If you look at their IPO range,
44:39
Unidentified speaker (UNKNOWN): uh
44:39
Unidentified speaker (speaker_2): I think the midpoint of that range was 1.33 trillion.
44:42
Unidentified speaker (UNKNOWN): Uh
44:43
Unidentified speaker (speaker_2): there were some talks about, you know, just the IPO being so oversubscribed that uh uh bankers may actually push uh the value. valuation higher. We'll see, we'll see how where the where that lands, especially after after the close on Wednesday when we have pricing. We're not expecting pricing to be too dramatically different than the $70 to $78 per
45:03
Unidentified speaker (speaker_2): share uh range, which corresponds to the the 1.2 to 1.4 trillion total fully diluted market cap
45:09
Unidentified speaker (UNKNOWN): uh
45:09
Unidentified speaker (speaker_2): that I mentioned uh before. I think
45:12
Unidentified speaker (UNKNOWN): the
45:12
Unidentified speaker (speaker_2): difference between sort of our view and and what the market is sort of looking at right now, especially with that the in that range that I mentioned, is that we're looking at the fundamental value of the business versus uh uh specifically an IPO, in my opinion, is a pricing event. So you're trying to raise what at what price can you raise like roughly 100 billion or
45:32
Unidentified speaker (speaker_2): 90-odd billion dollars at? I think that's a question that Enthropic is trying to answer, which is somewhat like is related, but it's a different. question than what the underlying value of the business
45:41
Unidentified speaker (UNKNOWN): is. Um
45:43
Unidentified speaker (speaker_2): I think the uh the best time to raise cash in in the AI cycle was June when I when SpaceX IPO that's was peak AI sentiment. Since then the sentiment has really shifted uh uh even in you know in in in in in the last couple of weeks with the news around NVIDIA and and all the all the other sort of you know key
46:03
Unidentified speaker (speaker_2): important players uh in the in the in the AI ecosystem,
46:06
Unidentified speaker (UNKNOWN): I
46:07
Unidentified speaker (speaker_2): think the market sentiment is is is a bit more bearish. And that's the market that Anthropic is IPOAing in. And which I think informs uh their decision to have that range of 1.2 to 1.4 trillion. Uh one final thing I'll say on this uh on this is that
46:22
Unidentified speaker (UNKNOWN): you
46:22
Unidentified speaker (speaker_2): know there's that there's that uh example or or or that that story of uh uh Google buying YouTube and and and the person who's responsible for that purchase uh being asked do you think you paid the right price and and and and the and and and the individual answering uh that no of course I did not pay the right price either I horribly
46:42
Unidentified speaker (speaker_2): underpaid or I horribly overpaid and we're gonna find out in 10 years right and and I think I think I think I think if you if you actually believe in in some of the bullish takes on AI,
46:52
Unidentified speaker (UNKNOWN): uh
46:53
Unidentified speaker (speaker_2): a $1.7 trillion valuation, we're gonna look back at that and say that that was like crazy. Like that was super low for a business that was gonna be this super wide business, super wide economic mode business, uh, and and capture so much of entreprise dollars. Uh, look at how stupid Ahmed was, right? Uh and and and there's also a future state of the world where uh AI adoption is much
47:13
Unidentified speaker (speaker_2): slower, model capabilities stall out. anthropic does not have a clear value capture mechanism. And we look back at the 1.7 trillion and say, huh, like that was that was very bullish of him. And it didn't work out. Right. So I think I think again, I I mentioned this before. I think we just have to remain intellectually humble and recognize that there's a there's a wide range of outcomes and try to sort of like say that, hey, like here's
47:32
Unidentified speaker (speaker_2): our best uh estimate or our best forecast on how what the future isn't looked like, uh, but also like recognize that that that it it is it is it is it it is there's a lot of uncertainty around enthroantic and and and broadly the AI model space.
47:47
Dave Sekera: One I'm curious because I know you also had built a uh an upside case as well as a downside case model. So maybe just really quickly review you know what were the assumption differences in on the upside and the downside case and what were the valuations of those.
48:01
Unidentified speaker (speaker_2): Sure. So we did we did bare we did uh model a bare case and a and a bull case uh for anthropic. Our bull case valuation was $3.4 trillion. Our bare case valuation was $500 billion. Now, somebody would can easily come and sort of push back and say the true bull case valuation in this RSI loop. And then suddenly you have
48:21
Unidentified speaker (speaker_2): ASI and what that means in terms of model capabilities and revenue potential would be would imply a greater valuation. than 3.4. Similarly, a bear could come and say, hey, like this company is going under, the equity value should be zero in a true bear case, right? So just keeping those, uh putting those uh those two sort of like views aside. Um
48:41
Unidentified speaker (UNKNOWN): going
48:42
Unidentified speaker (speaker_2): back to what I mentioned before in terms of like actual capture off of
48:47
Unidentified speaker (speaker_2): software and IT services budgets, uh, in our in our book. bold case over the next uh uh five years, the company is able to drive adoption, model capabilities continue improving, which opens up new markets within those two sort of respective uh uh uh budgets uh that I mentioned
49:03
Unidentified speaker (UNKNOWN): uh
49:03
Unidentified speaker (speaker_2): and then they're able to capture that value.
49:06
Unidentified speaker (UNKNOWN): Um
49:07
Unidentified speaker (UNKNOWN): I
49:07
Unidentified speaker (speaker_2): think I think our our I think our estimate is like north 15% of the of the of the sort of budget spend would be dedicated uh to anthropic. In this sort of bull case, they also end up
49:20
Unidentified speaker (speaker_2): capturing dollars from other AI labs. So Google and OpenAI in particular. So enterprise becomes anthropic space, not too dissimilar from your example about search earlier, where Google was the only search provider that mattered.
49:34
Unidentified speaker (UNKNOWN): Um
49:35
Unidentified speaker (speaker_2): in our bear bare case, you know, that's more of like AI just. takes far longer
49:41
Unidentified speaker (speaker_2): to diffuse.
49:43
Unidentified speaker (speaker_2): Companies are much more hesitant or reluctant to actually build out full sort of like you know agentic workflows. And there's things that support that, right? Change management, process redesign, I
49:57
Unidentified speaker (UNKNOWN): mean,
49:57
Unidentified speaker (speaker_2): vested stakeholder interests, all of them could like sort
50:01
Unidentified speaker (UNKNOWN): of live.
50:02
Unidentified speaker (speaker_2): limit the growth of
50:05
Unidentified speaker (speaker_2): AI. And that, and we're trying to capture that in our in our bear case.
50:09
Unidentified speaker (UNKNOWN): In
50:10
Unidentified speaker (speaker_2): terms of just like revenue forecasts, our base case looks at uh you know a five-year forecast of like 47%
50:17
Unidentified speaker (UNKNOWN): um
50:17
Unidentified speaker (speaker_2): uh compound annual growth rate. Our bear case, that number is north of 60%. And in our bare bear case, uh it's it's it's it's around uh 22%.
50:27
Unidentified speaker (speaker_2): So that just kind of tells you like where we think um
50:30
Unidentified speaker (UNKNOWN): uh
50:30
Unidentified speaker (speaker_2): some of those scenarios could lead.
50:33
Dave Sekera: So either way, I think investors need to be prepared that the stock price over the next couple of years can swing around quite dramatically, depending on how those quarterly results come in. Especially for a company, when you think about the way we look at valuation, you know, the value of a company is the present value of all the long-term free cash flow that it's going to generate over the course of its lifetime.
50:53
Dave Sekera: And in this company's case, it's not even going to generate free cash flow for the next couple years. So you're really hugely dependent on the valuation of this company being based in the out years as far as driving the valuation today. One last question here about the valuation. Now, of course, they're only going to IPO a small percentage of the company. Do we know when the
51:12
Dave Sekera: lockup periods expire? You know how many other shares or what percentage of the stock is going to become available? You know, and how quickly do we have a large percentage of the company that might become available in the next couple of months, next year or so? I mean, could that be a headwind towards the stock performance?
51:29
Unidentified speaker (UNKNOWN): For
51:29
Unidentified speaker (speaker_2): sure. So uh the way Anthropic has kind of uh uh has has has determined its lockup periods, period, they've staggered them. So there is there's a certain amount of shares that are gonna be uh they're gonna be available. to sell um from day one to day 76 post-IPO. Then there's a tranche that gets available from from day 76 and basically after uh uh
51:49
Unidentified speaker (speaker_2): the close of
51:51
Unidentified speaker (speaker_2): fiscal 26, which is gonna be, let's call it like early, early January, late February.
51:57
Unidentified speaker (UNKNOWN): Uh
51:57
Unidentified speaker (speaker_2): and then finally you have uh uh um a tranche that gets released um
52:04
Unidentified speaker (speaker_2): after that after that period, um basically leading up to a full year uh for the IPO. So in terms of like like what that means uh uh in terms of like selling pressure on the stop or like just like market pressure, like more more shares being available,
52:21
Unidentified speaker (UNKNOWN): I
52:21
Unidentified speaker (speaker_2): think that's certainly that's certainly something that investors uh should keep in mind when thinking about when thinking about anthropic and thinking about especially the IPO and and as more shares become available for for employees and founders and other investors, such as Amazon and Google, and whether they decide to sell some of their stakes. Um
52:40
Unidentified speaker (UNKNOWN): at
52:41
Unidentified speaker (speaker_2): the same time though, I I I would
52:44
Unidentified speaker (speaker_2): urge investors to develop like have like a longer-term longer term view on whether this business is. structurally fundamentally uh uh uh going to going to get market share you know the three questions i brought up earlier on adoption capabilities and value capture
53:01
Unidentified speaker (UNKNOWN): i
53:01
Unidentified speaker (speaker_2): think if you have like if you have a view on those on those uh questions and you're positive uh then some of the near-term fluctuations over the next year or so are not gonna going to matter uh that much and if you're not then perhaps like you know uh uh that's another reason uh to not own anthropic shares.
53:19
Unidentified speaker (UNKNOWN): All right.
53:20
Dave Sekera: Well, unfortunately, I think we're uh we're running out of time here, you know, very quickly. But I also want to make sure the uh the viewers of the podcast realize that there are several different articles uh written by Ahmed on Morningstar.com, you know, walking through a lot of the discussion that we've had today, as well as some other information, provides, you know, some more details as far as what our first forecasts are. But
53:40
Dave Sekera: unfortunately, Amit, thank you very much. I really appreciate the time that you spent with us here today. And that's going to be it for today's bonus episode of The Morning Filter. If you have ideas for future bonus episodes, please send them to themorningfilter at morningstar.com. And we hope to catch you every Monday morning for the Morning Filter Podcast. Drops at 9 a.m. Eastern,
54:00
Dave Sekera: 8 a.m. Central. Every week, Susan Debinski Dobinsky and I will review what investors should be watching for the week ahead, new research from Morningstar you shouldn't miss. And then we always wrap up with some stock picks, pans, or swap ideas based on our fundamental valuations. In the meantime, please like this episode and subscribe. Have a great week.
54:19
Unidentified speaker (speaker_0): This recording is for informational purposes only and should not be considered investment advice. Opinions expressed are as of the date of recording and are subject to change without notice. The views and opinions of guests on this program are not necessarily those of Morningstar Inc. and its affiliates, which together we refer to as Morningstar. Morningstar is not
54:39
Unidentified speaker (speaker_0): affiliated with guests or their business affiliates unless otherwise stated. Morningstar does not guarantee the accuracy or the completeness of the data presented herein. This recording is for informational purposes only. and the information, data, analysis, or opinion it includes or their use should not be considered investment or tax advice, and
54:59
Unidentified speaker (speaker_0): therefore is not an offer to buy or sell a security. Morningstar shall not be responsible for any trading decisions, damages, or other losses resulting from or related to the information, data, analyses, or opinions or their use. Past performance is not a guarantee of future results. All investments are subject
55:19
Unidentified speaker (speaker_0): to investment risk, including possible loss of principle. Individuals should seriously consider if an investment is suitable for them by referencing their own financial position, investment objectives, and risk profile before making any investment decision. Please consult a tax and/or financial professional for advice specific to your.
55:39
Unidentified speaker (speaker_0): individual circumstances. Morningstar Investment Management LLC is a registered investment advisor and subsidiary of Morningstar Inc.
55:47
Unidentified speaker (UNKNOWN): The
55:47
Unidentified speaker (speaker_0): Morningstar name and logo are registered marks of Morningstar Inc.
55:53
Unidentified speaker (speaker_3): The game's changed. Capital markets are evolving faster than ever. What's your next move? Morningstar helps clear the way for investors with capabilities. in data, credit ratings, indexes, and investments. Helping bring clarity across public and private markets with data built for the tools and AI platforms you
56:13
Unidentified speaker (speaker_3): use every day so you can move forward with conviction.
56:18
Unidentified speaker (speaker_3): Morningstar. Nothing in your way.
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.

More from The Morning Filter

The Morning Filter

5 Stocks to Buy After Earnings

Episode discussion: Dave Sacara discusses market volatility, NVIDIA valuation cuts, and buy recommendations for Alphabet and Crown Castle

Speakers’ claims: Dave Sacara outlines a volatile market environment driven by uncertainty around AI leadership and bond yields, noting that the 10-year Treasury yield near 5.12% may pressure equities if it rises further. He highlights Morningstar's fair value cut for NVIDIA to $250 due to increased cost of equity and leadership changes, while raising Alphabet's fair value to $450 despite missed search growth targets. Sacara identifies Crown Castle, CNH Industrial, Energy Transfer, and Pentair as attractive investment opportunities based on their moats, strategic pivots, and discounts to intrinsic value. He also notes that CoreWeave and Nebius Group face headwinds due to delayed profitability and equity dilution risks. The episode concludes with an analysis of defense stocks and consumer discretionary names, emphasizing the importance of earnings guidance and margin trends.

Market VolatilityNVIDIAAlphabet
42 min2 assets20 claims to verify
The Morning Filter

4 Growth Stocks to Buy While They’re Still Undervalued

Episode discussion: Speakers analyze AI selloff drivers, Fed rate expectations, and stock valuations across semiconductors, tech, and industrial sectors.

Speakers’ claims: Sarah Hansen and Dave Sekera discuss market volatility following midterm elections, attributing a growth stock selloff to political fears of new data center regulations and mechanical margin calls. They highlight an 80% probability of a December Fed rate hike to 4-4.25% based on inflation data. Sekera provides updates on specific equities, recommending Applied Materials and Healthpeak Properties while maintaining caution on Apple and Arista Networks due to valuation concerns. The hosts also address credit risks for neocloud firms like CoreWeave and analyze long-term AI infrastructure demand, noting that regulatory hurdles may increase costs but not eliminate demand.

AI InfrastructureFederal ReserveMarket Volatility
47 min2 assets19 claims to verify
All insights