Pitch The PM

Sean Emory analyzes Anthropic's $1.4T valuation, compute costs, and Zoom's indirect AI stake

1h 8m Listen to the original Episode page
AI InfrastructureValuationAnthropicZoomCompute CostsEnterprise SoftwareCloud ComputingInvestment Strategy

Summary

Sean Emory of Avory & Co. discusses Anthropic's path to a $1.4 trillion IPO valuation, calculating that it would need $350 billion in revenue to justify the price at a 20x free cash flow multiple. He critiques the prospectus's $40 trillion TAM as unrealistic, proposing a top-down approach based on S&P 500 operating expenses. Emory highlights risks including high compute costs, reliance on hyperscalers for distribution, and the potential for margin erosion as Anthropic competes for end-user relationships. The conversation also covers Zoom's significant indirect exposure to Anthropic through an early investment, which Emory views as a low-cost entry into the AI theme.

Episode sections

  1. 0:00 – 4:25

    Zoom's Anthropic Stake and Valuation

    Doug Garber and Sean Emory discuss Zoom's significant indirect exposure to Anthropic through an early investment. Emory notes that Zoom's stake, valued at $53 million, could be worth $5-7 billion at IPO, representing a large portion of Zoom's market cap. They analyze Zoom's valuation excluding this stake and its cash, highlighting its cheap free cash flow multiple.

  2. 4:25 – 13:04

    Zoom's Strategic Position in AI

    Emory argues that Zoom is well-positioned in the AI era due to the 'context' it generates through meetings and communications. He suggests that while Zoom may not directly compete with AI models for video conferencing, its data and governance capabilities make it a valuable asset for AI ingestion and potential acquisition.

  3. 13:04 – 17:12

    Anthropic IPO Valuation and Revenue Requirements

    The speakers break down the $1.4 trillion valuation of Anthropic's IPO. Emory calculates that to justify this valuation at a 20x free cash flow multiple, Anthropic needs $70 billion in FCF, implying $350 billion in revenue at 20% margins, compared to an $81 billion October run rate.

  4. 17:12 – 24:03

    Top-Down Market Sizing for Anthropic

    Emory critiques the $40 trillion TAM figure in the prospectus as unrealistic. He proposes a top-down approach focusing on S&P 500 operating expenses, estimating that if AI replaces 50% of these expenses ($875 billion), Anthropic would need to capture 50% of that market to reach the required revenue levels.

  5. 24:03 – 26:15

    Revenue Composition and Frontier Risk

    Emory highlights that 80% of Anthropic's revenue comes from models created in the last six months. This indicates high growth but also vulnerability, as the company must remain at the technological frontier to maintain margins, as users eventually trade down to cheaper, older models.

  6. 26:15 – 28:13

    Compute Costs and Capital Expenditure

    The discussion turns to Anthropic's profitability and compute costs. Emory suggests that current low costs may be due to discounted rates from investors like Amazon. He notes a $900 billion capital expenditure commitment for compute infrastructure over the next decade, which will likely normalize costs and impact margins.

  7. 28:13 – 28:56

    Compute Spend, RPO, and Vertical Integration

    Doug Garber questions the cancelability of Anthropic's massive compute liabilities and notes the company's shift toward vertical integration. Speaker 1 estimates $451 billion in compute spend coming online by 2030, with $30-40 billion already spent, and highlights a mismatch between contracted revenue and these long-term obligations.

  8. 28:56 – 31:33

    Hyperscaler Margins and AI as 'Token Factories'

    Speaker 1 argues that while Anthropic's compute spend is high, it is efficient because it produces intelligence. He contrasts this with hyperscalers like AWS and Google Cloud, which generate roughly $500 billion in combined revenue with high margins, suggesting that infrastructure investments hold significant value regardless of who produces the frontier models.

  9. 31:33 – 35:35

    Labor Productivity vs. AI Replacement

    Doug Garber challenges the view that AI will replace half of labor costs, suggesting it currently adds 5-10% to salary costs for efficiency. Speaker 1 expresses fear that in a recession, AI will prevent the typical cycle of rehiring, leading to permanent job reductions rather than temporary layoffs.

  10. 35:35 – 39:07

    Macroeconomic Growth and AI Exportability

    Speaker 1 disputes the productivity thesis, arguing that economic growth requires more dollars in circulation via population growth or debt, neither of which is currently strong. He questions whether AI intelligence can be exported like physical goods, noting that other nations may build their own internal intelligence ecosystems.

  11. 39:07 – 43:35

    Investing in the AI Stack: Orchestrators vs. Model Sellers

    Doug Garber asks where investors should position themselves in the AI stack. Speaker 1 argues that winners will control the end-user relationship and infrastructure, citing Amazon, Google, and Apple as examples. He believes Anthropic risks erosion if it remains a model seller to orchestrators rather than controlling the customer relationship.

  12. 43:35 – 48:07

    Current AI Harness Usage and Tools

    Speaker 1 details their use of various AI tools, including Perplexity, Claude Opus 5.5, and OpenAI's Fable. He emphasizes the importance of an orchestration layer (harness) to govern security, spend, and model selection, mentioning Block's 'Buzz' and Ramp as emerging harness platforms.

  13. 48:07 – 50:39

    Platform Companies as Future Orchestrators

    Speaker 1 predicts every software company will become a harness company, similar to ServiceNow for IT. He highlights AlphaCensus as a vertically integrated example and notes his firm's recent investment in a platform company positioned to become a major orchestrator, controlling governance and security for enterprise AI usage.

  14. 50:39 – 1:07:39

    Anthropic's Margin Guidance and S&M Concerns

    Doug Garber asks about the path to $70 billion in free cash flow and the 35% long-term margin guidance. Speaker 1 criticizes the 7-10% sales and marketing assumption in the guidance, arguing that Anthropic will need to spend significantly more on sales to compete with hyperscalers, making the 35% margin target potentially unrealistic.

Assets discussed

AssetKindStanceContext
AmazonAMZNequitybullishso it's one of our largest positions. If you take that, you have Amazon as well that has roughly $150 billion stake.
AlphabetGOOGLequitybullishLook at the companies that are really winning here, the hyperscalers, right? Who are building this. I mean, it's Google, Amazon.
MicrosoftMSFTequitymixedyou know, Microsoft, Anthropic. These are US companies. These aren't European companies. They don't really innovate that much.
AppleAAPLequitybullishYou think of like Apple, um, same thing, right? They control the experience, they control the devices and the iOS and the systems, metal meta with the customer relationships, they own the infrastructure as well.
Meta PlatformsMETAequitybullishYou think of like Apple, um, same thing, right? They control the experience, they control the devices and the iOS and the systems, metal meta with the customer relationships, they own the infrastructure as well.

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

Claims worth verifying

Zoom made a $53 million investment in Anthropic at a $15 billion valuation.

Unidentified speaker (speaker_1) ·6:07

How to verify: Zoom 10-K or 10-Q filing; Zoom investor relations page for investment disclosures. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Zoom's Anthropic stake could be worth $5-7 billion at IPO.

Unidentified speaker (speaker_1) ·3:26

How to verify: Calculation based on Zoom's stake percentage and Anthropic's IPO valuation; Zoom 10-K/10-Q. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Zoom has $8 billion in cash on its balance sheet.

Unidentified speaker (speaker_1) ·3:46

How to verify: Zoom 10-K or 10-Q filing; Zoom investor relations page. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Amazon has a roughly $150 billion stake in Anthropic.

Unidentified speaker (speaker_1) ·3:46

How to verify: Amazon 10-K or 10-Q filing; Amazon investor relations page. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic's October run rate is $81 billion.

Doug Garber ·0:00

How to verify: Anthropic S-1 filing; Anthropic investor relations page. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic's October run rate is $82 billion according to ticker trends.

Doug Garber ·0:00

How to verify: Third-party financial data providers (e.g., Bloomberg, FactSet) or Anthropic S-1. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Anthropic needs $70 billion in free cash flow to justify a $1.4 trillion valuation at 20x FCF.

Unidentified speaker (speaker_1) ·15:15

How to verify: Calculation: $1.4 trillion / 20 = $70 billion. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic needs $350 billion in revenue to generate $70 billion in FCF at 20% margins.

Unidentified speaker (speaker_1) ·15:51

How to verify: Calculation: $70 billion / 0.20 = $350 billion. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

80% of Anthropic's revenue comes from models created in the last six months.

Unidentified speaker (speaker_1) ·24:03

How to verify: Anthropic S-1 filing; Anthropic investor relations page. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic has $900 billion in capital expenditures towards compute and infrastructure over the next decade.

Unidentified speaker (speaker_1) ·27:40

How to verify: Anthropic S-1 filing; Anthropic investor relations page. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Zoom holds an early stake in Anthropic valued at $4-8 billion, against a Zoom market cap approaching $30 billion.

Unidentified speaker (UNKNOWN) ·28:13

How to verify: Check Zoom's SEC filings (e.g., 13F or 10-K) for holdings in Anthropic and historical market cap data for Zoom around the time of the stake acquisition. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic is expected to bring $451 billion of compute infrastructure online between now and 2030.

Unidentified speaker (speaker_1) ·28:35

How to verify: Verify against Anthropic's S-1 filing or official investor presentations detailing capital expenditure plans and infrastructure build-out timelines. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic has spent roughly $30-40 billion in total infrastructure compute spend from its start to the present.

Unidentified speaker (speaker_1) ·28:51

How to verify: Check Anthropic's S-1 filing for 'Risk Factors' or 'Management’s Discussion and Analysis' sections detailing historical capital expenditures and operating expenses related to compute. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic has a non-cancelable compute obligation with Broadcom through 2030.

Unidentified speaker (speaker_1) ·29:16

How to verify: Review Broadcom's and Anthropic's S-1 filings for 'Related Party Transactions' or 'Commitments and Contingencies' sections detailing long-term supply agreements. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic has $14-19 billion in Remaining Performance Obligations (RPO).

Unidentified speaker (speaker_1) ·29:36

How to verify: Check Anthropic's S-1 filing for the 'Contractual Obligations' table or notes to financial statements detailing RPO. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Hyperscalers AWS, Google Cloud, and Oracle have combined run-rate revenues of roughly $500 billion.

Unidentified speaker (speaker_1) ·30:16

How to verify: Sum the most recent annual revenues from AWS (Amazon), Google Cloud (Alphabet), and Oracle Cloud (Oracle) from their respective SEC filings (10-K). For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic's long-term margin guidance is 35%, with 7-10% allocated to sales and marketing.

Unidentified speaker (UNKNOWN) ·52:35

How to verify: Check Anthropic's S-1 filing for 'Management’s Discussion and Analysis' or 'Forward-Looking Statements' where long-term margin targets and S&M expense ratios are discussed. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic has $200 billion of cash.

Unidentified speaker (UNKNOWN) ·53:39

How to verify: Check Anthropic's S-1 filing for the 'Balance Sheet' or 'Liquidity and Capital Resources' section to verify cash and cash equivalents. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic's revenue per employee continues to increase.

Unidentified speaker (speaker_1) ·51:55

How to verify: Calculate revenue per employee using total revenue and employee count from Anthropic's S-1 filing or recent press releases. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic's sales are currently mostly through hyperscalers or product-led growth.

Unidentified speaker (speaker_1) ·54:12

How to verify: Review Anthropic's S-1 filing for 'Business' or 'Sales and Marketing' sections describing distribution channels and customer acquisition strategies. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Anthropic's IPO prices at up to $1.4 trillion valuation.

Doug Garber ·55:05

How to verify: SEC F-1 Registration Statement for Anthropic AI, Inc. filed with EDGAR. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

$350 billion in revenue is required to justify a $1.4 trillion valuation.

Unidentified speaker (speaker_1) ·55:07

How to verify: Calculation: $1.4T / $350B = 4x Price-to-Sales ratio. Verify against comparable SaaS/Cloud multiples in market data. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Doubling the investment requires reaching a $2.8 trillion valuation.

Unidentified speaker (speaker_1) ·55:24

How to verify: Speaker statement; verify logic against investor return requirements discussed in the podcast. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

$704 billion in revenue is needed to justify a $2.8 trillion valuation.

Unidentified speaker (speaker_1) ·55:24

How to verify: Calculation: $2.8T / $704B = ~4x Price-to-Sales ratio. Verify against comparable SaaS/Cloud multiples. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Combined cloud providers (AWS, Google, Microsoft) generate $500 billion in revenue today.

Unidentified speaker (speaker_1) ·55:44

How to verify: Sum of segment revenues from Amazon (AWS), Alphabet (Google Cloud), and Microsoft (Cloud + AI) in their latest 10-K/10-Q filings. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Do your own research

  • Review Zoom's latest 10-K or 10-Q filings to verify the exact value and percentage of its Anthropic stake relative to total assets and market cap.
  • Analyze Anthropic's S-1 prospectus for the detailed breakdown of the $40 trillion TAM assumption and the methodology behind the $81 billion October run rate.
  • Check Anthropic's disclosed Remaining Performance Obligations (RPO) and compare them against the $900 billion compute infrastructure commitments to assess cash flow coverage.
  • Verify the operating margins of AWS, Google Cloud, and Microsoft Azure from their respective earnings reports to validate the 25-30% margin benchmark.
  • Investigate Anthropic's customer concentration and distribution agreements with hyperscalers to determine the percentage of revenue flowing through third-party channels.
  • Review Zoom's stock-based compensation (SBC) trends over the last four quarters to confirm the reduction from 25% to low teens and its impact on free cash flow.
  • Assess the cancelability clauses in Anthropic's compute contracts with providers like Broadcom and Amazon to distinguish between fixed obligations and flexible liabilities.
  • Compare Anthropic's sales and marketing spend assumptions in the prospectus against actual spend by competitors like OpenAI and Google to evaluate the realism of the 35% long-term margin guidance.
Full transcriptexpand

0:00Doug Garber: The current October run rate is 81 in the prospectus. I'm looking at the ticker trends, data, it's at 82, a little bit above that.

0:08Unidentified speaker (speaker_1): Every software company is going to become a harness company in the world of AI. Context is king.

0:13Unidentified speaker (speaker_2): Hello and welcome to Pitch the PM. My name is Doug Garbert. After a decade at Citadel and Millennium, I interviewed top PMs on their highest conviction ideas. Make sure to stay to the end of the episode to see if I agree with the feature. thesis. And be sure to subscribe to our substack for more research updates.

0:31Doug Garber: Today I'm here with Sean Emery from Avery Co. Sean is one of the people I share ideas with, work up ideas, and I'm actually personally invested in my 401k in the AVRY ETF. Sean has a concentrated high A conviction strategy. He's fully aligned with his money, his partner's money, and his family's money. in this long-only strategy that turns over a third a year, and we've worked up ideas together. It took me a couple of years to get comfortable, but he does deep work, goes through the products, and the most important thing is alignment. He has all of his money in there and he thinks like a true owner.

1:10Unidentified speaker (UNKNOWN): Um,

1:11Doug Garber: and that's why I'm invested with him. They do a lot of data. work. Their thoughts are investing forward. They're thinking about where the world is going, very valuation conscious. Again, he thinks like it is his money and his parent family's money and his partner's money. So they've done it for 10 years, nearly 20% compounded per year in their SMAs. They just launched the Avery ETF. So people like myself can invest um and have access. Um, again, nothing on the show's investment advice. I just wanted to tell everyone my bias that I,

1:44Unidentified speaker (UNKNOWN): you know,

1:44Doug Garber: I'm invested here and I actually work up ideas with Sean. We do calls with CFOs, we go through models, we compare notes, and then we make our own decisions. Um so Sean, it's great to have you back on the show and

1:57Unidentified speaker (UNKNOWN): to

1:58Doug Garber: get an update on some of the episodes we've done over the years. We've done an episode on Zoom, we've done one. on Block XYZ. And today we're here to talk about anthropic ANT, which has backed its way into your portfolio as sort of a free option as a low to mid single digit position through your investments in Zoom. You had exposure there that you we weren't really paying for. But now all of a sudden it's a significant portion. Uh it's kind of hidden value. So why don't we start with

2:30Unidentified speaker (UNKNOWN): what

2:30Doug Garber: this means to the Avery portfolio vis-a-vis Zoom and any other positions that actually own Anthropic and the uplift to those positions.

2:39Unidentified speaker (speaker_1): Yeah. What's going on, Doug? You know, thanks for having me uh here. Um and the kind words. Obviously,

2:46Unidentified speaker (UNKNOWN): you know,

2:46Unidentified speaker (speaker_1): again, we've been we've known each other for quite some time here. Um, you know, anthropic, we're finally here. you know, that time where, you know, the public markets are finally getting uh their first touch and access point to, you know, true kind of model layer AI. You know, everything that is kind of been in the public markets have been mostly around software as opposed to, you know, direct exposure to you know some of these foundational companies. So Anthropics going public, you know, as it stands in our portfolio, like you mentioned with Zoom, there is some path pass through here. You know, Zoom made an investment a couple of years ago, 53 million. You know, that, you know, depending on, you know, this first stick of trading, what it looks like, you know, that could be worth somewhere around, you know, five, six, seven billion dollars uh on the balance sheet for Zoom, which equates to, you know, somewhere around uh, you know, 40% of their underlying market cap today, and plus another $8 billion of cash on that balance sheet. And so it's one of our largest positions. If you take that, you have Amazon as well that has roughly $150 billion stake. You have Blackstone, which has some indirect investments through their funds. And you you mush that all together, you you know, take your you portfolio weights and you're talking about, you know, single digit

4:06uh you know exposure in our ETF or in our strategies in anthropic alone. So you know that's kind of the pass through uh and how we you know how we see it. But you know we've been calling this out for years that you know Zoom has this uh you know pretty meaningful stake relative to the size of their um the business.

4:25Unidentified speaker (UNKNOWN): Yeah

4:25Doug Garber: I remember when we did the first Zoom episode nearly two years ago. I think, you know, we started to talk about this name in the 60s. We finally did an episode at 70 or 80. You know, it was never a high flyer software name, kind of a value name. And,

4:38Unidentified speaker (UNKNOWN): you know,

4:38Doug Garber: it was barely a line item in the sum of the parts at that point, right? It was um maybe five percent. Is there a path to monetize that? Do they have any plans to sell it? Have you asked management what their thoughts are on? harvesting? Because you don't really get credit for those kind of funny sum of the parts until you act on it.

4:58Unidentified speaker (speaker_1): Yeah. Obviously, there's two phases here, right? There's,

5:01Unidentified speaker (UNKNOWN): you know,

5:02Unidentified speaker (speaker_1): I think people were coming around to this idea about a year ago, uh, or even like six months ago, really. Um, and you know, as Anthropic became more and more the dominant player uh at the model layer, you know, really surpassing challenges. Chat GPT or OpenAI uh across different, you know, vectors, but you know, most importantly is I think the the revenue vector. And, you know, people are coming um trying to understand, you know, how do you get exposure? And obviously Zoom and some others in the in the markets have some pass through as well. Um, yeah. So step one is, you know, do you have exposure, right? And and the market will give you some credit for that, but you know, until it's realized, you know, it's it's it's somewhat meaningless. Um, as we got closer to this anthropic IPO, you know, that's step one, right? Which is monetize it at some sort of value. And step two clearly is, you know, what is the lockup? That is not public information, so it's hard to um you know to discern exactly what that lockup period is. There's lockups that they explain in the in the prospectus, but uh you don't know which tranche they're exactly in. They were early investors. Again, they made a $53 million valuation at a $15 billion

6:07or $53 million investment at a $15 billion valuation. So that just shows you how far back um they made that investment. You know, on the lockup, uh again

6:18Doug Garber: X Rand. Yeah.

6:19Unidentified speaker (UNKNOWN): This is

6:20Doug Garber: a hundred X tax wow in just a few years. Yeah,

6:23Unidentified speaker (speaker_1): it's quite the investment for sure. Um that's

6:25Doug Garber: pre-tax, after tax, obviously. they'll have to pay some capital gains.

6:28Unidentified speaker (speaker_1): Sure. Yeah. As they just had Zoomtopia, which is kind of their main annual product event. Uh, but then they they share, you know, an hour, hour and a half with uh investors. You know, we participated and and you know quickly or you know, the the main question is like, what are you going to do here? They kind of preempted the, you know, that question anyways without anyone asking and said, look, we're gonna going to see what happens when it monetizes. at what price, and you know, some of the dynamics there. And then we'll it will get back to you guys with what we're planning on doing.

6:57Unidentified speaker (UNKNOWN): I

6:57Unidentified speaker (speaker_1): will caveat it. Eric Juan at one point said, We'll have a lot of cash. And we think, you know, there's going to be some disruption in the AI space at some point. You already are seeing some of that today. And they will get, you know, they they they referenced reference the you know early 2000s as a as a reference point of what. where some of the best investments and some of the gobbling of the you know the companies that uh needed capital and those investments at that point in time were really, really strategic to you know the future software and technology companies uh thereafter and it it's almost as if he's setting up the idea that he's gonna hold this on the balance sheet until So

7:33Unidentified speaker (UNKNOWN): he

7:33Unidentified speaker (speaker_1): makes some pretty meaningful uh decisions around that to uh you know amplify their business.

7:40Doug Garber: Their company already has eight billion of net cash on the balance sheet.

7:45Unidentified speaker (UNKNOWN): Today's episode is brought to you by Fiscal AI, the modern data infrastructure powering lead-in investors and platforms. Fiscal AI delivers traceable financial data within minutes of earnings releases so we can move fast and with confidence.

7:59Doug Garber: From financials

8:00Unidentified speaker (UNKNOWN): to transcripts to IR decks to even shareholders. shareholder letters, every data point is structured and

8:05Doug Garber: ready to

8:06Unidentified speaker (UNKNOWN): use. Build your investment process on

8:09Unidentified speaker (speaker_3): institutional grade data that is 100% traceable to the source with fiscal AI.

8:15Doug Garber: Where does this leave the valuation if you were to strip out the cash and

8:21Unidentified speaker (UNKNOWN): the

8:21Doug Garber: anthropic stake, assuming it was sold? What is the implied

8:25Unidentified speaker (UNKNOWN): price

8:26Doug Garber: to cash flow?

8:27Unidentified speaker (speaker_1): Uh yeah. So you take a $2 billion free cash flow business, you know, rounding that up slightly. Uh you take a company that um has roughly $8 billion of cash, again, rounding that up slightly. The unknown is the anthropic uh value, given that it's going to trade, it's gonna, you know, be at some level when they're able to actually sell it. You know, is that $4 four billion? Is that eight billion? Uh let's say six billion. at the midpoint. So you take, you know, eight plus six, you get fourteen billion. You have a company's where you're the market cap, depending on the day, is you know, somewhere in those low to mid 20s. Um, and you know, two billion on that is you know, somewhere around, you know, five times free cash flow. Um, and so that's if you X out cash, if you X out uh additionally the uh anthropic stake, you you get a zoom bill. business that has actually accelerated growth here over the last you know five quarters. They just guided you know futuristically to you know essentially double-digit growth without including some of their emerging products. We'll see. But either way, it's a cheap business, you know, well capitalized. Um, some say over capitalized given how much cash they have on the balance sheet. Um, and so the you know that's kind of how we you know square those. two.

9:41Unidentified speaker (UNKNOWN): So

9:41Doug Garber: the thesis has played out from two years ago when they were called a mid-single digit grower. Now they're guided into double digit growth.

9:48Unidentified speaker (UNKNOWN): Um,

9:49Doug Garber: by what time frame?

9:50Unidentified speaker (UNKNOWN): Uh

9:51Unidentified speaker (speaker_1): I think this was twenty thirty, uh, was the end point of this, which they got up to around eight 8%, percent, 8.5%, eight and a half percent, I think was the midpoint of that guide growth. And then again, it that included just their core products today. And so without including any of their emerging products, you know, I think the street is now putting bogeys at that, at that, um, you know, that 10% range is kind of like the buy side bogey here.

10:17Doug Garber: And with the SaaS apocalypse, like the thought is that a lot of software is going to be disrupted. I feel like

10:24Unidentified speaker (UNKNOWN): the

10:24Doug Garber: original thesis was that with them was you know, Microsoft. Teams, Google, they're competing with them. It's really tough. They're gonna be, you know, that's the hard part. Now you have the thought anthropic can come in, but this seems too complex for anthropic to really compete with making a video thing from uh a prompt.

10:41Unidentified speaker (UNKNOWN): I mean,

10:41Unidentified speaker (speaker_1): look, they do four or five billion dollars of revenue, right? So you're looking at anthropic at a trillion dollar valuation. I'm not exactly sure, you know, going after five billion dollars of revenue. is the you know the obvious low-hanging fruit to go after. Um, additionally, you know, there's you know, you have to change behavior. You know, people have their meetings already established built into their workflows. Um, I would say though, I think strategically in the world of AI, you know, context is king, right? All this stuff that we're talking about right now is gonna get digested by somebody's AI. Um, and you know, someone's gonna read the summary and not listen to the recording, someone's gonna, you know, re watch it and then also read the recording or read the summary. And so a lot of engagements happen in this kind of form, which it's you know, either in person communication or virtual communication and all that context. In theory, if you are AI pilled and you think, you know, you're gonna surround your business around all the context that your business creates. use that to your advantage. I think Zoom strategically, everything from their phone systems to their meetings to their events, platforms, all that context is knowledge and information that in theory should be ingested inside of an organization's AI command center, and then use that as you know, kind of the agentic, you know, capabilities to do work

12:01um after the fact. Uh and so we use it a lot in that way, right? You'll have a meeting, it'll, you know, there's a lot of meeting, you know, post-meeting products and summarization products out there. Um, but again, I think a reliable meeting is worth a lot. And then, you know, safeguarding that governance, all the, you know, the important stuff that I think uh you know regulated industries care about. Um so Zoom, I think, is well positioned given the context they have. So yeah, maybe they're you know somebody acquires them for the context capabilities. We've always said that historically, which is cheap asset, really important to enterprises, really important to consumers. And so could there be a time and place? Yeah, I think if the valuation doesn't uh you know get the a meaningful lift between now and you know a couple years from now. Uh if not, yeah,

12:47Doug Garber: that proprietary context could be very valuable, right? That's what uh once you have them, you're kind of hooked in the organization. So they've got that proprietary context. Let's switch gears to anthropic AT. Yeah. The IPO is gonna price biggest IPO

13:04Unidentified speaker (UNKNOWN): on

13:04Doug Garber: on the books, even bigger than SpaceX, 26 billion of private placements Investments concurrently with around a $64 billion. open market deal. We've seen with SpaceX that's doable.

13:19Unidentified speaker (UNKNOWN): You know,

13:20Doug Garber: valuing the company at the high end around $14 trillion. Have you done the work to see what this is worth and how do you value

13:30Unidentified speaker (UNKNOWN): it?

13:31Unidentified speaker (speaker_1): Yeah, um

13:33Unidentified speaker (UNKNOWN): again,

13:33Unidentified speaker (speaker_1): I think uh, and we were talking offline a little bit, which is there's a lot of unknowns, right? Because you know, this is a different workflow.

13:42Unidentified speaker (UNKNOWN): That's our

13:43Doug Garber: job is to invest with unknowns. That's we're always investing with unknowns. It's just getting comfortable with doing it.

13:49Unidentified speaker (speaker_1): Yeah. Uh uh yeah, for sure. You know, clearly there's you know, this is a different way of doing things. Um, you know, it's not just extending the past into the future. It's more, you know, what is the future going to look like and is that gonna going to be fundamentally different than the past? And so a lot of times you'll have, you know, the ability to look at an IPO and it mimics something that has happened in the past. This is just not that, right? We're talking about token factories, data centers that create tokens, right? In theory, uh using energy and then extrapolate, you know, that into knowledge and people consume it. Um the way we tried to triumph. triangulate um how and why, you know, what this company could be worth, right? Is one, you take the, you know, the starting valuation. This is less so, you know, trying to figure out what it's worth, but more so what does it need to do to get to its current valuation. Um, and so you take a $1.4 trillion business and you say, okay, you know, you look at history. and you look at really well, you know, funded companies that are executed well through the duration of time. And what do they end up being worth at some sort of steady state or closer to steady state? And, you know, let's just say at the high end,

15:07you get a you know, high-valued company at, you know, 20 times their free cash flow that they generate, right? Um

15:14Unidentified speaker (UNKNOWN): so

15:15Unidentified speaker (speaker_1): in order to do that, right, we just back into that. You say 20 times, you know, some sort of free cash flow equals 1.4 trillion. That equates to roughly $70 billion of free cash flow needed. Um, so then what is the margin? What is a free cash flow margin to get to $70 billion, i.e., like what revenue times some sort of margin? So again, I'm just going down your best in class, you know, software vendor, uh, even though again, there's a lot to talk about in terms of their business model going forward. But let's just say again, 20% on that. You're basically saying this company needs to generate $350 billion of um

15:53Unidentified speaker (UNKNOWN): free

15:54Unidentified speaker (speaker_1): cash or revenue to generate $70 billion of free cash flow to then get to a $1.4 trillion valuation at 20 times that free cash flow. So yeah, again, you can step back and then ask yourself, okay, what is, you know, are they able to generate that $350 billion? What's the likelihood and probability of that? And so the next step was then

16:14Unidentified speaker (UNKNOWN): Well,

16:14Doug Garber: $350, the current October run rate is $81 in the prospectus. I'm looking at the ticker trends data, it's at 82, a little bit above that, as the current run rate makes sense when you do. you want to have a little good news in your back pocket when you do these IPOs, a little bit of a beat.

16:31Unidentified speaker (UNKNOWN): So

16:31Doug Garber: we have to get from 80 billion today run rate 4x essentially to get to this steady state 350. And then we have to assume there's a nice profitability margin at

16:43Unidentified speaker (UNKNOWN): the

16:43Unidentified speaker (speaker_1): end of that

16:44Unidentified speaker (UNKNOWN): on

16:44Doug Garber: the free cash.

16:45Unidentified speaker (speaker_1): Yeah.

16:46Unidentified speaker (UNKNOWN): Yeah,

16:46Unidentified speaker (speaker_1): yeah. So exactly. And then

16:48Unidentified speaker (UNKNOWN): we

16:48Unidentified speaker (speaker_1): yeah, when you step back, you say that you know what's the probability. of 350 billion? Um, you know, again, it's doing

16:56Unidentified speaker (UNKNOWN): you

16:56Unidentified speaker (speaker_1): know multiples of what they've done to this point. There's obviously competitors out there. Um, but if you take a top-down approach, obviously in their prospectus, they had the forty trillion dollar number. Um, that is like all knowledge work. I think you have to cut that down, right?

17:10Doug Garber: About the TAM number?

17:11Unidentified speaker (speaker_1): Yeah.

17:12Doug Garber: I think TAM should is really like a VC thing. I think in the public markets you gotta kinda, it's about real cash, but there's some crossover. I I I guess uh IR teams kind of forget that uh

17:24Unidentified speaker (UNKNOWN): you

17:24Doug Garber: know now they went from TAM because it was such BS for the last cycle. They're like, all right, let's look at Sam, right? And a lot

17:30Unidentified speaker (UNKNOWN): of people

17:31Doug Garber: I talk to are

17:31Unidentified speaker (UNKNOWN): like,

17:31Doug Garber: yeah, don't even talk to me about TAM. Damn, It's like

17:34Unidentified speaker (UNKNOWN): I

17:34Doug Garber: don't know. I used to make this stuff up uh in IR like as big as you can. It's it's just like

17:39Unidentified speaker (UNKNOWN): it's

17:39Doug Garber: not tangible. It's I call it West Coast math.

17:42Unidentified speaker (speaker_1): I mean 40 trillion is a crazy number. That was kind of my point. Um that includes China, that includes India, you know, like i in theory

17:50Unidentified speaker (UNKNOWN): if you

17:50Unidentified speaker (speaker_1): step

17:51Doug Garber: China, interesting. I don't know how they're getting revenue from there.

17:54Unidentified speaker (speaker_1): Right. That's my point, which is you know,

17:55Unidentified speaker (UNKNOWN): if

17:55Unidentified speaker (speaker_1): AI is is truly super intelligent and kind of um, you know, uh is recursive in nature and I I kind of highly doubt you know we may we may be in a world where AIs are isolated to their territories um you know over time and so you know these TAMs and and SAMs may be just fundamentally different than you know legacy just given the importance of what the what the model's delivering um to two societies. Um that's a whole nother you know question there but

18:26Unidentified speaker (UNKNOWN): in

18:26Unidentified speaker (speaker_1): general the way I looked at it was, okay, let's let's do let's do assume that, you know, AI can replace a lot of um, you know, knowledge work.

18:37Unidentified speaker (UNKNOWN): And

18:38Unidentified speaker (speaker_1): let's just focus on the US. Um, w what we did from a top-down perspective, and I'll and I'll give you a bottoms-up perspective as well, so we can complement the two ends, which is on one side, you say, okay, well, you know, then what is all of the operating expense expenses uh within all the big companies in the S P 500? Essentially, like, what is the expenses that these companies are running at the operating level? Um, those are people doing things, doing tasks, that's GNA, that's sales, that's marketing, that's you know, uh cogs. Um, not all of that can be replaced, but let's just assume AI is as good as you know, some have, you know, characterized it as, you know, 10 years from now. What is that value? That's $3.5 trillion, right? So that is the opportunity to totally replace all operating expenses in the S P 500. Within that, though, then you're saying, okay, let's assume not all of that can be replaced. So you take, you know, let's say 50% of that. That's 800 and um, you know, roughly 875 75 billion is 50% um roughly uh a 50%. So it's two fifty percent. You're saying half of that three point five trillion can be replaced.

19:46Unidentified speaker (UNKNOWN): And

19:46Unidentified speaker (speaker_1): then

19:47Unidentified speaker (UNKNOWN): that's a

19:47Doug Garber: that's a big assumption too. I went, I'm not sure I'd underwrite fifty percent, right? I'd go with five percent, and then how much five, ten percent of an increase, but then

19:55Unidentified speaker (UNKNOWN): I

19:55Unidentified speaker (speaker_1): think it's more. I think it's

19:56Unidentified speaker (UNKNOWN): definitely more than

19:57Unidentified speaker (speaker_1): five and ten, but it it's probably you know, not fifty. Um, but then there you know, it's a competitive field, right? So this isn't going to be winner take everything. Um and you know there's going to be multiple players, I believe, uh that are in this game. And so again, if you say fifty percent of the S P five hundred's operating expenses are replaced by AI. And then it's a duopoly uh you know one player over here, one player over here, that's getting gets to that eight hundred and seventy-five billion dollar number. That's you capturing fifty percent of the market um and replacing fifty percent of all operating expenses. Again, these are just frameworks to try to figure out if that one point four trillion, that three hundred and fifty billion dollars of revenue on twenty percent margins with a multiple of twenty gets you like are you in the ballpark? And then next is like this one's more simple.

20:49Unidentified speaker (UNKNOWN): So

20:50Doug Garber: you're saying they gotta AI has to replace fifty percent of all the S P's operating expenses. And then from there, what share does anthropic need to have of that?

21:02Unidentified speaker (UNKNOWN): So

21:02Unidentified speaker (speaker_1): eight hundred and seventy-five billion dollars is replacing fifty percent of all operating expenses for the S SP P five hundred. And then

21:12Doug Garber: they need three fifty of the eight seventy five. Yeah,

21:14Unidentified speaker (speaker_1): and then their fifty percent uh of that spend, their their fifty percent of the market, let's say, of that replacement, that equates to three eight hundred and seventy five billion. So they would, you know, at three hundred and fifty billion, in theory, if they captured the fifty percent of the fifty percent, um, they could capture eight $875 hundred and seventy five billion dollars of spend, would migrate to them. Yep.

21:39Doug Garber: The question is how much of the economic activity is in those S P 500 companies is number one. And number two, the 50% assumption. Those are the two.

21:47Unidentified speaker (speaker_1): Yeah. These are the

21:48Doug Garber: markets here, they kind of have. They're in that kind of 40-ish percent range.

21:53Unidentified speaker (UNKNOWN): Um,

21:55Doug Garber: but you know, that's a big step from where we are now. What what's the current percent of uh expenses that are AI in these companies? Look,

22:02Unidentified speaker (speaker_1): software is only. two to three percent of spend, right? So the software. So that everything else is like human tasks, like human elements, right? You're paying a sales rep, they're they're using HubSpot, right? They're using Salesforce, and that's two percent of their the of the company's budget, but they're paying the person, you know, 100, 200, 500,000, a million bucks, depending how good they are at selling. Um, and then you have SEO.

22:24Unidentified speaker (UNKNOWN): The

22:24Doug Garber: labor component. You're talking about the li But the labor component.

22:27Unidentified speaker (UNKNOWN): I

22:28Doug Garber: mean, that's not fifty percent of cogs. We're still gonna have people, right? People are providing the context. It's just

22:35Unidentified speaker (speaker_1): for sure. I mean, I

22:36Unidentified speaker (UNKNOWN): I

22:37Unidentified speaker (speaker_1): slex that that ups you know, this top down view is I think you're trying to capture the exponential taking place, right?

22:47Unidentified speaker (UNKNOWN): Um

22:47Unidentified speaker (speaker_1): and so again, I think we're edging on the sort side of uh caution from a from a high end standpoint to figure out if the 350 billion is even remotely in that ballpark. Um another way to look at it, and this goes the other way, which is like, okay.

23:04Unidentified speaker (UNKNOWN): And

23:04Doug Garber: the 350 is somewhere, call it twenty eight, twenty nine run rate from the current consensus brokers out there.

23:12Unidentified speaker (speaker_1): Yeah. I mean some of the um

23:14Unidentified speaker (UNKNOWN): I'm sorry,

23:14Doug Garber: yeah. 28, 29. I mean, that's kind of where I'm looking at these numbers, call it 270, 300 for 28, roughly, 2029, 440. These are all kind of word of mouth brokers. It's funny, the numbers you actually need to underwrite the deal

23:27Unidentified speaker (UNKNOWN): are

23:27Doug Garber: the only numbers not in the prospectus. I don't know how they spend millions of dollars on these S1s. And the company doesn't talk about the numbers,

23:35Unidentified speaker (UNKNOWN): but they

23:36Doug Garber: tell the brokers who then call get on the phone. if you're lucky and can tell you their numbers on the phone. But I somehow someone's gonna have to explain this to me one day.

23:43Unidentified speaker (speaker_1): Yeah, nobody knows where no one can consciously say where they're gonna be in three years from now. And I think have a a a real coherent argument. You can directionally say, like, I believe AI is gonna continue to expand, and the amount of demand and consumption for AI is going to be, you know, continually. continue to grow at a pretty healthy pace over the next, you know, three to five years. Who that accrues to is still up in question. Once that, you know, to to to share and it talks a you know, cum, you know, uh essentially articulates a little bit of the competitive dynamics here, 80% of Anthropics revenue comes from models they've created in the last six months. And so this is kind of like who's at the frontier consumes all the the dollars or generates all the dollars and token use for the most part. And but it also shows the vulnerability of that. So if you're underwriting three years from now, I mean, y you know, if eighty percent of revenue is coming from models within the last six months, you your your your viewpoint has to be that they're gonna They're going stay to at the frontier, you know, indefinitely here. Um, 'cause Um, because that's where the margin, the dollars are all accruing to, specifically for brand new things people are doing. You you you you start at the frontier, you develop and build, and then you you you push that to cheaper, more efficient

25:03models that can then run whatever you built. Um that's uh what we're

25:08Unidentified speaker (UNKNOWN): seeing in practice today.

25:09Doug Garber: Concept of,

25:10Unidentified speaker (UNKNOWN): you know,

25:11Doug Garber: We've started to see it recently where people are trading down from the frontier to cheaper. models, open weight models to get their costs down. And that's one of the reasons, you know, Enthropics revenue still growing, you know, 40 plus percent quarter over quarter. But

25:25Unidentified speaker (UNKNOWN): you

25:25Doug Garber: know, token consumption's amazing, but the mix has really hurt the revenue. And you know, there's a lot of competition for the open weight, and people trade down. It's like this invisible asymptote argument where you don't need the frontier for everything, right? People are gonna put harnesses in that save their cost once they put in that layer.

25:45Unidentified speaker (speaker_1): Yeah.

25:46Unidentified speaker (UNKNOWN): Um so they're

25:47Doug Garber: underwriting a lot here, but I get it. This this name is

25:51Unidentified speaker (UNKNOWN): Yeah,

25:51Doug Garber: one more it's the future, right? In the markets, everyone you every long only is gonna have to have it, you know, close to an equal weight position. Retail's retailers gonna can be excited about it.

26:00Unidentified speaker (UNKNOWN): Um, you know,

26:02Doug Garber: it's it's certainly trendy. It's but it's priced on if you're saying 350 billion when they're doing an 80 billion run rate.

26:10Unidentified speaker (UNKNOWN): Uh that's

26:11Doug Garber: a few, that's a lot of steps right there to get there. And then what about the profitability?

26:15Unidentified speaker (UNKNOWN): Um

26:16Doug Garber: I'm hearing long term 35%, 15% in kind of the medium term, profitable on an adjusted basis now.

26:25Unidentified speaker (UNKNOWN): Yeah,

26:26Unidentified speaker (speaker_1): it's tough. I mean the the if you historically and the numbers as they stand today are actually I think uh there was an upside um to what people thought was going to come out in the prospectus, right, in the S one. Um, both on mostly on profitability and the efficiency of this model so far. Now there's a lot of nuances there that I think are somewhat um hidden, which is obviously they have investment stakes from you know Amazon and and and some others out there. And ultimately, I I'm not so sure that the the underlying cost of running compute uh is at you know spot or you know normal enterprise agreement rates versus getting heavily discounted for this for the ability. to get access to the value, you know, that round and an investment for someone like Amazon. And so I expect fully that at some point their rate in terms of their compute spend starts to normalize. And I mean that from using cloud providers.

27:32Now, for looking forward, you know, we we also know that they have roughly 900 billion. of capital expenditures towards you know compute and infrastructure on a go forward basis. Again, that's over the life

27:47Unidentified speaker (UNKNOWN): of like a decade.

27:50Doug Garber: That nine that $900 billion liability. I'm looking here at the um the Alpha Sensanthropic dashboard. They have a new you know primer, which is pretty cool to look at here. It kind of breaks it down by company. And then if you're you read the foot in the footnotes, they have it right there. There's liabilities currently, and then there's liabilities that come in the future, right? To get

28:10Unidentified speaker (UNKNOWN): to

28:10Doug Garber: this 900 billion.

28:13Unidentified speaker (UNKNOWN): How

28:13Doug Garber: much of those are cancelable? How much of those can they move around to match their revenue? Because there is a mismatch here between contracted revenue and these liabilities. And then the company's also now starting to say, hey, we're not just going to lease these things. We might build them ourselves and and vertically integrate.

28:35Unidentified speaker (speaker_1): It's a lot. It's um it's roughly over a decade. I think 451 billion of that is expected to come online between now and 2030. And so, you know, we're going from what felt like they've spent, I think, roughly thirty $30 to forty $40 billion in total infrastructure compute spend from the start of the company to today.

28:56Unidentified speaker (UNKNOWN): And

28:57Unidentified speaker (speaker_1): some will say, yes, that's highly efficient. And it is, right? I mean, that's, you know, to generate run rate 80, you know, something billion dollars on 2030, you know, $30 billion of compute spend. Again, I think some of that is artificially uh lowered given the investments that some of the hyperscalers made in them. to give them discounted preferential pricing. But on a go forward basis, you know how much of this is cancelable? Some of it is unknown, but there are we we do know that really those uh the one that they have, I believe, with Broadcom um is between now and 2030. And I'm I'm almost sure that's not cancelable. Uh everything beyond that, again, it's it's somewhat unknown, but either way, a lot of this is still, you know, somewhat of an obligation on roughly, you know, 19 or 14 billion dollars of RPO remaining performance obligation. So, like you mentioned, their forward visibility is rather low. Um, and their expense and capex on a go GoFord forward basis is rather high. Now we know that different from other industries, this CapEx is very valuable because it does, these are token factories, right? And ultimately it does produce some level of intelligence. And so this isn't something that is easily replicated on the

30:16infrastructure side. And we know that companies like the hyperscalers, AWS, US, uh, you know, Google, GCP, Oracle, you know, they have run rate revenues combined, the main hyperscalers. of roughly you know $500 billion, uh, with pretty high margins. So whatever compute and infrastructure they're building, I think is gonna have significant value over time, even if they're not the ones producing, you know, the frontier model. And so,

30:42Unidentified speaker (UNKNOWN): you know,

30:42Unidentified speaker (speaker_1): I think there's a little bit of um,

30:44Unidentified speaker (UNKNOWN): you know,

30:45Unidentified speaker (speaker_1): when you when you take that into account, I think it it you know dilutes the fear, I guess, a little bit on a go go-forward forward basis. basis. But

30:51Unidentified speaker (UNKNOWN): again,

30:52Unidentified speaker (speaker_1): there is a mismatch, like you mentioned, in revenue, uh, bookings, you know, contracted revenue on a go forward basis versus what seems like a pretty hefty bill um, you know, down the line. And this business model is func you know fundamentally changing from what perceived to be highly efficient. Doesn't mean it won't be efficient on a go-forward go forward basis, given we don't know the level of revenue they'll generate, but we just tried to triangle. triangulate like what is the upside revenue they can generate. So, you know, we remember we said uh 875 billion is kind of S P 500 operating expenses, 50% of that, and then they're 50% of that market. Um, that gets to 875 billion

31:31Unidentified speaker (UNKNOWN): with

31:32Unidentified speaker (speaker_1): competitive.

31:33Doug Garber: pushback on AI being half of the operating costs. I don't even think labor's half the cost, and like what would the people do? I think it's like I

31:40Unidentified speaker (UNKNOWN): mean

31:40Doug Garber: that's

31:40Unidentified speaker (UNKNOWN): the person

31:41Unidentified speaker (speaker_1): that's what people are I think

31:42Unidentified speaker (UNKNOWN): are salary,

31:43Doug Garber: but then

31:43Unidentified speaker (UNKNOWN): questioning

31:44Doug Garber: you add, you know, this person spending five, ten percent of their salary on tokens to make them more efficient as opposed to a complete replacement.

31:54Unidentified speaker (UNKNOWN): Um are

31:55Unidentified speaker (speaker_1): you doing more and

31:56Unidentified speaker (UNKNOWN): more work in these AI? You

31:57Doug Garber: need to train the AI. You need to train the AI and give it the context to do your job. And it needs to iterate but

32:03Unidentified speaker (UNKNOWN): um

32:05Doug Garber: it's getting really smart. It's improving.

32:09Unidentified speaker (UNKNOWN): No,

32:09Unidentified speaker (speaker_1): it's improving

32:09Unidentified speaker (UNKNOWN): at a

32:10Doug Garber: very rapid pace of iteration and improvement. And I guess that's

32:14Unidentified speaker (UNKNOWN): you

32:14Doug Garber: know, the labor pools, bear cases, uh anthropics, bull case. So that's the uh the five percent for the uh society fund.

32:24Unidentified speaker (speaker_1): Look, right now it's not impacting it, right? You see jobs. they continue to grow. We just got the ADP, you know, weekly jobs number, you know, that continues to trend positively. My fear on on on jobs isn't right now when things are functioning. It's when one day we'll have some sort of crisis, right? A real crisis where there's real layoffs and it's not just you know isolated to a certain industry or or or something else and it's you know more of a, you know, a recession of sorts, you know, brought you know, more broad based.

32:54Unidentified speaker (UNKNOWN): At

32:54Unidentified speaker (speaker_1): that point in time, that's where I fear the

32:58Unidentified speaker (UNKNOWN): the

32:58Unidentified speaker (speaker_1): the the next cycle of which historically people fire people,

33:03Unidentified speaker (UNKNOWN): uh,

33:03Unidentified speaker (speaker_1): then they move into temp workers, and then they migrate into, you know, full time.

33:08Unidentified speaker (UNKNOWN): And

33:08Unidentified speaker (speaker_1): then the cycle begins and ultimately now you have unemployment going from elevated to, you know, casually going lower and then you know stay stable for some sort of period until you have another event. Um my fear is instead of going from firing to then um temporary to rehiring, is that you find ways to do more with less fundamentally. And we're in 2026 twenty twenty six and we have these persistent agents starting to f you know form use from from meta. You know, you have other, you know, Claude and Duet and Dots, and these are the early days of a persistent agent that has a goal uh that's somewhat recursive, but not, you know, maybe to this to the extent that um, you know, people fear.

33:55Unidentified speaker (UNKNOWN): But

33:55Unidentified speaker (speaker_1): if you look out five years from now, these things are gonna going be to much more intelligent, much more self um serving to themselves. Um and I I fear that net cycle, you just don't get the rehiring uh that is expected.

34:09Unidentified speaker (UNKNOWN): Um

34:09Doug Garber: once a human programs then, like everyone's gonna have a work agent. So you still need the person, maybe of a few. You're talking about a fundamental step up in productivity, right? Where you can essentially grow the economy, companies can could be more do more with less inflation. because they're using tokens instead of labor to it's pushing down the cost of not necessarily of cost, but the output per labor hour for

34:36Unidentified speaker (UNKNOWN): white collar

34:37Doug Garber: jobs because human plus tokens equals more output. So it's the same thing we had in the 2000s where you had,

34:46Unidentified speaker (UNKNOWN): you know,

34:46Doug Garber: we defied the

34:49Unidentified speaker (UNKNOWN): when

34:49Doug Garber: the internet came, we had a step up. in productivity, right? We had this huge growth with very little inflation. And we're like, how is that possible? Usually you have growth with inflation. The end the key variable was productivity. And that seems to be what we're heading towards now is another big step up in productivity, which allows, I guess, more growth with less inflation. Obviously, we've had some inflation here with energy prices. prices and the conflict and some of our policies abroad on the energy side.

35:20Unidentified speaker (UNKNOWN): Um, so

35:21Doug Garber: I see another productivity step change. That's my personal view that but I have

35:28Unidentified speaker (UNKNOWN): time

35:28Doug Garber: will tell, right?

35:28Unidentified speaker (UNKNOWN): Yeah,

35:28Unidentified speaker (speaker_1): I I know it's not a macro show by any stretch, but the

35:31Unidentified speaker (UNKNOWN): I have a

35:32Doug Garber: you have asked to do that, so we we we'll get that going.

35:35Unidentified speaker (speaker_1): I have trouble, I have trouble with the productivity. version uh because it it requires more revenue, right? Uh it it requires more activity, dollar activity to take place. And I fear I fear that um we're like where's like historically, if you look back in you know in the history,

35:57Unidentified speaker (UNKNOWN): the

35:58Unidentified speaker (speaker_1): the economy's economies got bigger through multiple like the economy has to get bigger for that unit. per hour of output to be more productive, right? Like that that output to be more productive. And so historically, you had significant um uh immigration uh in into the US, you had population growth, whether it was uh you know organic population growth, you know, um um from new births or uh immigration to the country, and that added more people. And so you had the ability to have more dollars in circulation, which you know, in theory has you know more economic output doing that. Today we don't have either of those. Number two is more debt, right? And we're we at some point, if you go back to like this the eighties, nineties, and two thousands, and then even post um the the financial crisis, is we continue to add layered and layers of debt. And we're at that. point where m p potentially that is no longer, right? So debt adds more dollars in circulation as well. Um

36:58and then, you know, lastly is uh given the internet, we were able to sell abroad. And so even if we couldn't tap into our own pool of dollars in the US, we could tap into other uh, you know, countries and tap into their GDP to obviously credit create uh you know create our own uh uh input in terms of uh revenue to our companies. And in a world of AI, I do fear whether uh we're gonna be as unless we are the computer, the factory of the world. Um

37:29Unidentified speaker (UNKNOWN): how many

37:29Doug Garber: data centers have been built here? We have the best chips. Look at the companies that are really winning here, the hyperscalers, right? Who are building this. I mean, it's Google, Amazon. you know, Microsoft, Anthropic. These are US companies. These aren't European companies. They don't really innovate that much. Obviously, China's coming up. They're, you know, always trying to clone and moving up the curve quickly. And we've added entire industries. I mean, this is trillions of dollars of capex for industrial, ENC, electricians, plumbers. Then we need more energy. We need more natural gas to fuel that. That pays state taxes. So we've added an entire l industry onto our economy.

38:09Unidentified speaker (UNKNOWN): Or

38:09Doug Garber: you could say it's just a huge expansion of the cloud compute, right? It's just

38:13Unidentified speaker (speaker_1): Yeah, that was cloud,

38:14Unidentified speaker (UNKNOWN): which was

38:15Unidentified speaker (speaker_1): productivity. But

38:16Doug Garber: this is like super cloud, right? This is uh

38:18Unidentified speaker (UNKNOWN): This

38:18Unidentified speaker (speaker_1): is intelligence. So are you going to export your intelligence to another country? I don't know. Um, and I'm not so sure that's

38:24Unidentified speaker (UNKNOWN): what I'm saying. the same thing. We're They're gonna be their

38:25Doug Garber: intelligence, they're gonna be using right? There's gonna be a few other countries that do this themselves internally, but all the other countries

38:33Unidentified speaker (UNKNOWN): um are

38:34Doug Garber: gonna use the U.S. intelligence, right? That's kind of what they're doing with through American companies. We're getting a little off traffic of what topic of what anthropic

38:42Unidentified speaker (UNKNOWN): is.

38:43Doug Garber: It's

38:43Unidentified speaker (UNKNOWN): all relative. So I guess you know, it's

38:44Unidentified speaker (speaker_1): all uh you know

38:45Unidentified speaker (UNKNOWN): relevant

38:45Unidentified speaker (speaker_1): to their future, but yeah.

38:47Doug Garber: Well, this is a macro call. I mean, this is over, this is a 1.4. trillion company, and there's some clones coming right behind it.

38:55Unidentified speaker (speaker_2): Sure.

38:56Unidentified speaker (UNKNOWN): Um, and

38:57Doug Garber: then it's not just this part, then you have the rest of the stack, right? You've already got

39:01Unidentified speaker (UNKNOWN): the

39:01Doug Garber: implementers like Palantir, you've got

39:04Unidentified speaker (UNKNOWN): you know

39:04Doug Garber: the hyperscalers. It's like, where do you want to be?

39:07Unidentified speaker (UNKNOWN): Where

39:08Doug Garber: do you want to be in the stack as an investor? I mean, that's the get who's gonna end up winning the rent. Right now, obviously, the semi-companies. and the chip companies are,

39:17Unidentified speaker (UNKNOWN): you

39:17Doug Garber: know, have got a lot of the rent in terms of cash flow. The hyperscalers are investing.

39:23Unidentified speaker (UNKNOWN): You know,

39:23Doug Garber: they're free cash negative now as a group. People are a little concerned now.

39:28Unidentified speaker (UNKNOWN): And

39:28Doug Garber: then now you have this other layer of the model. So where that's the big question if you're thinking about large institutional money, where do you want to be in the AI? or the fabricator? A lot of people want to be, you

39:40Unidentified speaker (UNKNOWN): know,

39:40Doug Garber: if there's only one of them, but like,

39:42Unidentified speaker (UNKNOWN): where

39:43Doug Garber: do you want to be in the stack?

39:44Unidentified speaker (speaker_1): Yeah. Um, yeah, if you I mean, if you look back in time, you say, okay, in most technology cycles, you have a couple different um characteristics that persisted, you know, regardless of the cycle.

39:55Unidentified speaker (UNKNOWN): You

39:55Unidentified speaker (speaker_1): ultimately want to control you know the end user, you know, relationship strategically, also control that infrastructure to the relationship. You look at Amazon with commerce controlling the experience and the infrastructure and logistics. That aid ate away at FedEx and UPS, right? And some of the players that tried to have microbusinesses associated with that. And because they couldn't

40:16Unidentified speaker (UNKNOWN): compete. They

40:16Unidentified speaker (speaker_1): didn't have the relationship with the customer. And therefore, um, you know, they were at risk, right? And so they had the relationship with the customer, subsidized subsidize some of the logistics built out. You have Google, you know. with search. Uh, they obviously own, you know, their own cloud business, and that's assists them in terms of going end to end from the consumer experience down to the to the infrastructure. And those two things work it congruent with each other. You think of like Apple, um, same thing, right? They control the experience, they control the devices and the iOS and the systems, metal meta with the customer relationships, they own the infrastructure as well. not just the AI infrastructure of now, but even in the past, that gave them a leg up on companies like Snap, which had to rent out compute from Google Cloud. So they couldn't really control their destiny. So really those companies, I think, that have you know distribution, that have data, that have workflow or network um, you know, advantages, you know, capture that influence and ultimately the economics, you know, within that. So again, this is. a different position from selling, you know, interchangeable components to somebody who owns the customer. So again, selling a model to someone else. And so I think ultimately, you know, the ones that are orchestrating, control the infrastructure, orchestrate these models in the world of AI, uh, and then have a

41:35customer relationship end-to-end end to end are going to be gonna the winners. Clearly that starts to pivot towards the hyperscalers, but you're seeing anthropic, we just talked about that going from they obviously have Claude, they have codex and or uh cloud um code, uh, and they have a relationship to some degree with engineers and software developers,

41:54Unidentified speaker (UNKNOWN): but

41:54Unidentified speaker (speaker_1): do they have enough of a relationship with you and I and just their, you know, in their in their Cloud Claude product? Um, and do they control the infrastructure? Amazon Direct, the $900 billion we talked to, is their efforts to go down. And then ultimately, I think eventually they have to go up and get closer and closer to the customer and not just be reselling through the hyperscalers, which I believe 50% of their revenue goes through the hyperscalers, roughly 50% of the revenue. And so in their IPO, they said they control the relationship, but do they? I mean, in some of the AI products we use, you know, we're one click away from going from this model to that model. Um, and we have the relationship with the harness. So I think the orchestration layer is super important. It controls governance, security, it controls ideally um the model layer. I think eventually, and it's already happening, the model itself, what model it is, should be orchestrated by a parent model uh that is. either built or assumed by, you know, whatever entity you're part of. And that ultimately abstracts the models themselves. So if you don't own the relationship and you're just getting orchestrated by a third-party orchestrator,

43:05Unidentified speaker (UNKNOWN): then

43:05Unidentified speaker (speaker_1): ultimately I think your economics are going to erode away.

43:08Unidentified speaker (UNKNOWN): And

43:09Unidentified speaker (speaker_1): so you have to own the infrastructure, you have to own the relationship. It's going gonna to look m very close and similar to, you know, the past.

43:19Unidentified speaker (UNKNOWN): When

43:19Doug Garber: we spoke before, you were vibe coding your own AI usage. Um, sounds like you've advanced now, you're using harnesses. What are you doing for AI? What's your harness system to control your agents?

43:35Unidentified speaker (speaker_1): Yeah. Um

43:37Unidentified speaker (UNKNOWN): we

43:38Unidentified speaker (speaker_1): use um a lot of different things. Um one specifically, you know, we use a lot of perplexity computer.

43:45Unidentified speaker (UNKNOWN): Um,

43:46Unidentified speaker (speaker_1): you know, that is uh a version of our harness,

43:49Unidentified speaker (UNKNOWN): um,

43:50Unidentified speaker (speaker_1): where you have a model orchestrator. It's not cheap. Um, you know, and so we have

43:58Unidentified speaker (UNKNOWN): you know,

43:58Unidentified speaker (speaker_1): again, our data on one side or some of our, you know, uh centralized in one place so that we can move. off a harness, you know, if we needed to. Um, we have our our skills repository on platform, off platform, again, so we can move if we needed to. Um we have persistent agents that you know we're using that um allow us to do things, you know, in in real time and they're getting smarter and smarter. Uh we have open source models we're using or open weight models we're using and you know, some of the frontline stuff. So again. you know, back to anthropic here is you know, Claude, you know, Opus 5.5 is a great model, right? And it's one that it's at the frontier. So if we're building new things, we're using it. Um, along with uh OpenAI's, you know, leading models with Fable. And and so,

44:43Unidentified speaker (UNKNOWN): you

44:43Unidentified speaker (speaker_1): know, having that orchestration layer for a company is important because then you can govern your other users inside your organization. as well. Put capacity limits on them, um, have instructions for what they can and can't do, uh, govern, you know, what files underneath, uh, if we're, you know, depositing those into like a box enterprise. Um, and so those are important aspects that I think matter for entities that aren't just a single person. Um then there's like block

45:16Unidentified speaker (UNKNOWN): you have

45:16Doug Garber: like an IT department control and an AI department control in your harness layer telling you what employees have access to what, how much they can spend, monitoring the return on this project or you know why are you using this model for this? Can you do that? So that's kind of where we're going. Yeah I've heard um you know one of the companies transient And AI, they're kind of for finance specifically. They came out of a lot of the bulge brackets that they're they're um some some large head hedge funds are using them. So I've heard good things like that's people are like, hey, the last few months,

45:45Unidentified speaker (UNKNOWN): token

45:45Doug Garber: max, and let's make sure we're getting a return on the tokens because everyone's like, all right, we want to max productivity. It's like, hold

45:53Unidentified speaker (UNKNOWN): on.

45:54Doug Garber: Let's make sure we're using the right model. So they're kind of finding that right harness. And you know, it comes with forward deployed engineers. And I'm seeing actually, I think jobs are going up because everyone has to implement in license finance. You need to add a whole AI layer to implement this. Otherwise you'll be left behind. Sure. So my one of the things you see me repost on LinkedIn is like,

46:12Unidentified speaker (UNKNOWN): you

46:12Doug Garber: know, more jobs are being created with for AI implementers.

46:16Unidentified speaker (UNKNOWN): No,

46:16Unidentified speaker (speaker_1): that's what I mean we we had that thesis the whole time that we're going to have more software engineers than ever. Um you know they're the ones that part

46:22Unidentified speaker (speaker_3): where

46:23Unidentified speaker (UNKNOWN): I got

46:23Unidentified speaker (speaker_3): it.

46:23Unidentified speaker (speaker_1): Yeah. If you need 'cause need because if you need help, like you have to start looking at it uh in a way a software engineer would look, right? I mean,

46:30Unidentified speaker (UNKNOWN): i

46:30Unidentified speaker (speaker_1): do you think uh you know Microsoft Outlook was down forever? Um I'm pretty sure they had uh you know their top leading software engineers having to figure out what's going on there. Um as opposed to just having some sort of AI model that was, you know, trying to figure it out. Uh I joked

46:44Unidentified speaker (UNKNOWN): around.

46:45Doug Garber: It's a good thing we have Zoom, so we were uh still online, right?

46:48Unidentified speaker (UNKNOWN): Yeah,

46:48Unidentified speaker (speaker_1): exactly. Yeah, so we we use Zoom team chats. Another harness, just uh uh you know, plug uh one of our investments is um you know, block uh came out with Buzz. Um, and so you know they just announced um general availability uh with with um with with managed uh a managed product. Um and that's a harness that um I think has real uh viability uh out there. So if you're looking for something that is interesting, that's model agnostic, that can be, you know, built for your company's, you know, operation buzz um from block is uh pretty attractive. Um

47:25Unidentified speaker (UNKNOWN): you

47:25Unidentified speaker (speaker_1): know, for sure. I'm on the wait list for ramp. Ramp apparently has a harness, and we'll see what happens there. But um, you know, they're creating their own harness. I don't know if it's gonna be going to vertical specific to, you know, payments and finance versus you know something agnostic. So um I mean those are two versions of the

47:40Unidentified speaker (UNKNOWN): thing. I was at

47:40Doug Garber: lunch yesterday with a couple huge family offices and very big investors and they were like a lot of them are saying hey they like alpha census because it's already vertically integrated. You have a good UI, you've got a lot of the data you've got transcripts and they've kind of built it in for you so you don't have to go through this headache of employing an entire team. It's kind of the off off-the-shelf the shelf vertical one for, you know, unless you want to customize it and put in those resources.

48:07Unidentified speaker (speaker_1): Yeah. I'd say that's more like um

48:09Unidentified speaker (UNKNOWN): that's

48:09Unidentified speaker (speaker_1): almost more like a dashboard and and like a depot uh you know a repository for for information as opposed to an orchestrator, uh, you know, you're

48:19Unidentified speaker (UNKNOWN): not

48:20Unidentified speaker (speaker_1): like building new agents in there yet. Um every company, every software company is going to become a harness company and you know, like the services now, the service now, people are using them for ITSN, so IT, you know, service management for CRM, for you know, employee management to all these things. And so they're they they're like they're one that's set up. And again, full disclosure, we just made an investment here recently. Is they, you know, they're we think their position to be a harness. And ultimately, you are just the orchestrator of all a company's um job. functions, and you become the governance security guardrails for anything coming in and anything going out, uh, you know, governing spend, governing usage. Um, and so those companies that have historically been platform companies, I think have the ability to become a really viable player in this space and actually just put pressure on the models themselves. And so

49:15Unidentified speaker (UNKNOWN): you know,

49:16Unidentified speaker (speaker_1): that th those are some of the dynamics or things we're looking at where we think value is going to accrue. If you go back to our let like we wrote a letter like four years ago and we said,

49:24Unidentified speaker (UNKNOWN): we

49:24Unidentified speaker (speaker_1): think the most valuable place of all this stuff long term is going to be the orchestration layer. That's, you know, that term has developed into a harness. That wasn't a term used back in the day, but orchestration. And we still hold that true today. That's where you you remove complexity on the uh at the top end and the infrastructure layer. So the orchestration is trying to remove the complexity. What am I spending? And then how am I using it? And what is the experience? And those two things are the comp like the complexities that you and I have. Like in theory, me and you do not care whether it's running at AWS, Google Cloud, uh, Microsoft Azure. In theory, we do not care if it is Opus 5.5 or Fable or OpenWaite or Open Source. We just care if the intelligence, the output relative to the cost is valuable. And ultimately, that is all you know what everyone actually cares about. And that orchestration layer is going to be the key element that does that for you. Who that accrues to is the battleground.

50:24Doug Garber: And then the other key is. the proprietary context of what you're instructing it to do and making sure it has the right data behind it.

50:33Unidentified speaker (speaker_1): Yeah.

50:33Unidentified speaker (UNKNOWN): Um,

50:33Doug Garber: so it's a lot of pieces you gotta get going to get that productivity.

50:38Unidentified speaker (UNKNOWN): Um last

50:40Doug Garber: question on anthropic here. We're getting to $350 billion of revenue, and I'm using my, you know, my um in sync model here for the IPO, which I've used for a while, doing sensitivities, It's kind of pre-built here on the margins.

50:54Unidentified speaker (UNKNOWN): How

50:54Doug Garber: are you getting to the 70 billion of free cash? What are the margin assumptions? How should we think about the efficiency on the cost per compute? And then also, you know,

51:07Unidentified speaker (UNKNOWN): the

51:07Doug Garber: pricing. Like pricing's not necessarily Really going down, but the mix is hurting the the dollar per token number. numb.

51:13Unidentified speaker (speaker_1): Yeah. Um

51:15Unidentified speaker (UNKNOWN): again,

51:15Unidentified speaker (speaker_1): these are moving parts. Again, because the the the past isn't the future. And so we have not seen a company scale with $900 billion of spend on compute running and training models. Um, now, fundamentally, though, you know, AI labs like anthropic should be unusually productive internally, right? I mean, they have the latest and greatest models. Uh, in theory, they should have. the best use of them, uh, because they've seen them the last, you know, you know, over time. And so fundamentally, on the operating expense side, kind of what we were saying before, they should be very, very efficient. And you saw that in some of the numbers, right? The revenue per employee. Um, you know, that number continues to go up and to the right. So they're very productive on that element. Um now, again, the complication. of of the spend on the compute is is is all about, you know, utilization uh of it. And, you know, are they going to utilize yeah all that spend and maximize that and drive that revenue performance, right? And so if we take what they've looked they they laid out a guide, right? They laid out a guide for everybody. And

52:23but I'm not so sure like we can just take that as is, like we have historically for other companies, just given so many variables that you you know are here.

52:33Doug Garber: About you mean the 35%

52:35Unidentified speaker (UNKNOWN): of

52:35Doug Garber: the long term guide, right? Because you can't guide

52:38Unidentified speaker (UNKNOWN): the

52:38Doug Garber: next 12 months or whatever, 18. So you always put out a long term guide.

52:42Unidentified speaker (speaker_1): Yeah.

52:43Unidentified speaker (UNKNOWN): You

52:43Doug Garber: know, in in a few years beyond the horizon, and that's what

52:46Unidentified speaker (speaker_1): Yeah, like what their stable state economic model is gonna fundamentally look like. That's like their best view of it. And uh

52:52Unidentified speaker (UNKNOWN): one of

52:52Unidentified speaker (speaker_1): those numbers

52:53Doug Garber: to sell the deal as the 35% margin. I'm trying to say what is what is that 35% based on it in that long-term got?

53:01Unidentified speaker (UNKNOWN): I mean,

53:01Unidentified speaker (speaker_1): most of it's

53:01Unidentified speaker (UNKNOWN): the deal's really training

53:03Doug Garber: 27-28.

53:04Unidentified speaker (speaker_1): Yeah, I mean, look, it the the one number I I don't like in there is they have I think it was like seven to ten percent for sales and marketing. Um, and that's inclusive of the take. rate that the hyperscalers uh collect. And I'm not so sure that uh unless they become the hyperscaler, um, they're which can happen, right? They could go out and, you know, gobble up all you know, all all these neo clouds that are down seventy five percent that have access to land power and energy. Um, but so will everyone else, right? In theory, if they're such a bottleneck. Um they've

53:39Unidentified speaker (UNKNOWN): got

53:39Doug Garber: they're gonna have 200 billion dollars of cash, so they're in a good spot, and they have the customer, so they can make add value by giving that security.

53:47Unidentified speaker (speaker_1): Yeah.

53:47Unidentified speaker (UNKNOWN): Uh

53:48Doug Garber: in theory.

53:48Unidentified speaker (UNKNOWN): And

53:49Unidentified speaker (speaker_1): so I

53:49Doug Garber: I would be surprised if they go vertically integrated. It's just

53:52Unidentified speaker (speaker_1): they

53:52Unidentified speaker (UNKNOWN): have

53:52Unidentified speaker (speaker_1): to, they

53:52Unidentified speaker (UNKNOWN): have

53:53Unidentified speaker (speaker_1): to. And that's what I mean, that's what $900 billion of CapEx in the future is telling us. And so again, it's about the utilization of all that stuff. And so my biggest number again was that seven to ten percent sales and marketing number. Historically, like the so they have to go into these enterprises, right? As of right now, most of their sales are either through the hyperscalers or kind of like product led growth, right? People are just hearing about Anthropic, hearing about Claude, going in and turning on, you know, Claude in um in um the app store and kind of doing your traditional search through Claude cloud and and then starting to develop new workloads um over time within their operation. But I I think they're gonna have to spend more um on sales and marketing. And so that 35% number just seems uh a little bit elevated relative to any other, you know, capital intensive business that's out there. The only other comparable you could say uh again, and it's all dependent on whether they become the hyperscaler themselves. is we know the cloud players generate roughly $500 billion of revenue today.

54:56Unidentified speaker (UNKNOWN): And

54:57Unidentified speaker (speaker_1): they're running almost every workload in the world, right? $500 billion today. Again, you know, remember that when you're thinking of the numbers that they have to get to.

55:05Doug Garber: $350. I remember your map.

55:07Unidentified speaker (speaker_1): Remember $350 is to justify 1.4. I think that's generally, you know, if they continue to execute, they could they could possibly get to 350. But now if you want to double your money, I want to go from 1.4 to 2.8 trillion.

55:21Doug Garber: Yeah, because why buy it if there's not that same upside?

55:24Unidentified speaker (speaker_1): Yeah. So like we're off we're focused on the 1.4, but like you need 2.8. You need, you know, 4.5, right? Um, for this to be a viable investment. And so at those same numbers, just to keep throwing it, is you know, at 2.8. two point eight, you need seven hundred and four billion dollars of revenue. Again, remember, the five clouds combined generate five hundred billion dollars today and they've been running workloads for the last 20 years.

55:50Unidentified speaker (UNKNOWN): Um,

55:51Unidentified speaker (speaker_1): and they're gunning for the same workload. So Anthropic's likely to go down into infrastructure, go vertical from the ground up. We know those uh c hyperscalers have roughly, you know, depending on which one, but call it you know, 25 to 30 percent operating margins. Um, AWS, Google GCP, Microsoft Azure, that's the known known. So there is precedent if you own full stack, um, that you can have margins that you know get to those levels. Um, so I think if you think they're gonna get the 35% margins or 30% margins or 25% margins, you have to believe that they're gonna go full stack, execute, and be the cloud for individuals and enterprises alike. And if you believe that, then I think you could you you could probably justify, you know, somewhere in those mid 20s to 30s. Um, if not, you know, I think you are a using and utilizing third-party infrastructure for most of your stuff. People are buying you from third parties. Harnesses and orchestrators are choosing your model, but not because of your because of you. Um, and so you still have to stay at the frontier to make the next sale. And so

57:04all of that combined is you know, it's hard to, you know, to figure out the actual operating, you know, margin number. You have to follow the product, and the product is gonna going to tell us, you know, over the next six, twelve, eighteen months. is who's at the the frontier.

57:18Unidentified speaker (UNKNOWN): You

57:18Unidentified speaker (speaker_1): know, when are we getting to kind of the superintelligence where it's recursed recursive? And if it we get to that point, does that actually erode margins away? Um, you basically have to be the infrastructure at that point. And so if you don't have the customer relationship, which I'd argue enthropic probably doesn't have the customer relationship the way they explicitly say in the in the IPO perspectives. Look, it's a great company. I mean to go from where they are today to $80 billion is amazing, but they still have to continue to execute at a very high level on the model, own the infrastructure, and get to the customer relationship. If that, then you know, all bets are off. But

57:55Unidentified speaker (UNKNOWN): and

57:55Doug Garber: they've got to continue to be the leading frontier one when there's a lot of people chasing them and there's a lot of new models. and it's not very sticky. People switch, they've just got to continue to create a new frontier. And you know, I think right now one of the reasons maybe sales and marketing as a percent of revenue might be a little low. I mean, things I hear is like you got to be doing at least 10 million of ARR for them to even they're so busy with inbound they don't have to do any marketing, right? Everyone wants them to kind of come in and install it. I mean,

58:25Unidentified speaker (UNKNOWN): you

58:25Doug Garber: know, so you can't. get their attention, right? Um, that that's why they don't need sales and marketing right now. But eventually, right, it'll get more competitive. Right now we're in

58:35Unidentified speaker (UNKNOWN): a

58:36Doug Garber: tight uh t uh, you know, a shortage of compute. What are you hearing right now on um compute? Obviously, pricing's going up on shortages, people locking in. When do you talk to your companies post quarter?

58:48Unidentified speaker (UNKNOWN): Yeah,

58:48Unidentified speaker (speaker_1): I mean everyone's trying to figure out how to be more efficient with their with their token spending. And, you know, at the end of the day. So it's like you hear the Ubers of the world, you hear, you know, the the Zooms of the world. Uh you're essentially just watching their gross margins, right? Because that's ultimately, you know, where all the uh expenses are hitting, you know, some of the vendors that are consuming this stuff. Um and for the most part, the bigger ones are are figuring out ways to have multiple models, again, creating almost their own. uh orchestrator for inside their product so that if we're using zoom and we're summarizing and it's you know a a two hour you know summarization maybe they're using this model versus that model and then creating their own models right like um and and again not doing those you know frontier training models but you know taking uh the existing models and um post-training them you know relative to their to their workload workloads. so that's kind of what you're seeing in the field, which is, you know, people kind of getting um a little bit uh frustrated with the amount of spend. Because it's either you push people out to make room, you push vendors out, software vendors. But again, software is only two to three percent of your budget. So, you know, it there's not that much wiggle room to push that out. Um

1:00:00and, you know, do you really want to stand up your own software tools inside of your organization if you have 2,000 people and then train them on a new product? I don't know. And then you have to hire more people to keep it up and running and govern it and securitize, you know, keep it security associated with it. So,

1:00:17Unidentified speaker (UNKNOWN): you know,

1:00:17Unidentified speaker (speaker_1): everyone's, you know, from a compute side, um, on one side, you're hearing, you know, the people that are uh more inclined or incentivized to say that there's you

1:00:26Unidentified speaker (UNKNOWN): know

1:00:26Unidentified speaker (speaker_1): all these kinds. constraints.

1:00:28Unidentified speaker (UNKNOWN): But

1:00:28Unidentified speaker (speaker_1): then you hear, you know, new companies coming out with you know different types of techniques and you know, model efficiencies. Um, and you're seeing that like across the board. Like my my like everyone's gunning for this. There's so much money here. We're talking about a $1.4 trillion valuation. Like, you know, if if you have AI in your name, you're still raising capital. Um, and so, you

1:00:49Unidentified speaker (UNKNOWN): know,

1:00:49Unidentified speaker (speaker_1): until that dries up. you know, ultimately are you gotta ask so many questions, right? Are these people double ordering to, you know, to to you know, hold their hold the compute away from the competitors? There's there's rumors of that. Um, you know, that's why you hear Andy Jassy or so you

1:01:04Unidentified speaker (UNKNOWN): know,

1:01:04Unidentified speaker (speaker_1): AWS or even, you know, Zuckerberg at Meta say, if we have too much compute, we'll will still be able to use it eventually over time. And, you know, for me, that's just re reading between the lines that like if you're smart, strategic, and you think this is important to the longevity of your business, I'm gonna buy more compute, not less, and I'm gonna take it away from the hands of my competitors so that they're constrained indefinitely until I feel like we're at the leading edge. Um

1:01:31Unidentified speaker (UNKNOWN): so

1:01:31Unidentified speaker (speaker_1): I look, it's a the wild west. Uh, there's like 280 neo clouds now, they're all trying to be a new neo cloud. Um, there'll be a lot of winners, there'll be a lot, a lot of losers. Uh, I think AI is real. So I don't I'm not trying to sound negative, I'm just trying to sound disciplined, which is really just around, you know, let's just let it play out. There'll be a crash. And if you invested in Cisco back in 2002, Two thousand two, you are so happy right now. Um, if you you know, if you invested it a year before that. you know, you lost all your money. So

1:02:05Doug Garber: value invest in, you know, you got zoom in with the anthropic stake kind of for free now. And magically you have a few percent of your portfolio exposed to AI without direct exposure. So that's uh something for nothing.

1:02:21Unidentified speaker (UNKNOWN): And then

1:02:21Unidentified speaker (speaker_1): Okta, like during the SaaS apocalypse, we bought Okta, and look, you were buying that thing at you know, three times revenue. you know, fourteen percent margins at the time. You're paying maybe, you know, low teens, free cash flow margins on that thing. And our belief was like, we like AI, we believe AI, we use a lot of AI, so we're clear believers. Um, and we like anthropic, otherwise we'd sell like Zoom or something like that. We think it's a a pretty uh important company to the future. Um the but the you know, octa Okta and cybersecurity and identity and verification I think are super important. And so those were the obvious places for us to go. Hey, they don't have to spend on CapEx, they just have to, you know, procure and and and um verify identities. And you're seeing these agents go into you know bank accounts and all these other things, and all those things are are going to have to be identified, verified before they enter a a corporate entity's network. And so identity, anyways, we just saw sold it after it, but you know, because it ran up so you know a lot, which is which is

1:03:19Unidentified speaker (UNKNOWN): was

1:03:19Doug Garber: it two two three X?

1:03:21Unidentified speaker (speaker_1): Yeah, it was uh you know 70, you know, bucks roughly, two, you know, two ten roughly. Um in a short period.

1:03:27Unidentified speaker (UNKNOWN): Like a

1:03:27Doug Garber: three X in in six months. Yeah, not not bad.

1:03:29Unidentified speaker (speaker_1): Yeah, and so we stayed disciplined on the on the exit and just said, you know what? You know, at this level, fifty 50 times, you know, multiple and great company.

1:03:37Doug Garber: If you were Vise advising in the Zoom board as a shareholder, and I think not a lot of people have eight, nine percent of their portfolio, maybe just

1:03:46Unidentified speaker (UNKNOWN): uh

1:03:47Doug Garber: you and Yang have that weight of your net worth in Zoom. Would you advise liquidating and selling it at 14 one four trillion valuation, 17? one seven, What would you be doing?

1:04:03Unidentified speaker (speaker_1): Yeah, I mean I I would definitely uh liquidate it, you know, maybe give um a small little special dividend uh back to the investors. Then if I decide I want to reinvest that back in this in anthropic, I'd do so. Um, you know, on my own on my own terms, uh at this level. They already did a great job, fifty three million to one point or fifteen 15 billion to you know, one 1.4 point four trillion for them is you know, they should pat themselves on the back uh and give uh um you know give the money back and and have the uh the the shareholders make their decision if they want to uh carry that forward. Um and then utilize that cash. I say I I say you use that capital and you just you go out into the software world, you find vendors and you create one of the larger platforms out there that's all context context-based. space. Um Um, and you build a meaningful you know business around context. Uh, they they they took over the CTO from you know Microsoft, who uh led their AI initiatives, you know, back in the day. And so there's a lot of talent starting to come back to Zoom because I think they see them strategically important given that context layer. And so furthering that along, becoming more they already are, they have Zoom phones, phone, Zoom, you know, context. contact center for you know support and such, and all of that is additional context end to

1:05:18end. Anyways, I would I would do um some sort of uh you know special dividend. Uh actually the level of capital they have, I would just, you know, fundamentally just uh incorporate a dividend as well. Um just to get that that type of investor into the ecosystem. It doesn't have to be large, um, but you just grow it annually because you already you're you're spitting off. so much capital relative to the size of your business. And then you have so much cash. So like you have

1:05:44Doug Garber: buybacks. What's their dilution on stock based comp versus buybacks? How much is that eating up?

1:05:49Unidentified speaker (UNKNOWN): It's now negative.

1:05:50Unidentified speaker (speaker_1): Uh so that's good. Um they had that true upload.

1:05:53Unidentified speaker (UNKNOWN): But

1:05:53Doug Garber: the two billion, they're spending a few hundred million on buybacks to stocks Just offset at the comp. So the real free, you know, the real free cash isn't two billion to f to shareholders. No.

1:06:03Unidentified speaker (UNKNOWN): you're offset for

1:06:04Doug Garber: SPC.

1:06:04Unidentified speaker (speaker_1): They're migrating down to like, you know, uh low teens uh re uh SBC versus revenue. Uh they used to be like 25, which was elevated, super elevated, right? That's like Snowflake type stuff. Um and so you know that's migrating down to like your 12% or so. As long as you have revenue continuing to grow, then you know, you're in a good spot there. Um and revenue's accelerating, so you feel comfortable with that. So but they they totally took down their their SBC initiatives from years past. Um, you know, shareholders definitely made it known. So

1:06:37Doug Garber: awesome. Well, Sean, thank you for coming on. It's always great to to catch up and get an update on my investment.

1:06:46Unidentified speaker (UNKNOWN): This is our quarterly meeting,

1:06:47Unidentified speaker (speaker_1): guys. This is our quarterly meeting.

1:06:49Unidentified speaker (UNKNOWN): We're

1:06:50Doug Garber: we've got we've been talking about it. We gotta start doing it so I could um hold you accountable to hey

1:06:56Unidentified speaker (speaker_1): what's going on here and there.

1:06:57Doug Garber: Why why why'd you make a mistake here? Or why you you know, give you a little bit of credit for your three X's, why you making these changes, why you overweight, just kinda

1:07:06Unidentified speaker (speaker_1): sure

1:07:06Doug Garber: get an update. We gotta we gotta do

1:07:08Unidentified speaker (UNKNOWN): it. I like it, I like it.

1:07:09Doug Garber: Um so

1:07:10Unidentified speaker (UNKNOWN): we'll

1:07:10Doug Garber: get it going. It'll be a fun conversation. It's um it's on the books for twenty 2027, twenty seven, my friend.

1:07:14Unidentified speaker (UNKNOWN): Perfect.

1:07:15Unidentified speaker (speaker_1): Love

1:07:15Unidentified speaker (UNKNOWN): it. Thanks

1:07:16Doug Garber: for your time. Awesome

1:07:17Unidentified speaker (speaker_1): Doug. Always

1:07:18Unidentified speaker (UNKNOWN): appreciate it,

1:07:18Unidentified speaker (speaker_1): man. Good stuff. Alright,

1:07:20Doug Garber: buddy. Bye.

1:07:20Unidentified speaker (speaker_3): Please make sure to follow Pitch to PM on your favorite podcast app and never miss out on episodes. This show is for entertainment purposes only. Please consult a professional before making any decisions. This show is copyrighted by Pitch the to PM. Written permission must be granted before rebroadcasting.

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