Panel debates Anthropic IPO valuation, leadership transition, and AI economic impact amid competitive pressures.
Summary
Speakers analyze Anthropic's IPO pricing, noting strong revenue growth but questioning pricing power and token commoditization risks. The discussion highlights the challenge of transitioning founder-led leadership to operator-focused management in public markets. Competitors like OpenAI and emerging chip suppliers are scrutinized for their impact on Anthropic's market position. Speakers also debate the broader economic implications of AI, including productivity gains versus job displacement risks.
Key takeaways
- 1 Anthropic's amended S-1 showed 12x year-over-year revenue growth, projecting over $100 million in annual revenue by 2027 if growth decelerates to 2x.
Unidentified speaker (speaker_2) · 11:00
- 2 Anthropic's accounting treats revenue from AWS customers as revenue while taking a cost of goods, which may inflate revenue growth figures compared to competitors like OpenAI.
Unidentified speaker (speaker_3) · 38:27
Episode sections
- 0:03 – 4:30
Anthropic IPO Context and AI Interaction Norms
The panel discusses the pricing of Anthropic's IPO and engages in a lighthearted debate about anthropomorphizing AI models like Claude and ChatGPT. Speakers share anecdotes about how politeness and prompt engineering affect responses from different LLMs.
- 4:30 – 7:57
AI Integration in Financial Data Terminals
Speakers compare AI integration in financial platforms, noting that FactSet and Bloomberg Terminal use AI to provide direct data answers without conversational filler. They discuss the advantages of audit trails and the historical sales strategy of the Bloomberg Terminal.
- 7:57 – 10:52
Anthropic's Growth Drivers and Competitive Landscape
Cory Johnson attributes Anthropic's rise to enterprise adoption of Claude Code. He outlines valuation challenges including potential revenue loss from model distillation, competition from open-weight models, and the resurgence of OpenAI.
- 10:52 – 13:55
Valuation Skepticism and Pricing Power Risks
Jacob Sonenshine questions Anthropic's pricing power, arguing that benchmark performance does not guarantee real-world utility. He warns that token prices may commoditize, potentially undermining revenue projections based on current high-value token assumptions.
- 13:55 – 19:57
IPO Trading Dynamics and Lockup Analysis
The panel discusses the immediate post-IPO trading behavior, comparing it to SpaceX and Snowflake. They analyze the impact of passive index fund buying requirements and the unique dynamic of insiders being allowed to sell shares on day one.
- 19:57 – 21:56
Supply Chain Constraints vs. Semiconductor Models
Nancy Tengler contrasts Anthropic's asset-light model with traditional semiconductor companies like Intel. She argues that while Anthropic lacks control over chip supply, its ability to secure data center capacity and financing positions it well for future growth.
- 21:56 – 23:45
Supply Chain and Leadership Complexity
Speaker_1 discusses the complex stack required for NeoClouds to secure chip supply from NVIDIA, noting that economies of scale depend on multiple factors aligning. Speaker_0 highlights the critical role of leadership in high-stakes AI companies, citing the immediate stock repricing of NVIDIA following news about Jensen Huang.
- 23:45 – 31:02
Anthropic Leadership and Financial Discipline
The panel debates whether Dario Amodei can transition from startup founder to public company leader. Speaker_2 argues that successful tech behemoths often evolve to have operator-led leadership rather than just technologists, suggesting Anthropic needs financial discipline to generate cash and manage liabilities.
- 31:02 – 34:18
Valuation and Private vs. Public Market Dynamics
Speaker_0 contrasts private market valuations, where investors agree to keep prices high, with public markets that reprice assets constantly. He expresses concern that a leader focused on 'existential threat' rhetoric may not be ruthless enough for public shareholders compared to one focused on margins and profitability.
- 34:18 – 40:31
Regulatory Risks and S-1 Disclosures
Speaker_1 raises concerns about catastrophic risks from agentic AI and whether insurers have modeled these risks, noting the California Attorney General's lawsuit against OpenAI. Speaker_3 criticizes the brevity of the S-1 filing regarding data center commitments and the lack of detailed pricing information, which complicates analysis.
- 40:31 – 45:02
Competition and Semiconductor Diversification
The group discusses competition from China, with Speaker_1 arguing that AI technology is broadly available and that semiconductor supply is diversifying beyond NVIDIA to include AMD, Broadcom, and others. Speaker_0 believes NVIDIA's dominance will last at least 18 months, while others see competition starting immediately.
- 45:02 – 48:49
Investment Implications of Chip Supply
Speaker_2 suggests that if NVIDIA loses share, investors should buy AMD as the 'Robin to NVIDIA's Batman' and consider TSMC if valuations are cheap. Speaker_3 argues that stricter export controls on GPUs to China due to the Jensen Huang scandal will hinder Chinese AI models, making it bullish for Anthropic and OpenAI.
- 48:49 – 50:38
Market Broadening and Compute Economics
Speakers debate whether AI development is bottlenecked by chip access or if market broadening will allow diverse models to emerge. One speaker argues Anthropic's approach is inefficient compared to competitors leveraging cheaper compute, while another counters that Anthropic's cost of compute is dropping quarterly.
- 50:38 – 1:17:50
NVIDIA's Market Position and Share Buybacks
The host asks Nancy Tengler about NVIDIA's recent decline. Tengler attributes the drop to 'wrong reasons,' citing the market's tendency to price in competition, but highlights NVIDIA's massive share buyback authorizations as a financial cushion.
Assets discussed
| Asset | Kind | Stance | Context |
|---|---|---|---|
| Anthropic | private company | mixed | Obviously, this is all this is obviously we all use it in one way or another. Obviously, it's changed all of our lives, and obviously this is an amazing thing. But uh, what's it really all worth, right? And we're actually gonna find out today, right? I mean, anthropic, like I said, went, I don't want to say went from zero to hero, but you know, it was an also ran, uh, and now |
| SpaceX | private company | neutral | And there there are some things about the SpaceX and uh and is probably the closest analogy, because there are new rules in place for big IPOs. |
| S&P 500 | index | neutral | force uh entry of these startups, these new uh these newly uh public companies into their uh SP 500 index, the Nasdaq did not. |
| OpenAI | private company | neutral | get the same supply, i.e., OpenAI, Google, Amazon, Meta, Microsoft, everyone wanting to get the chips to run the tokens. |
| AlphabetGOOGL | equity | neutral | Yeah, sometimes these people step away of their own accord. Eric Schmidt came into Google early to be the steady hand and the adult voice as those two founders m retreated to, you know, working on science projects. And it's easy to imagine Dario sort of saying, I'm going to focus on the technology development of this. |
| TeslaTSLA | equity | neutral | And what that means, if you need a catalyst for that evolution, because everybody, nobody, no employee, no other executive, no shareholder has the money to become an activist. That's I that's I'm just being c crazy for a second. That's not the same. What I'm saying is nobody wants to remove the visionary. Yeah. And Tesla, nobody wants to remove Musk. So what my point is, is if there's some |
| Meta PlatformsMETA | equity | neutral | So one of the reasons we launched the Quiod Research Product is because there's the quiet period so much shorter. Facebook went public from the time of the S1 for Facebook to the Facebook book IPO was 107 days. Alibaba was 138 days. |
| AlibabaBABA | equity | neutral | So one of the reasons we launched the Quiod Research Product is because there's the quiet period so much shorter. Facebook went public from the time of the S1 for Facebook to the Facebook book IPO was 107 days. Alibaba was 138 days. |
| AmazonAMZN | equity | neutral | that great mostly impacts uh their wholesalers. So when an Amazon or a Google, which are their two biggest customers, |
| DeepSeek | private company | neutral | When when this when the DeepSeek thing came out, I thought that was a good thing, right? Because it basically just spurred competition. |
| NVIDIANVDA | equity | mixed | you saw a moment when the Huang news broke where NVIDIA went down, AMD Broadcom went up, you saw the crisscross. |
| Broadcom Inc.AVGO | equity | neutral | share to AMD and Broadcom? Instead of trying to pick winners, |
“Stance” is how the speakers talked about the asset, not a recommendation.
Claims worth verifying
ChatGPT advised the host not to buy Anthropic stock on IPO day because it was fully priced.
Unidentified speaker (UNKNOWN) ·1:50
How to verify: Check the transcript at [1:50] for the host's claim; verify via independent AI interaction logs if available. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic's revenue grew 12x year-over-year in the amended S-1 filing.
Unidentified speaker (speaker_2) ·11:00
How to verify: Review Anthropic's amended S-1 filing on SEC EDGAR for revenue growth figures. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourceSnowflake rose 29% in the first 12 months after its IPO but had a 2.5% annualized return over six years.
Unidentified speaker (speaker_1) ·16:22
How to verify: Check Snowflake's (SNOW) stock price history from September 2020 to 2026 on financial data providers. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary source5-7% of Anthropic's revenues were from distillation.
Unidentified speaker (speaker_3) ·9:54
How to verify: Review Anthropic's S-1 filing for revenue breakdowns or management discussion on distillation usage. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic insiders are allowed to sell shares on day one of the IPO.
Unidentified speaker (speaker_3) ·18:12
How to verify: Check the S-1 filing for lockup agreements and exceptions for insider sales. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourcePassive index funds will buy approximately 16% of Anthropic's float.
Unidentified speaker (UNKNOWN) ·17:52
How to verify: Calculate based on index inclusion rules for Nasdaq, Russell 1000, and MSCI indices and Anthropic's float size. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Anthropic is designing its own chips and is not completely locked into NVIDIA architecture.
Unidentified speaker (UNKNOWN) ·21:55
How to verify: Review Anthropic's S-1 filing for capital expenditure plans and chip design disclosures. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourceClaude responds better to polite prompts than other LLMs.
Unidentified speaker (speaker_3) ·2:21
How to verify: Cite the 'quant' research mentioned by the speaker; verify via independent benchmark studies on LLM responsiveness. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
OpenAI has taken the lead over Anthropic in token generation volume.
Unidentified speaker (speaker_3) ·10:43
How to verify: Check Anthropic's S-1 filing or public reports for token usage comparisons with OpenAI. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourceSnowflake's IPO was on September 18, 2020.
Unidentified speaker (speaker_1) ·16:22
How to verify: Check historical stock data for Snowflake (SNOW) for its IPO date. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourceNVIDIA stock repriced immediately following news about Jensen Huang.
Unidentified speaker (speaker_0) ·23:05
How to verify: Check NVIDIA (NVDA) stock price charts and news archives for the date of the 'Jensen Huang news' referenced in the transcript (likely late 2026 or 2026 based on context, verify via SEC filings or major financial news outlets like Bloomberg/Reuters). For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic is like a 1970s Cadillac going up against the oil crisis, having built on getting the most money to get the most chips the quickest ahead of everyone else.
Unidentified speaker (UNKNOWN) ·49:31
How to verify: Check Cory Johnson's public commentary or Epistrophy Capital Research publications for this specific analogy and view on Anthropic's strategy. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Anthropic's cost of compute is going down, going down, going down every quarter relative to what they started from.
Unidentified speaker (UNKNOWN) ·50:08
How to verify: Verify against Anthropic's amended S-1 filing or public financial disclosures regarding compute costs and training efficiency metrics. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourceNVIDIA authorized $230 billion in share buybacks, spent $19 billion, and authorized another $150 billion.
Unidentified speaker (speaker_1) ·51:47
How to verify: Check NVIDIA's (NVDA) SEC filings (Form 8-K or quarterly reports) for share repurchase authorizations and execution reports. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic does not have 10 years of cash on its balance sheet; it has a few years.
Unidentified speaker (UNKNOWN) ·55:43
How to verify: Review Anthropic's amended S-1 filing for cash, cash equivalents, and restricted cash balances versus net burn rate projections. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic has at least $230 billion on its balance sheet net of debt, more than all hyperscalers combined at their 2020 peak.
Unidentified speaker (speaker_3) ·57:03
How to verify: Check Anthropic's amended S-1 filing for total cash and debt; compare with historical cash positions of Microsoft, Alphabet, Amazon, and Meta from 2020 filings. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic has less than $70 billion in obligations against $230 billion in cash, allowing it to honor commitments for two years if operations shut down.
Unidentified speaker (UNKNOWN) ·57:34
How to verify: Analyze Anthropic's amended S-1 filing for 'commitments and contractual obligations' and compare to total cash and equivalents. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Primary sourceAI is currently augmenting or replacing entry-level programmers, but this does not support the entire economy's need for efficiency gains in heavy processes.
Unidentified speaker (speaker_1) ·1:12:04
How to verify: Review economic reports on AI productivity gains by role (e.g., NBER working papers) and industry-specific efficiency metrics for large enterprises. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Voters are focusing wrath on data centers, which could turn to AI broadly due to employment changes.
Unidentified speaker (speaker_3) ·1:12:52
How to verify: Search news archives for 2026-08-19 through 2026-11-19 for reports on local opposition to data center construction and political statements regarding AI and employment. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
The speaker's 14-year-old does not like AI due to existential fears of job replacement.
Unidentified speaker (speaker_2) ·1:13:34
How to verify: This is an anecdotal claim; verify via the speaker's social media or public statements during the specified date range if available, otherwise mark as unverified anecdote. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
College students are changing career goals to work for AI companies because traditional entry-level jobs are disappearing.
Unidentified speaker (UNKNOWN) ·1:14:04
How to verify: Check university career center surveys or labor market data from 2026-08-19 through 2026-11-19 for trends in student major selection and job applications in tech. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.
Do your own research
- Review Anthropic's amended S-1 filing for detailed pricing models and cost of goods sold disclosures.
- Analyze NVIDIA's share buyback authorization and spending trends to assess financial health.
- Compare Anthropic's inferencing margins against competitors like OpenAI and Mistral.
- Monitor export control regulations affecting GPU sales to China and their impact on global AI competition.
- Evaluate Anthropic's cash burn rate and refinancing risks given current interest rate environments.
- Assess the competitive landscape for enterprise AI tools, specifically Claude Code adoption metrics.
- Check for any legal or regulatory actions against AI companies regarding data privacy and model safety.
- Verify claims about AI productivity gains in non-tech sectors through independent case studies.
Full transcriptexpand
0:03Unidentified speaker (speaker_0): Oh boy. I every day I feel like this week I'm starting with we finally got here. Uh guys, we did finally get here. We have a great group of folks coming in today. We're talking, it's anthropic day. It's
0:14Unidentified speaker (speaker_1): anthropic day. Waiting to exhale.
0:15Unidentified speaker (speaker_0): Yeah, yeah. And we just literally were well, we can exhale a little bit, right? Because now we've actually priced the stock, right? We've been waiting, not exhaling all day long. It's pretty funny, you know. Uh I like to refer to uh Claude as one of my you know my favorite employees. Okay. Uh but it, you know, Claude started out as you know, Claude started out as a uh let's call it a uh an intern, uh what seems like about a year ago, but has risen to the star employee. Uh and this is Claude's big day with Anthropic, and we're finally here, and we're gonna talk a lot about Anthropic today. But
0:48Unidentified speaker (speaker_2): I I have an imaginary relationship with Claude.
0:50Unidentified speaker (speaker_0): Do you? Is it a good one or a bad one?
0:52Unidentified speaker (speaker_2): It's a fantastic relationship.
0:54Unidentified speaker (speaker_0): Okay. Yeah. It's only only you have only one employee, Claude, or do you like the uh what do you think
0:59Unidentified speaker (UNKNOWN): of the other
0:59Unidentified speaker (speaker_0): guy?
1:00Unidentified speaker (speaker_2): I have Claude and GPT work together extremely well. Everybody's respectful and everybody builds on each other.
1:06Unidentified speaker (speaker_0): It's funny that you mentioned that because I did the unthinkable today. I actually did go to uh ChatGPT chat GPT. Uh and I I you know like I I'm I it's actually funny, I treat them like like real people, right? So I'm afraid to
1:19Unidentified speaker (UNKNOWN): you're cheating.
1:19Unidentified speaker (speaker_0): I'm like I'm afraid to like uh to like sort of sour what their opinion is of me, right? So I'm like I'm I'm constantly guarding myself, like what am I gonna say to this thing that it might you know hold it against me in the future or think that I'm a bad person. And so I thought for the longest time it would be bad for me to ask Chat ChatGPT GPT if I should invest in. you know in Anthropic on IPO day. So and I did.
1:43Unidentified speaker (UNKNOWN): And
1:43Unidentified speaker (speaker_0): uh and Chat GPT uh answered me with candor uh and said uh uh no, don't buy
1:50Unidentified speaker (UNKNOWN): it, don't buy
1:50Unidentified speaker (speaker_0): it today.
1:51Unidentified speaker (UNKNOWN): Don't buy it.
1:51Unidentified speaker (speaker_0): Don't buy it today. It's fully priced, wait till everything simmers down. So
1:56Unidentified speaker (speaker_1): you
1:56Unidentified speaker (speaker_3): know, there's something to what you were saying there. I'm Corey Johnson, by the
1:59Unidentified speaker (UNKNOWN): way,
1:59Unidentified speaker (speaker_3): from the
1:59Unidentified speaker (UNKNOWN): Corey
1:59Unidentified speaker (speaker_0): Johnson.
2:00Unidentified speaker (UNKNOWN): Hi Corey Corey Johnson. Johnson.
2:01Unidentified speaker (speaker_3): Uh before For our listeners are very or viewers who don't know us, there I was I had dinner with her or lunch with a quant maybe two years ago. And she told me she was running um uh uh prompts through di OpenAI, through Chat GPT, the same prompt through Claude, and trying other LLMs. And what she found was the nicer she was, the better responses she got, number one. And she found that Claude was more responsive to kindness than some of the other LLMs. This is again going back a few years. And she encouraged me to talk to these um these these uh LLMs as if they were one of my children, or as if if I if I was talking to a woman and just be kind and polite and not treat it like one of the dudes and demand something. And indeed, uh I've I've subsequently done some research on this, and they are because they are prompted to be a human-like human like interface, they do. respond better to kindness and even pleases and thank yous.
2:54Unidentified speaker (speaker_2): I need to try that approach because the other day.
2:57Unidentified speaker (UNKNOWN): Because you've been a jerk.
2:58Unidentified speaker (speaker_2): I asked
2:58Unidentified speaker (UNKNOWN): really yeah,
2:58Unidentified speaker (speaker_2): I've been a jerk. Like the other I'm
3:00Unidentified speaker (UNKNOWN): just respectful,
3:01Unidentified speaker (speaker_2): but like I I was getting uh frustrated the other day 'cause I I was under time pressure. I needed some data and I was asking it to to like help me crunch a little bit of data, just do it quicker. And like it gave it did not answer my question. And
3:15Unidentified speaker (UNKNOWN): I
3:15Unidentified speaker (speaker_2): literally woke wrote the words, come on, period. What I asked for is blah, blah, blah and then it gave it to me. But now I want to see if I I have better results than that good result if I'm nicer.
3:27Unidentified speaker (speaker_0): I
3:27Unidentified speaker (speaker_2): said, come on.
3:28Unidentified speaker (UNKNOWN): Was this
3:28Unidentified speaker (speaker_0): with Claude or with Chat GPT?
3:30Unidentified speaker (speaker_2): This was with Gemini. Oh
3:32Unidentified speaker (speaker_0): with Gemini? Yeah.
3:35Unidentified speaker (speaker_1): I don't engage in the anthropomorphism anthropomorphization of these things. I jump. just go in, I ask something, and I walk away. I don't even say thank you. Oh
3:43Unidentified speaker (speaker_0): really? I don't
3:43Unidentified speaker (UNKNOWN): say thank you, but
3:44Unidentified speaker (speaker_0): I say please. I say please. I
3:46Unidentified speaker (UNKNOWN): just
3:46Unidentified speaker (speaker_0): I just have to do it. Yeah.
3:48Unidentified speaker (UNKNOWN): It feels
3:48Unidentified speaker (speaker_1): instinctual, feels impulsive.
3:51Unidentified speaker (speaker_2): In your position, is let's not let's not pretend that it's not a machine.
3:54Unidentified speaker (speaker_1): And I think in a year from now we won't be doing that as much. I think it'll be in other applications and it'll just be a thing that's there. You will not be acting like it's your girlfriend. or your date or your boyfriend or your colleague. It'll be behind the scenes more
4:08Unidentified speaker (UNKNOWN): if it
4:08Unidentified speaker (speaker_1): works out.
4:08Unidentified speaker (speaker_3): Maybe this stuff just wasn't meant for New Yorkers. Yeah. Maybe it was meant
4:12Unidentified speaker (UNKNOWN): for lead with kindness. I
4:15Unidentified speaker (speaker_0): mean, I think I I should call it you, right? You, right? It's a New York thing. Use, use,
4:20Unidentified speaker (UNKNOWN): use. Well if you
4:21Unidentified speaker (speaker_3): go to
4:21Unidentified speaker (UNKNOWN): If you go to the
4:22Unidentified speaker (speaker_3): plural, if you go to both if you have a model router
4:24Unidentified speaker (UNKNOWN): and
4:24Unidentified speaker (speaker_3): you can send the query to both chat ChatGPT GPT. and
4:27Unidentified speaker (UNKNOWN): to
4:27Unidentified speaker (speaker_3): Claude and to uh GLM. Right.
4:30Unidentified speaker (speaker_1): Yeah.
4:30Unidentified speaker (speaker_0): Wow.
4:30Unidentified speaker (UNKNOWN): But
4:31Unidentified speaker (speaker_1): but for example, the applications are doing this. I use FactSet,
4:34Unidentified speaker (UNKNOWN): uh-huh,
4:34Unidentified speaker (speaker_1): which is the terminal program. It's like Bloomberg Terminal, and it's a wonderful application because it has all
4:39Unidentified speaker (UNKNOWN): the
4:39Unidentified speaker (speaker_1): data there. And and I go in and I type in what was the change in Invaluation valuation for Snowflake from its IPO in 2020 to a year later, and it goes boom. and it gives it to me. And there's no chit chat, there's no like thank you ask fact set. It's just like good, got it. And I move home.
4:58Unidentified speaker (UNKNOWN): You put the fact set data
4:59Unidentified speaker (speaker_2): into Excel.
5:00Unidentified speaker (speaker_1): I don't have to put anything in because it's all in fact set. It's got a prompt interface. It's just like these chatbots.
5:07Unidentified speaker (speaker_0): Bloomberg has that. Have you tried?
5:09Unidentified speaker (UNKNOWN): Ask B. Ask B
5:10Unidentified speaker (speaker_0): is pretty
5:10Unidentified speaker (UNKNOWN): good. I
5:10Unidentified speaker (speaker_2): I feel not advanced now because I've been downloading facts that data into Excel and I do a little something in Excel to see what I calculate, put into GPT or Claude and say, you
5:20Unidentified speaker (UNKNOWN): know,
5:20Unidentified speaker (speaker_1): this will replace that with data that has an audit trail, so it'll show you via hyperlinks this is the change in valuation over six months, and you click it and it shows you exactly how it's computing what the forward price to sales is. Well and so you don't have to wonder and you don't have to upload, and then it'll generate a table that you can then use to see
5:37Unidentified speaker (UNKNOWN): how it's gonna be. be able to do that. But now
5:38Unidentified speaker (speaker_2): we are getting into user share between between the the big models and then everybody else that exists, Bloomberg fax that whatever.
5:46Unidentified speaker (speaker_0): Yeah. I mean that's a reality, right? That that's where it's moving pretty fast. It's moving pretty fast. I gotta tell you, I think they launched Ask Bee. It was only a couple of months ago, maybe within six months. Yeah. Um and I see that it it's getting better. I definitely like it. I
5:59Unidentified speaker (UNKNOWN): mean, you're a you're
6:00Unidentified speaker (speaker_1): a Bloomberg terminal user.
6:01Unidentified speaker (UNKNOWN): I'm I'm a a
6:01Unidentified speaker (speaker_0): Bloomberg. Bloomberg terminal. Okay.
6:02Unidentified speaker (UNKNOWN): Okay, I'm a fact set
6:03Unidentified speaker (speaker_1): person. But they're both good applications of AI.
6:06Unidentified speaker (speaker_0): Yeah. Well, it's amazing, right? Because you know, Bloomberg is notoriously difficult to navigate unless you were born with a with a language or you know it.
6:14Unidentified speaker (speaker_1): And that's why
6:15Unidentified speaker (speaker_3): that was a sales strategy. So that was Mike's Mike Bloomberg's strategy was if you if you want help, you hit F what was that it F1, you hit it twice, right?
6:24Unidentified speaker (UNKNOWN): Right.
6:24Unidentified speaker (speaker_3): Or hit the help button twice. But his notion
6:26Unidentified speaker (UNKNOWN): is a very good thing.
6:27Unidentified speaker (speaker_3): was if you have to talk to your salesperson every time you want to learn something, you'll have to be in a regular dialogue with your salesperson. So there's never been a manual for the Bloomberg terminal for that reason.
6:35Unidentified speaker (speaker_0): That's
6:35Unidentified speaker (UNKNOWN): amazing.
6:35Unidentified speaker (speaker_1): But it's good to have service there if this stuff doesn't work, which is an added advantage. If none of it works, you can call up someone at Bloomberg or at Factset and say, I don't understand why it is generating garbage.
6:45Unidentified speaker (speaker_3): In a moment.
6:46Unidentified speaker (speaker_1): Yeah. Which you can't you cannot call anthropic as far as I know and say why is it generating garbage.
6:53Unidentified speaker (speaker_0): Oh no, I'm saying you could ask. You can ask
6:55Unidentified speaker (UNKNOWN): you. I have been
6:57Unidentified speaker (speaker_0): politely angry at at some of these things when like you're trying to get something from them and they're just not they're just not hearing you. Trevor
7:03Unidentified speaker (UNKNOWN): Burrus, Jr.:
7:03Unidentified speaker (speaker_2): I
7:03Unidentified speaker (UNKNOWN): mean
7:03Unidentified speaker (speaker_2): you can say I don't think it's impolite to say that's not what I'm asking for.
7:07Unidentified speaker (UNKNOWN): No,
7:07Unidentified speaker (speaker_0): that's correct. You hope it that's what that's how it learns.
7:09Unidentified speaker (UNKNOWN): Right. That's how it works. Yeah.
7:11Unidentified speaker (speaker_2): That's
7:11Unidentified speaker (speaker_1): how it becomes more sick of fun.
7:12Unidentified speaker (UNKNOWN): it. Saying
7:12Unidentified speaker (speaker_2): things like come on, I think like we can cut that out. We don't need that.
7:16Unidentified speaker (UNKNOWN): I
7:16Unidentified speaker (speaker_0): mean, look, it should know if it's frustrating you. Maybe
7:19Unidentified speaker (speaker_2): a little bit. You
7:20Unidentified speaker (UNKNOWN): know? Yeah.
7:20Unidentified speaker (speaker_0): Maybe apply some more uh apply some more tokens.
7:23Unidentified speaker (UNKNOWN): You're
7:23Unidentified speaker (speaker_1): going to train
7:24Unidentified speaker (UNKNOWN): up
7:24Unidentified speaker (speaker_0): the token speed a little bit.
7:26Unidentified speaker (speaker_1): Post-training it
7:26Unidentified speaker (UNKNOWN): to
7:27Unidentified speaker (speaker_1): be yeah.
7:27Unidentified speaker (speaker_2): We don't need chips for training. We need conversation for training. Yeah. We need chips.
7:33Unidentified speaker (speaker_0): Obviously, this is all this is obviously we all use it in one way or another. Obviously, it's changed all of our lives, and obviously this is an amazing thing. But uh, what's it really all worth, right? And we're actually gonna find out today, right? I mean, anthropic, like I said, went, I don't want to say went from zero to hero, but you know, it was an also ran, uh, and now it's like at the front of the pack. It's setting the uh it's setting up.
7:56Unidentified speaker (UNKNOWN): So let's talk
7:56Unidentified speaker (speaker_3): about that. Yeah. Yeah,
7:57Unidentified speaker (speaker_0): so
7:57Unidentified speaker (UNKNOWN): we've got to be able to do So we've
7:57Unidentified speaker (speaker_3): so we've we did a uh a a bunch of work on in this. We um I launched a new company in the face in the of this anthropic. IPO, or a new product at least, uh called the Quiet Period, which is research for the time after the S1 drops, or you know, we start it many months in advance until the uh the street can start to publish on these things. We know the underwriters won't be able to publish for another month or so. And one of the things we looked at was was the competitive edge and how it was that Anthropic, a company that started in two 2021, thousand twenty one, has become one of the biggest companies in the world. and bigger than OpenAI. And the the short answer is that it was it was clawed code. And it was enterprises trying to change the way that they do business by using clawed code throughout their enterprises. I got to interview hundreds of their partners as they came into San Francisco. I don't or know Anthropic didn't know I was doing this, but we they they were uh meeting near me, and so I would just stop. these people uh on a daily basis to talk to them about how they were using um Anthropic's products, how they were using Claude.
8:57Unidentified speaker (UNKNOWN): And
8:57Unidentified speaker (speaker_3): it was really interesting. These kind of uh uh evangelists within all kinds of corporations, the biggest and smallest companies you can imagine, were coming to San Francisco to get trained up by Anthropic on ways to use Claude within their organizations. And we saw it in the revenues of this company, how they really took off. in the last year with those incredible growth numbers. That was all about Cloud Code.
9:18Unidentified speaker (speaker_1): So what's it worth now, Corey?
9:21Unidentified speaker (speaker_3): Whatever the market's going to pay for it.
9:23Unidentified speaker (UNKNOWN): So can I
9:24Unidentified speaker (speaker_0): love that?
9:24Unidentified speaker (speaker_3): How how might we evaluate that? I'd love to I'll I'll I'll give you an answer if you guys will give one. I'm happy to hear
9:29Unidentified speaker (UNKNOWN): it. Go ahead. No, you first go ahead.
9:31Unidentified speaker (speaker_2): Go
9:31Unidentified speaker (speaker_0): ahead. Yeah.
9:32Unidentified speaker (speaker_3): All right, sure. I think we've got to assume that the growth rate comes down a lot, that maybe it's going to be gonna more like what we saw in the sequential growth rate from Q2 to Q3 over the course of the next few months. I think you've got to assume, you've got to answer some questions when you figure this out. Um what role will distillation play? The company said five to seven percent of revenues were from people essentially s using Claude to steal Claude's code until they could catch them. But five to seven percent of revenues is a huge number for this company. So is distillation gonna be part going of to
10:02Unidentified speaker (UNKNOWN): the company.
10:03Unidentified speaker (speaker_3): their business going forward? Another question you got to answer to figure out the valuation is: are open weight models a real threat? And maybe we can talk about that later. But are are people routing models, right? This week at Microsoft Ignite, Microsoft announced a product that will let a user write a prompt and it will route that model to the best uh right that prompt to the model that might answer it more cheaply. Does that mean the Chinese models are gonna going to steal market share from these guys, which is what we've seen? a little bit in the last year? These guys being OpenAI and Anthropic. And then thirdly, what is the role of OpenAI? In our report, we we we report that that uh OpenAI has actually taken the lead over Anthropic in terms of number of tokens uh being generated. Um and so you've got to assume some competition that wasn't there 18 eighteen months ago when Anthropic was running away with this thing. open OpenAI AI's is back. back and that competition is real.
10:52Unidentified speaker (speaker_2): And in terms of in terms of assessing where we are in the life cycle of the revenue growth rate, um if you look at the year over year growth, I think uh year to date to September, year to September this year, I think it was like twelve X revenue growth in the amended S one. Now so it's gonna grow I I have to imagine. in 2027 more than 2x, which means by the way, you're you're gonna get over, you're probably gonna get over $100 million in annual revenue by 2027. Um but but if you look at
11:27Unidentified speaker (UNKNOWN): I'm at 112, but
11:28Unidentified speaker (speaker_2): if you look at
11:28Unidentified speaker (speaker_3): what it's worth.
11:29Unidentified speaker (speaker_2): Right. So and it's like if you look at that 12x and and and then you decelerate it, and then you're also looking at quarter over quarter growth, I think. quarter over quarter, it's important to look at like how like even if there's like a an acceleration in Q over Q growth, like is it what's the magnitude of the acceleration? You know what I mean? Like
11:50Unidentified speaker (speaker_1): I don't think you can make any prognostications about growth in quarters out or years out until the question of pricing is addressed, because most of the revenues consumption revenue, that is something that is based on currently the most valuable tokens. They have the most valuable tokens because they ace all of the benchmark tests, which everyone knows can A be gamed, B are not connected to real work that real companies do. And so what is the status of their pricing power? And I've seen no one address it. Everyone assumes they'll maintain pricing power, which is going to be the multiplier for all of their token volume. And so whether they're in the f in the forefront or OpenAI is in the forefront, it's assumed, oh, they'll have these valuable tokens. The rest of the market is way below them in token price. And
12:42Unidentified speaker (UNKNOWN): I think it's
12:42Unidentified speaker (speaker_1): I think it's going to commoditize and they're not going to have anywhere near the prices they have, which means that then all of the revenue projections for the out year. Jacob, might not be two, two and a half times. But
12:53Unidentified speaker (UNKNOWN): if that were
12:54Unidentified speaker (speaker_1): might
12:54Unidentified speaker (UNKNOWN): be
12:54Unidentified speaker (speaker_1): much less.
12:59Unidentified speaker (UNKNOWN): Can I just pull the
12:59Unidentified speaker (speaker_0): emergency break up for a second? Go for
13:01Unidentified speaker (UNKNOWN): it.
13:02Unidentified speaker (speaker_0): I feel like we need to wait before we do that, everyone wants to know one thing, right? Let we're going to get into all this stuff. But I feel like we needed to start with where we are this second. But before we get started here, I want to bring on Nancy Tel. Tangler. Nancy, I think, is gonna join us. There you go. There you are, Nancy.
13:19Unidentified speaker (UNKNOWN): I know
13:19Unidentified speaker (speaker_0): we now I have
13:21Unidentified speaker (UNKNOWN): I
13:21Unidentified speaker (speaker_0): needed to get another
13:22Unidentified speaker (UNKNOWN): I
13:22Unidentified speaker (speaker_0): needed to get another bullish anthropic person. And so uh we I c I felt like I had to pull up the emergency brake. I need to balance this out. Hi Nancy.
13:32Unidentified speaker (UNKNOWN): Hi
13:32Unidentified speaker (speaker_0): gentlemen. How are you? How are you? I don't want to put you in the hot seat or anything.
13:36Unidentified speaker (speaker_1): But you are wearing a green jacket.
13:38Unidentified speaker (UNKNOWN): I'm wearing my rally jacket.
13:40Unidentified speaker (speaker_0): That's your rally, and that's good to go. So that's the I'm going to start with you then. The big question is: stocks trading now 97.25. Uh, if you didn't get it at the IPO price, um, do you buy it here? Do you buy it at the market on a day like today?
13:55Unidentified speaker (UNKNOWN): I don't think so. I think I think
13:57Unidentified speaker (speaker_3): you let this
13:58Unidentified speaker (UNKNOWN): one sit for a bit. Um i the market wants to go down. You got all those.
14:03Unidentified speaker (speaker_3): short sellers now coming out of their holes.
14:05Unidentified speaker (speaker_1): And I think you
14:06Unidentified speaker (UNKNOWN): just
14:07Unidentified speaker (speaker_3): wanna sit back and and look for opportunities to to pick this one off and some others at at a cheaper price. Uh if if you're a tra uh if you're a long term investor though, I'll say this it's not gonna matter where you bought it
14:19Unidentified speaker (speaker_1): because
14:20Unidentified speaker (speaker_3): five years from now, three years from now, you're gonna be glad you owned it and
14:23Unidentified speaker (speaker_1): you're gonna
14:23Unidentified speaker (UNKNOWN): wonder
14:24Unidentified speaker (speaker_1): why you didn't
14:24Unidentified speaker (speaker_3): own more. So I I I just think we we can debate, you know, all the ins and outs. There's a lot of unknowns, clearly. Uh and uh you know the AI trade is is the trade everyone loves to hate now. So um good luck seeing the market go up if uh without technology stocks. So I I think um cooler heads
14:46Unidentified speaker (UNKNOWN): will
14:46Unidentified speaker (speaker_3): ultimately prevail, but for now we're sitting
14:48Unidentified speaker (UNKNOWN): on the
14:49Unidentified speaker (speaker_3): sidelines.
14:49Unidentified speaker (speaker_0): You're you're but you're you're a buyer. How about you? What are your thoughts on that?
14:59Unidentified speaker (speaker_2): So from my memory of the first day of trading of SpaceX and then looking at that chart of anthropic today, it you have the initial surge and then you have it what looks like uh what looks like a pyramid. It's like the surge, like y when it's surging, you don't know. where the peak will be, then you get this little peak, and you get like the pyramid going down. I'm not a technician really at all. But the point is that I think you now have to respect, as Nancy just as uh referred to, you now have to respect that it is, it is showing us now, right now, that it is trading. I'd like to know what your dog thinks, but it is trading just the way uh um a a massive tech IPO used. you would think would trade, which means instead of thinking about intraday and tomorrow and my and my technical levels, you don't know what the techno level tech levels of support are gonna be. You have to let the volatility settle out. It will the volatility will decline. It'll still be volatile, and then think about your value valuation framework, which obviously Nancy has.
16:00Unidentified speaker (speaker_1): Let me just remind you that
16:02Unidentified speaker (UNKNOWN): IP.
16:02Unidentified speaker (speaker_1): IPOs over many years can underperform even when they're great companies. And my prime example is Snowflake, which went public September 18th, 2020. It rose by 29 percent in the first 12 months after the IPO, which bucked the trend because most IPOs trade down. But then, then Denmark, Mark, the annualized return in six years since 2020 is 2.5%. percent per annum for Snowflake. And it's still one of the greatest companies
16:31Unidentified speaker (UNKNOWN): in the world. From
16:32Unidentified speaker (speaker_2): what starting point?
16:32Unidentified speaker (speaker_1): From an enormous starting point. From the first day close at a huge pop. Six years after the IPO, it's still a wonderful company. It's more valuable now, and it's been a terrible annualized stock return. So you can pick a great company, buy it at IPO. and have a terrible annualized return over many years. Yeah.
16:56Unidentified speaker (speaker_3): So there are a lot of the there are a lot of we did an analysis for our quiet period report about um the lockup. And there there are some things about the SpaceX and uh and is probably the closest analogy, because there are new rules in place for big IPOs. And those are rules about the the role of passive investing of of of index funds. While the SP uh changed their plans to force uh entry of these startups, these new uh these newly uh public companies into their uh SP 500 index, the Nasdaq did not. And the Vanguard Total Return Index did not, the MSCI index indices. So when I I look at we actually did this kind of cool chart with sort of bubbles showing the release of shares. And uh there's after today, there's not much of a release of shares for a few months. But you've got, again, you've got the NASDAQ, the Russell 1000, the MSCI uh World, and the Morningstar uh total return or the Vanguard total return uh fund also all being required to own a lot of shares. We think that's gonna suck up about sixteen percent of the float. Yeah,
17:52Unidentified speaker (UNKNOWN): and
17:52Unidentified speaker (speaker_0): that's
17:52Unidentified speaker (UNKNOWN): right there. I will say one important thing.
17:54Unidentified speaker (speaker_0): I mean the beginning of December, so you're gonna see a surge in demand because of that.
17:58Unidentified speaker (speaker_1): But didn't we see that with
17:59Unidentified speaker (UNKNOWN): SpaceX?
18:00Unidentified speaker (speaker_1): We saw that with SpaceX where passives had to get gain. the
18:02Unidentified speaker (UNKNOWN): We did, but you
18:02Unidentified speaker (speaker_0): also had a lot, I believe there's a lot more supply coming, right? So, Corey, were you suggesting that there's just not a lot more supply coming from the
18:08Unidentified speaker (UNKNOWN): R. Well, you said
18:08Unidentified speaker (speaker_1): for 90 days, Corey, right?
18:12Unidentified speaker (speaker_3): Different. And really different from just about any IPO I've ever seen. Um, insiders are allowed to sell today.
18:17Unidentified speaker (speaker_0): Yeah.
18:18Unidentified speaker (speaker_3): There's a there's a lockup of over three hundred million shares effective today. I've been talking, I talked to an insider late last night sitting right here at my desk, um, uh a friend who's got a venture capitalist has got an office nearby, and she's got some a lot of anthropic shares from having worked there. And they're facing this decision. What do they want to do today when they've got some liquidity and a method of liquidity that they didn't have before? So one of the ways anthropic is putting more shares out there is not through selling them for the company, but allowing their in inve their uh employees to sell today and their non-executive director. directors uh to sell today. And so I think that changes the dynamic a little bit.
18:55Unidentified speaker (speaker_0): So I mean I y you you've done a great job at not telling me if you would buy it here today, Corey. Would you buy it here today?
19:03Unidentified speaker (UNKNOWN): I don't give an
19:03Unidentified speaker (speaker_3): individual compliance will not let me give uh individual investment advice. Um so I well I I I I hate to use that as a but it is a fact. Um uh how about
19:12Unidentified speaker (speaker_0): this
19:12Unidentified speaker (UNKNOWN): are you bullish on are you
19:14Unidentified speaker (speaker_0): bullish on the stock? at this level on a day like
19:17Unidentified speaker (UNKNOWN): today?
19:17Unidentified speaker (speaker_3): There are numbers of of there are a number of forks in the road that which we try to identify in our report and and with a lot of them I take a more bullish uh take than I typically do in in um in uh in other companies. But I believe, for example, that the challenge of the cheaper uh token models is not gonna going to prove to be as great to them as it might seem here. And I'll also offer two one reason is I think those models, to your point, if some of those models are teaching to the test. Well they're they benchmark well, they don't perform as well in real life. And that the users are we see anecdotes of users saying that. The other thing I'll point to maybe a bigger, more macro way of thinking of this, is, you know, in the semi if you said to Intel going public or or any semi in NVIDIA going public, said, wait a minute, every
20:02Unidentified speaker (UNKNOWN): year
20:02Unidentified speaker (speaker_3): you're going to cut your price in half, you're going to double the performance of your product, and that's supposed to be good for your business going forward, clearly that's a a losing business model. But and that's sort of what they're saying here. Even into this IPO, even in last week, releasing new models, saying they're cutting the prices of old models and it's increasing consumption. And that's what we've seen thus far for those things. It didn't stop happening with semiconductors. And
20:25Unidentified speaker (UNKNOWN): maybe it won't ever stop happening with the same. semiconductors. The crucial
20:27Unidentified speaker (speaker_1): difference, Corey, is Intel was in charge of its fabs, which meant it was in charge of supply. Ethropic is not in charge of supply. What they have done is put together financing, billions and billions, and plan hundreds of billions to rent essentially from other people, which gets to the question of what happens at data centers, what happens to the pace of build-out of data centers, what happens to other people trying to
20:50Unidentified speaker (UNKNOWN): do that.
20:51Unidentified speaker (speaker_1): get the same supply, i.e., OpenAI, Google, Amazon, Meta, Microsoft, everyone wanting to get the chips to run the tokens. It's a battle for supply. Some of it won't be available, some of it will be available at different prices. And so this is a company whose Moore's law as it would be to lower the price of their cost to serve the consumption-based consumption based revenue is not within their control. It's not in their factory. It's going to market and hoping to get the best cost uh and bring down their own operating cost. And it's a large variable for them.
21:22Unidentified speaker (speaker_3): Thus far they've done that, right? Thus
21:24Unidentified speaker (UNKNOWN): far they've
21:25Unidentified speaker (speaker_3): they've done the best job of capturing uh future commitments from data centers, having the contracts thus far with flexibility that they can handle. Now they have come out of this with enough to handle all of their capital needs for the next few years at least. Forget what
21:38Unidentified speaker (UNKNOWN): we're doing.
21:39Unidentified speaker (speaker_3): whatever cash flow they might continue to generate. And they and they've they've locked in capacity when they where they need it. They've made the arrangements with companies where they've need it. They're not completely locked into an NVIDIA architecture now with using AMD and and we understand them designing their own chips, you know, performing
21:55Unidentified speaker (UNKNOWN): the other thing that depends better.
21:56Unidentified speaker (speaker_1): Some of that depends on other things that happen in the market with neo clouds and with availability of chips and with the bank of NVIDIA that funds the NeoClouds and others who are supposed to serve the chips that they're supposed to rent. So there's a whole stack of things that have to come together for them to actually have the supply that they need to thereby create economies of scale. And I'm not sure if you're
22:19Unidentified speaker (UNKNOWN): not
22:20Unidentified speaker (speaker_0): the N-word, right, how long can we get a go thro through without saying NVIDIA? Jensen.
22:25Unidentified speaker (UNKNOWN): You
22:25Unidentified speaker (speaker_0): said it, you said it, right? And so that brings me to something that's really been on my mind the last couple of weeks, right? None of us were thinking about this until, I don't know, last week. That, you know, what a an a what an important impact uh the leadership can have on a company. We're describing what is a high-stakes model here, right? Running these companies is not the same thing as running a lot of the tech companies that we become all familiar. and love with, right? This is a this is a lot of moving parts. Um and uh any one of these moving parts changes uh that can materially affect not only your business, uh, but also the entire ecosystem. Sure. Right? And uh so it's very, very complex model. Um and so I love to take a step back and think about like leadership, right? We saw what happened, we saw very clear. quickly a repricing in NVIDIA's stock the second all this news came out about Jensen. And you know, we have certainly been litigating that here all
23:25week and all of us, you know, all over the place. Um and you know, it looks to me uh that you know the market does believe that uh leadership it has an important role, right? Because you know, NVIDIA hasn't exactly bounced back yet. Now we can talk more about NVIDIA in a second, but I'd love to talk about you know, Dario's what what do we think? Like, you know, he's you know, running a startup, as we've been saying, is very different than running a company that is a public company. It takes a different kind of skill. However, this is not an ordinary startup. Obviously, there's so many moving parts the financing, the technology, the lobbying that it took to get them from zero to hero, all the stuff that we've been talking about. First of all, um, do we believe that that he can do it? Or is this something that we think that, you know, this is the guy, or is there gonna have to be another, is there gonna be
24:17Unidentified speaker (UNKNOWN): a
24:17Unidentified speaker (speaker_0): leadership change?
24:18Unidentified speaker (UNKNOWN): Yes.
24:19Unidentified speaker (speaker_0): What do you think? Can I can I just want to see uh get your thoughts uh first, Nancy? What are your thoughts on
24:24Unidentified speaker (UNKNOWN): that?
24:25Unidentified speaker (speaker_1): Oh, um
24:26Unidentified speaker (UNKNOWN): I
24:27Unidentified speaker (speaker_1): always think it's interesting when
24:28Unidentified speaker (UNKNOWN): those of us who I mean I run a little company, but don't run companies uh opine on on whether or not Elon Musk knows what he's doing or Daria Mding. I I I think that um he'll he'll be just fine. I mean, I haven't agreed with everything that he has has done or said. He's
24:46Unidentified speaker (speaker_1): clearly
24:47Unidentified speaker (UNKNOWN): adopted his message uh or adapted his message. away from um existential threat to humanity
24:55Unidentified speaker (speaker_1): to
24:55Unidentified speaker (speaker_3): uh hey we're gonna figure this out and we're gonna keep it safe. Uh I yeah I I don't I'm not concerned about that. I think founder leaders uh can can succeed for a long time with lots of advice and internal um you know jockeying with him and his sister.
25:16Unidentified speaker (speaker_2): I agree with you. I'm not I'm not like horrified about the future of humanity. I mean we did have Y2K and we're still sitting here. But like I I I think that there's a lot of good that can come out of a few um items of uh that are interrelated items of change. Um I don't think that change is happening today or tomorrow at Anthropic Enthropic or OpenAI. Dario's rank running the company, and he's obviously quite a mind, okay? And it's quite a technologist. But one item of change is that if you look at not all, but a lot of extremely successful technology companies that have become behemoths, that have stood the test of time, ultimately became, Uh ultimately had leaders at some point that were not the original technologists, technologist, but that were great operators. If anthropic um decides and and OpenAI cooperates and they all slow down and stop competing so fiercely and they slow things down, then you can get to that second item of change, which is financial discipline. Because then what happens is that the revenue still grows. I don't know at what rate, but the revenue still grows. But you probably get
26:23great leverage over your operating expenses, because we're already starting to see. that see Anthropic head in that direction. And then by the way, then you generate cash because they're actually not that far from generating cash. And so some financial discipline, some good operators, it you still need Dario to be helping. I'm not calling for him not to be CEO tomorrow. That's not what I'm talking about. But you start generating cash, and and by the way, then the whole their whole liability. liabilities, their lease liabilities don't look so scary. And so in my opinion, a little bit of this financial discipline and leadership evolution is a healthy thing. Oh,
27:01Unidentified speaker (speaker_0): it's it's a Nessa.
27:03Unidentified speaker (UNKNOWN): Well they are
27:03Unidentified speaker (speaker_3): generating cash, right? They generated about four billion last
27:06Unidentified speaker (UNKNOWN): quarter alone. They're just spending twice as much. And this and this is
27:08Unidentified speaker (speaker_3): a brand new company. Look, yeah, this I uh uh I d in defense of a founder, right? I don't know what you're you doing. were in twenty twenty one, but I didn't think I should go hang out in a park down here near me with people I know to start this company. And and the the the vision and the skill that he has shown to take an idea with a bunch of people sitting around in a park to create one of the largest businesses in in the in the fastest growing business in the history of the world shows that he's the guy to do it right now because no one else has done that and everyone else did want to do that. And I think it pr it's proof that he's a he's the right manager for the moment.
27:42Unidentified speaker (speaker_2): Certainly not calling for him to to leave he should be CEO right now. Again,
27:46Unidentified speaker (speaker_0): I think the bigger
27:47Unidentified speaker (UNKNOWN): question we should be asking is should Sam Altman be running OpenAI?
27:52Unidentified speaker (speaker_1): Why is
27:52Unidentified speaker (UNKNOWN): that Well I think, you know, here's a guy who lied to his board, got fired, got brought back, has wandered around the AI trade in terms of well, we're going gonna to get into retail, we're going to gonna get into pornography. Um and uh you know has a
28:06Unidentified speaker (speaker_2): a mantra at the company that says focus. I mean if I if I have to pick, I I know which one I'm gonna bet on. And I I think you gotta let the guy run his company. As as Corey rightly points out, it's the fastest growing company we've ever seen. Less than six years revenue grew twelve 12x X year over year last quarter. I I think he's doing fine.
28:30Unidentified speaker (speaker_1): Yeah, sometimes these people step away of their own accord. Eric Schmidt came into Google early to be the steady hand and the adult voice as those two founders m retreated to, you know, working on science projects. And it's easy to imagine Dario sort of saying, I'm going to focus on the technology development of this. training AI models, building the next versions of things, and let me have a financial manager run things. And to your point, Jacob, sometimes those things happen when there's an event in the business model. For example, since it's my favorite theme, pricing. If pricing changes, if pricing power starts to slip from this company, you could have a manager who comes in and talks a lot about leverage, costs, cost discipline. as the technology people retreat to we're going to focus on great new product, and there's a there's sort of a buy model here, a hybrid model. One person's the the the adult in the room who's financial and the other person is continues to build you know amazing tools.
29:30Unidentified speaker (UNKNOWN): What you just said,
29:31Unidentified speaker (speaker_2): in my opinion, is key. Because I and correct me if I'm wrong, but part of what you just said is that for some of this, let's not call it change. but evolution in terms of how the company operates, sometimes you need a catalyst for that evolution.
29:43Unidentified speaker (UNKNOWN): Sometimes you need a catalyst.
29:44Unidentified speaker (speaker_2): And what that means, if you need a catalyst for that evolution, because everybody, nobody, no employee, no other executive, no shareholder has the money to become an activist. That's I that's I'm just being c crazy for a second. That's not the same. What I'm saying is nobody wants to remove the visionary. Yeah. And Tesla, nobody wants to remove Musk. So what my point is, is if there's some degree of evolution in terms of how the company is operating, financial discipline, um, you know, flipping to ki cash flow positive for real, um, there needs to be a catalyst, which means, which means there would have to be an event that causes stock prices
30:21Unidentified speaker (UNKNOWN): to be able to do that. Sometimes
30:22Unidentified speaker (speaker_1): there sometimes there's a sea change. And it's
30:24Unidentified speaker (UNKNOWN): which
30:24Unidentified speaker (speaker_2): also means that this is probably not. an amazing entry point.
30:28Unidentified speaker (speaker_1): Well, there could be an event for the stock, or there could just be an astute management that says, you know what, this is going to be a different profile of growth. And so we need someone who's going to kind of talk to the street about profitability and about margins and about costs.
30:41Unidentified speaker (UNKNOWN): Yeah,
30:42Unidentified speaker (speaker_0): I want to I want to talk about that for a second. You know, obviously, again, running a private company is very different than running a public company. Granted, again, it's a it was an a very large and fast-growing private company, but there's this reality. Now you're answering to different uh many masters, uh, namely the market. Um and uh you know that reprices as I see it every millisecond right over your shoulder. Right. Um where
31:07Unidentified speaker (UNKNOWN): can't see it, Mark. What's it doing? Yeah,
31:08Unidentified speaker (speaker_0): it's at 96 uh 96 uh 9634.
31:12Unidentified speaker (UNKNOWN): Yeah, yeah.
31:12Unidentified speaker (speaker_0): It's down a little bit, you know. But the point is and is that, you know, look, uh valuation, obviously very, very important, but valuation in the private market is basically a lot of people sit around and say, what do you think it's worth? I don't know, what do you think it's worth? And you know, then let's have lunch, right? And they all agree. And so nobody in that group is incentive to have that price go down, that valuation go down. So they all make sure it just keeps going up. You know, sometimes it goes up faster than others, uh, but now you're answering. to everybody, right?
31:39Unidentified speaker (speaker_2): Meaning in the private market, you have to have one private investor saying, if I buy, will you put in money later on this? I'll put in money later. Yeah.
31:46Unidentified speaker (UNKNOWN): And then you and now you go public and
31:49Unidentified speaker (speaker_0): very, very different story. So there's gonna be this reality, right? So the ability to say, you know, I'm I'm gonna I'm leaning into this like this sort of the softer side that we've heard, this and the this this um this predilection with the word um with the word uh existential threat. Like we're the softer we're the softer AI company, right? Um and that seems to be very much a Dario thing. Um and you know you could certainly do that uh in a private company.
32:17Unidentified speaker (speaker_1): You're saying Dario is verbiage about we care about the existential thing.
32:21Unidentified speaker (speaker_0): Yeah, so as a as an investor, right, in a company like that, I want a a ruthless leader that that's gonna be worried about the financials, worried about profit, worried about that growth, not necessarily somebody who's saying, well, we're all we're gonna try to make everybody happy, we're gonna play with our with with our competition, uh, we're gonna not sell to mean people or something like that. So to me, that as a shareholder um is uh is kind of just
32:49Unidentified speaker (UNKNOWN): You're
32:49Unidentified speaker (speaker_1): worried about being too soft.
32:51Unidentified speaker (speaker_0): I'm not saying he's soft, but I'm saying you can say that stuff in a private market. In the public market, you want a person in there that's an operator that is going to be focused on the margins, which are clearly and we're we'll get into that too, the margins are going to be under pressure. You hinted at it several times, but I would
33:06Unidentified speaker (speaker_3): I totally disagree. Not every public company has to sell cigarettes or weapons. And some companies can make decisions about the businesses they want to be and to Nancy's point about pornography, right? Some companies can make a decision, you know, there's maybe more money in selling stuff like that, and we're not going to do it, we're a different kind of company. And that does that's not the sign of a private or a public company. I do think that this company, uh and I and I I generally don't take people to their word of these things, but I think they are focused on a notion of public good in ways that other companies aren't. And I don't, you know, I don't I don't think that's a bad thing in any way. And I don't think that's an unprofitable. It is clearly not an unprofitable decision. It's also It clear clearly has not hindered their growth. Secondarily, I think you can look at what uh XAI, now we call it Grok, whatever, when that when that business was failing as a large language model, only then did they start to allow for more curse words and more p pornography and turn to that because nothing else was working and guess what that didn't work either.
33:58Unidentified speaker (speaker_1): Let me ask since we've had a week of outages that no one can fully explain. what is the catastrophic risk that I hear is potentially there for anthropic and others, not from extinction events, but from agentic AI that goes haywire and ruins things, which is not Bendario's concern. He's been concerned about being beneficent toward humanity, but as I understand it, the actual risk to your point, Mark, that comes up in the market is an insurer is not fully modeled. What would be the premium to insure against massive systems outages and other forms of destruction that are not intentional and maybe not malicious, but that simply happened because it's a new technology. And when does that get priced into the stock? Where, again, to your point, the you know, the serious conversation is not about, oh, the notion notions of being a good, you know, uh uh a good steward and a kind steward, or are you in pornography or not? Have things happened in the market that are destructive that uh insurers then say, oh, this is a big deal.
35:02Unidentified speaker (UNKNOWN): Well,
35:03Unidentified speaker (speaker_3): and we saw that this week, right, with the with Attorney General of California suing OpenAI for letting their agents run amok and saying that they have a warranty risk, right? That because they allowed these things to happen and caused financial harm to other companies, they have a financial responsibility to make good of it.
35:18Unidentified speaker (speaker_1): Without malicious intent and without catastrophic human damage, simply
35:22Unidentified speaker (UNKNOWN): this stuff
35:22Unidentified speaker (speaker_1): is this stuff is a problem. Yeah.
35:25Unidentified speaker (UNKNOWN): I
35:26Unidentified speaker (speaker_1): don't know. I don't know what's modeled and I don't know what can be modeled by insurance insurers and then by the company in its statement of risk in Safe Harbor to say, you know, we have the best of intentions, but actually this is the
35:37Unidentified speaker (UNKNOWN): same thing. In the
35:38Unidentified speaker (speaker_3): S one they say that they don't they can't get insurance for this or that they haven't gotten insurance for
35:42Unidentified speaker (UNKNOWN): the money.
35:44Unidentified speaker (speaker_2): By the way, does anybody think the amended S one and the original S one was a little too long? I know that there's a lot going on with Anthropic. It's a massive company.
35:51Unidentified speaker (UNKNOWN): Oh let's talk. And
35:52Unidentified speaker (speaker_2): they and they I I I understand that it's it's it's a massive company that on a revenue basis again is very soon going gonna to be at over $100 a hundred billion dollars on an annual run weight. It's worth two trillion dollars. They owe it to the to the to the public to to have this huge document and explain a lot of stuff. But actually the S one was, in my opinion, as noisy as the whole public debate is
36:14Unidentified speaker (UNKNOWN): about
36:14Unidentified speaker (speaker_2): stock. Yeah. So like and there there were some questions unanswered. So like I felt like the S one was a little noisy and a little long and not so succinct.
36:22Unidentified speaker (UNKNOWN): But you
36:23Unidentified speaker (speaker_1): could use AI
36:23Unidentified speaker (speaker_3): to
36:23Unidentified speaker (UNKNOWN): too. do that. And it left out the
36:25Unidentified speaker (speaker_3): initial S S1. one Yeah.
36:27Unidentified speaker (UNKNOWN): So
36:27Unidentified speaker (speaker_3): let's talk. Let's talk. So one of the reasons we launched the Quiod Research Product is because there's the quiet period so much shorter. Facebook went public from the time of the S1 for Facebook to the Facebook book IPO was 107 days. Alibaba was 138 days. SpaceX was 23 days. That is There's not enough time to fully understand these businesses for most of the big funds out
36:49Unidentified speaker (UNKNOWN): there that
36:49Unidentified speaker (speaker_3): want to do the work.
36:50Unidentified speaker (speaker_2): I don't. I understand it
36:51Unidentified speaker (UNKNOWN): well, but not exactly
36:52Unidentified speaker (speaker_2): well.
36:53Unidentified speaker (speaker_3): When the S one came out for Anthropic, it left the commitments the the biggest question, right? How much money have they committed to build data centers? And what are those commitments and when are they due? That section was
37:03Unidentified speaker (UNKNOWN): completely
37:04Unidentified speaker (speaker_3): blank for the first week, which gave us only 14 days to try to figure that stuff out. And that's I uh to me that's insufficient and that's a failure of the uh the SEC's rules around these. The new rules around IPOs.
37:17Unidentified speaker (speaker_0): Did you see anything in the uh in the S one that like was was negative? Like the actually that you that we didn't already know that you're like, uh I don't like this. This is something new, or is it kind of like that everyone just like, oh, this is what we expected it to be? Trevor
37:31Unidentified speaker (UNKNOWN): Burrus, Jr.
37:31Unidentified speaker (speaker_1): I saw the absence of a discussion of pricing. No discussion of pricing.
37:36Unidentified speaker (UNKNOWN): Just
37:37Unidentified speaker (speaker_1): discussion of price.
37:38Unidentified speaker (speaker_3): Wait,
37:38Unidentified speaker (UNKNOWN): I want
37:39Unidentified speaker (speaker_2): to hear
37:39Unidentified speaker (UNKNOWN): the thing.
37:42Unidentified speaker (speaker_2): Go go ahead, Corey. Yeah.
37:44Unidentified speaker (UNKNOWN): No,
37:44Unidentified speaker (speaker_3): I think I think there is a lot of discussion, the S1, about um cutting price on models, their their um approach to pricing. Whether we can't we can't possibly get enough.
37:53Unidentified speaker (speaker_1): That's their intention, but there's nothing in detail about what they're actually charging, for example, for their unit of work.
37:59Unidentified speaker (speaker_3): For each token. Yeah, absolutely. Which is
38:01Unidentified speaker (UNKNOWN): their cost
38:01Unidentified speaker (speaker_1): of goods. And cost of goods. is the whole server complex they build, some of which is used internally for training. And so there isn't really a incremental cost for the cost to serve tokens from them that you would say.
38:14Unidentified speaker (UNKNOWN): Yeah, they're
38:14Unidentified speaker (speaker_3): they're report. So their gross margin reporting is we think it's going to be very different than what you see from OpenAI. And essentially that great mostly impacts uh their wholesalers. So when an Amazon or a Google, which are their two biggest customers, when Amazon goes out and sells the use of Anthropic to their existing AWS customers, Anthropic counts all of that as revenues and then takes a cost of goods. So it makes the revenue number look bigger. And when that was the fastest growing sales channel, it made the growth look faster than OpenAI. We still think it's a bigger business than OpenAI, OpenAI, but the accounting matters a lot there in terms of calculating what is revenue and what isn't.
38:51Unidentified speaker (UNKNOWN): And
38:51Unidentified speaker (speaker_1): it's an omission to your point, Mark.
38:52Unidentified speaker (UNKNOWN): Well you
38:53Unidentified speaker (speaker_0): know like w so what are what are the what are the like my approach is always like I have to talk myself into a position, right? Everyone has a different approach to things, right? Other people like talk themselves out of position. So I'm always like reading for like, you know, I'm looking for the shadow data. Uh I'm looking for like what are they not disclosing? What's the edge what are the edge you know the edge risks here. Um, you know, what what you know what is what is a what is a worst case scenario? Nancy, are you thinking about that? Like a worst case scenario? Like what could really go what could go like violently wrong uh for anthropic?
39:31Unidentified speaker (speaker_1): Existential threat to humanity, I guess.
39:36Unidentified speaker (UNKNOWN): Um
39:36Unidentified speaker (speaker_1): I I think, you know, the thing that stood out. to me was that the uh inferencing margins were at about 80% in the S in the original S1. And that that was surprising to me. I think those numbers may be at risk as we enter this new um blue world where maybe data centers are bad and they're at least going to be slowed down. That's gonna reduce capacity and increase prices. So potentially. Uh I'm a little bit more cynical about politicians, and I think we probably will see most of, if not all, of the data centers being built because they pay taxes and they hire constituents. But generally, that would be the risk to me that they don't have a good handle on costs at this point, uh, given the exponential nature of the growth.
40:24Unidentified speaker (speaker_0): Yeah, I I would agree. I would agree. Nancy
40:27Unidentified speaker (speaker_2): what about China or anybody? What about China? Because not only do you have the Jensen Huang issue, and then Congress is going to look into that. And by the way, as long as he has a legal um uh overhang and he's not speaking to Congress until December 1st, then he can't come back to NVIDIA Video anytime soon. Okay. And then we also, by the way, this morning and AMD
40:50Unidentified speaker (speaker_0): we
40:50Unidentified speaker (UNKNOWN): got new
40:50Unidentified speaker (speaker_2): news with AMD.
40:51Unidentified speaker (UNKNOWN): And D. So
40:52Unidentified speaker (speaker_2): here's my question. Ouch.
40:53Unidentified speaker (UNKNOWN): You know,
40:53Unidentified speaker (speaker_2): are these are we talking about China cockroaches now? Like it coming out like the way we had with private credit. So one of my questions, Nancy, it seems like you've done like a ton of work on anthropic or anybody else, but like one of my questions is um is is competition from China a um a a a risk um to to US tech companies as and their ability to push forward?
41:16Unidentified speaker (speaker_3): No.
41:18Unidentified speaker (UNKNOWN): No,
41:18Unidentified speaker (speaker_3): it's not.
41:19Unidentified speaker (speaker_1): No, the technology that Anthropic has developed is developed also by many labs. Anthropic is the richest at the moment, but it's developed through labs around the world. DeepSeek AI, the big Chinese started from a year ago, contributes things. By diffusion, they all borrow from each other. The technologies are out there. So the large language model technology underlying this is more common throughout the world than anthropic or OpenAI would have you believe, because they have these private models that they gatekeep. The technology is broadly available, and what's going to happen, I think, is the semiconductors, which are the heart of it, will move beyond NVIDIA GPUs because of their crisis, and they will move to AMD, they will move to all of Broadcom's partners, the Google TPU, the Amazon Trainum and Inferentia chips, the Meta chip, the Jalapeno chip that Broadcom's helping to build at OpenAI. Everyone will have semiconductors. The substrate will broaden out for compute. And that actually could it be a good thing for an anthropic if the whole cosmos of chips is not just just Jensen
42:26Unidentified speaker (UNKNOWN): a chip.
42:27Unidentified speaker (speaker_1): Huang and his GPUs and all of his designs, which command a premium, but many, many chip vendors that actually now provide cerebro systems, now provide a broadening substrate of compute, bring down the costs for inference. Everyone could benefit. And Chinese models will be contributing to the sauce, and open source models will be contributing, and there'll be an opening. a flowering of what has so far been: let's everyone go to NVIDIA and beg them for chips. Elon Musk standardizing on NVIDIA because he wants to get chips, and so he's making a commitment. You have to make a commitment. If it broadens out, everyone can relax, take a deep breath, and say the technologies will be broadly available, and then people can compete in different ways.
43:12Unidentified speaker (speaker_0): Question is is when? Like when when is I mean
43:14Unidentified speaker (UNKNOWN): we all
43:14Unidentified speaker (speaker_0): agree that competition you think? When do
43:16Unidentified speaker (UNKNOWN): you think?
43:16Unidentified speaker (speaker_0): Like when you talk about uh next year or 24. Yeah,
43:19Unidentified speaker (UNKNOWN): so I think it
43:20Unidentified speaker (speaker_1): starts now.
43:20Unidentified speaker (speaker_0): So my the best thing I have been able to do is say, okay, I can give myself a runway where I know that Nvidia is gonna enjoy their position of power, right? So I'm a I'm a fan of it. Uh and I believe that that they will they they have a head start, they still have a head start. They're still head and shoulders over all those other models, and they're not just sitting, they're not just sitting around either. They recognize the competition, and the competition has obviously spurred them to push their technology to the next level.
43:47Unidentified speaker (UNKNOWN): Talking
43:47Unidentified speaker (speaker_1): about NVIDIA.
43:48Unidentified speaker (speaker_0): NVIDIA, okay. And and I think the same thing with competition, right? When when this when the DeepSeek thing came out, I thought that was a good thing, right? Because it basically just spurred competition. And then when you started to see the, you know, the labs starting to compete with each other and recognize that there is competition, you saw that their pace of innovation picked up massively, right? One person makes the next step forward, they start with the agents and the other guy has to have
44:12Unidentified speaker (UNKNOWN): a lot of people.
44:14Unidentified speaker (speaker_0): Absolutely. And I, you know, but the point is they still enjoy that position right now, uh, at least for the next, I was always assuming uh, let's call it 18 months. Right? So you're saying it's not 18 months, you think it's
44:26Unidentified speaker (UNKNOWN): a good idea. happening. It's now.
44:27Unidentified speaker (speaker_1): It's happening now.
44:28Unidentified speaker (speaker_3): Competition is happening now. I was at uh Oracle's Investor Day and they talked about new uh semiconductor products that they're bringing into the into the uh data centers. And in particular, they talked about inference focused chips from Sarah Bros and from Positron, which I was very happy to see because I'm an investor in that scene round, and that's been a good one for us so far in paper. But um uh I you know we're definitely seeing some competition show up, uh but we're not seeing
44:50Unidentified speaker (UNKNOWN): the company.
44:51Unidentified speaker (speaker_3): competition and where they're being manufactured, right? We're there we're still seeing Taiwan semiconductor get most of that and it's capacity that is being diverted away in some ways away from NVIDIA, which would like to see more, and frankly, away from others who would like to see that.
45:02Unidentified speaker (UNKNOWN): Can I just
45:03Unidentified speaker (speaker_2): say real quick about some of these chip names? Um, you know, with you saw a moment when the Huang news broke where NVIDIA went down, AMD Broadcom went up, you saw the crisscross. And then that's that stopped. And then they became basically correlated again for a few days. I don't know why. Maybe it's if you lose the mind that is Huang, then you lose some starting point of tech advancement. I don't know. But the point is that if you what you guys are describing, you guys are not describing a deteriorating pie for AI chips. You're describing competition. And so if in that case, if TSM is way down, and by the way, we think the shakeout from pre-Huang, the shakeout of chip selling, all that might be close to over, you look at support levels, it's super cheap valuations, basically, assuming the right earnings number, then you buy TSM, which is in the back of the supply chain. And if you do believe that NVIDIA loses a little bit of shine, a little bit of share, you buy AMD, which has always been the bat the robot. the Robin to NVIDIA's Batman.
46:04Unidentified speaker (speaker_1): Yep. And they have the x86 chips for the pre and post stages that you now know we can't do
46:12Unidentified speaker (UNKNOWN): without
46:13Unidentified speaker (speaker_1): Intel and AMD. And it seems generally AMD gains in share and servers, so they might be sort of a little better here for CPUs, but it broadens out because that's the work that has to get done. to make stuff work in enterprises throughout the world is all the things to tie it into real applications and not simply running massive GPU farms for more tokens endlessly. Yeah.
46:38Unidentified speaker (speaker_0): Yeah. Yeah.
46:39Unidentified speaker (UNKNOWN): So
46:39Unidentified speaker (speaker_3): since the Jensen thing, I've got a, I've got a uh I want to run this by you. I haven't written this yet, but I've been working with this idea on why the NVIDIA scandal is good news for anthropics, bullish for anthropic. And I love your takes on this and give me more information before I write my piece. Easier or harder?
47:04Unidentified speaker (UNKNOWN): For
47:04Unidentified speaker (speaker_1): NVIDIA?
47:05Unidentified speaker (speaker_3): Anyone. For anybody. No,
47:07Unidentified speaker (UNKNOWN): it's a good idea. Say it again, Corey. Say it one Corey. Say it
47:08Unidentified speaker (speaker_0): one more time.
47:09Unidentified speaker (speaker_3): Is it is is it easier or harder going forward to sell the best GPUs into China? Harder. It's hard. Okay. So so if China needed those GPUs and what was doing now we now know so much on the spy craft world to steal these chips, to import these chips, to access these chips, to train the models that we thought was the biggest competition to anthropic, that is the open weight models in China, if they needed the the GPUs that they will not be able to get in the future. Right. or be able to get more difficultly in the future, that means that those models will be worse in the future. And that's gotta be good news, bullish for anthropic and OpenAI, if the cutting edge models will continue to come from those two labs and not from their competition in China because they can't get the chips.
47:55Unidentified speaker (UNKNOWN): Well,
47:55Unidentified speaker (speaker_2): by the way, Corey, the what part of what I think you're describing is a continuation of the same theme that I think we'll see for a for a while from now. which is that chipmakers' gain is the chip buyer's loss, and vice versa. Um, like if you have cheaper chips or like fewer chips being bought or the data center adjacent companies, you know, take a hit in terms of estimates, then you know that you get some bidding up of software and anthropics, the same theme.
48:20Unidentified speaker (speaker_1): If you have more chip availability because NVIDIA doesn't have a stranglehold on the designs, a thousand flowers will bloom and you will get people being able to build more on AMD GPUs, on all the Broadcom partners, Google TPU, Amazon Trainium, Meta's prog uh chip, Microsoft's chip, OpenAI's chip, everyone, Oracle's And own dedicated chip.
48:47Unidentified speaker (UNKNOWN): Did you
48:47Unidentified speaker (speaker_3): really quote Mao?
48:49Unidentified speaker (UNKNOWN): That was
48:49Unidentified speaker (speaker_3): perfect.
48:50Unidentified speaker (speaker_1): Yeah.
48:51Unidentified speaker (speaker_3): Well done.
48:52Unidentified speaker (speaker_1): So everyone will
48:53Unidentified speaker (UNKNOWN): be
48:53Unidentified speaker (speaker_1): everyone will be able to build more. And so, Corey, I think it's a wash because if China has somewhat of a bottleneck, uh all these other places will be able to build without sort of begging Jensen
49:08Unidentified speaker (UNKNOWN): for the use.
49:10Unidentified speaker (speaker_3): They're going to build them for OpenAI and Anthropic because they put them to best use, not the Chinese competition that may never really emerge as real competition because they can't get the chips. Trevor
49:19Unidentified speaker (UNKNOWN): Burrus, Jr. No,
49:20Unidentified speaker (speaker_1): the broadening of the market will serve all open source developers and all new frontier labs, and they will build things that don't
49:28Unidentified speaker (UNKNOWN): again
49:28Unidentified speaker (speaker_1): need to be
49:29Unidentified speaker (speaker_3): they can't build it if they can't get the chips.
49:31Unidentified speaker (UNKNOWN): They
49:31Unidentified speaker (speaker_1): can build it if it doesn't need to be the biggest collection of chips because it's trying to fuel a frontier model that has gamed the benchmarks, claiming bigger is better. Anthropic is the 1970s Cadillac going up against the oil crisis. They have built on getting the most money to get the most chips the quickest ahead of everyone else and rush to market telling you you cannot do anything but the best frontier model. And I think that's
50:01Unidentified speaker (UNKNOWN): the same thing. a very important thing. Well, every cost of
50:02Unidentified speaker (speaker_3): compute says the opposite of that. Their cost of compute is going down, going down, going down every quarter.
50:08Unidentified speaker (UNKNOWN): Their cost of
50:08Unidentified speaker (speaker_1): compute is going down relative to what they started from. Everyone else, everyone else
50:14Unidentified speaker (speaker_3): As opposed to what?
50:15Unidentified speaker (speaker_1): As opposed to everyone else going out and building something that's reasonable to a task, that's well suited to a task, that doesn't have to be the 10X model. That's what you can have when you have a flowering of chips and you have semiconductors that aren't have don't have to be built to the largest reference program that was dictated by NVIDIA because if you want our chips, you got to party with us.
50:38Unidentified speaker (speaker_0): Speaking of our chips, though, I have to bring it back down to the ground, right? Because you
50:42Unidentified speaker (UNKNOWN): know
50:42Unidentified speaker (speaker_0): I'm staring at that number uh over your shoulder. Um I I want to just quickly uh just talk about because because we're bashing a little bit on NVIDIA. now. So I kind of feel like NVIDIA is kind of down for the wrong reasons uh at this point. Nancy, what are your thoughts on that? I know you and I both like NVIDIA. We're long time uh fans of NVIDIA. I think it's uh down for the wrong reasons. What are your thoughts?
51:07Unidentified speaker (UNKNOWN): Um
51:07Unidentified speaker (speaker_1): yeah well I guess any r it's down, right, Mark? So um the market's always right. Uh that was lesson lesson one. Um yeah, I I think in here um it it makes sense. I mean we're gonna be adding to it. It's just a question of is it this week or next week. But you know I look at the the overall picture we can debate whether or not you know competition's gonna put a dent and it it it should. It always does. I mean just use the coach purse personality analogy where the margins got so big uh that competitors came in and then, you know, coach stumbled on design and, you know, um cannibalized their own business by doing uh what they did with uh outlets and the and the business was was moribund for a decade. So that could happen. Uh I just don't think we're there yet. And I think the company has um a pretty big cushion in terms of sharing. share buybacks. You know, they they authorized two hundred and thirty billion, they've spent nineteen, now they've authorized another hundred and fifty billion, uh which I suspect has
52:12Unidentified speaker (UNKNOWN): been to do.
52:12Unidentified speaker (speaker_0): Nancy, I hate to cut you off here, but unfortunately we're gonna have to hold that thought. We're gonna have to go to a commercial break, but we're gonna come back in a few. Uh I don't know if Nancy you'll be able to join us after, but I wanna want to thank you for being part of this conversation. Yeah,
52:26Unidentified speaker (speaker_1): my pleasure.
52:27Unidentified speaker (UNKNOWN): And that
52:27Unidentified speaker (speaker_3): green jacket without winning Augusta.
52:29Unidentified speaker (speaker_0): Yeah, yeah. We'll be back in a few guys. We're back. And now we have Julie with us. Nancy had a leave us, but uh Julie has been working double duty today. Uh f triple duty following this from the very beginning. Uh and at one point we were just talking, even playing uh Grand Theft with Auto for all of us. And that's obviously the other news.
52:52Unidentified speaker (speaker_1): IPOs and GTA's together. Biggest IPO today, Julie's first time playing Grand Theft Auto. Which is more exciting. I mean the Grand Theft Auto,
53:01Unidentified speaker (UNKNOWN): obviously.
53:02Unidentified speaker (speaker_1): For
53:02Unidentified speaker (UNKNOWN): me, probably
53:02Unidentified speaker (speaker_1): like just on a personal level. Yes. Exactly. Because I'm not trading the IPO. I'm just talking about it. But it which is also fun.
53:09Unidentified speaker (speaker_0): And
53:09Unidentified speaker (speaker_3): no green jacket, Jules. jewels. No. Nancy's Nancy took it to another level there. here.
53:13Unidentified speaker (speaker_1): As she always does.
53:15Unidentified speaker (UNKNOWN): As
53:15Unidentified speaker (speaker_0): she always does. I mean
53:16Unidentified speaker (speaker_1): the shoes, which
53:16Unidentified speaker (UNKNOWN): weren't on camera. We didn't get to see the shoes.
53:21Unidentified speaker (speaker_0): Your shoes are on point today, too.
53:23Unidentified speaker (UNKNOWN): Those
53:23Unidentified speaker (speaker_0): are pretty cool. I have to say. Maybe we could uh we could zoom out a little bit um and talk more about uh where anthropic fits um in the bigger economy, right? Um you know, we have been talking all about getting genius
53:38Unidentified speaker (UNKNOWN): the
53:39Unidentified speaker (speaker_0): next level models and and uh all these um
53:43Unidentified speaker (UNKNOWN): uh
53:43Unidentified speaker (speaker_0): you know these these nebulous thoughts. Let's let's get to reality here, right? I I think I saw I like I saw a couple of um a couple of interesting uh I think I saw a Sazi interview uh which was pretty cool with Gunlock, and he was just like, Oh, you know, look at the balance sheet of this company. You know, he made it sound like, you know. my eyebrows went up because that stuff always gets me uh that always gets me I started as a bond person so I'm always looking at interest rates and trying to figure out what what the impacts are. We haven't really talked about that with respect to anthropic a lot yet. I think that was the first the
54:20Unidentified speaker (UNKNOWN): balance
54:20Unidentified speaker (speaker_3): sheet.
54:20Unidentified speaker (UNKNOWN): The
54:21Unidentified speaker (speaker_0): balance sheet right Okay. in terms of the impact of of interest rates and how you know obviously interest rates are tied to the broader economy. And so that is another thing, yet another variable that is not in you know under the control uh of its neophyte leader uh that uh they're they're gonna have to manage. So wha how you know, it used to be that like just about a week ago, we were talking about um NVIDIA being at the center of that of that solar system, the sun of that solar system. right? And uh and so how we have all recently learned how important interest rates are, right? Because they're the bank of NVIDIA. Uh but now it looks like Anthropic is taking over that new top triangle uh uh that's gonna be out there and gonna be, you know, sort of how how interest rates move, how the economy performs, all that stuff, I think that can impact Anthropic's business.
55:15Unidentified speaker (UNKNOWN): I'm
55:15Unidentified speaker (speaker_2): concerned. I I'm concerned. And I realize it's really easy for me to say that I'm concerned because I sit here and some of you guys have money at stake. I and I and I don't, I'm my 401k, but I you know, I'm writing about people's money, so it's easy to say. But I have to say, when you when you look at the amended S1 and you look at the balance sheet, it does not appear if they continue to burn a little bit of cash. I know Corey has something to say about cash burn, but if you can turn if you continue
55:43Unidentified speaker (UNKNOWN): to burn
55:43Unidentified speaker (speaker_2): a little bit of cash.
55:43Unidentified speaker (UNKNOWN): They're not burning cash, but go on.
55:45Unidentified speaker (speaker_2): If you continue to burn burn a little bit of cash and and and that continues, I don't think they have 10 years of cash on the balance sheet. They don't. They have a few years.
55:56Unidentified speaker (speaker_1): No, but they have revenue
55:57Unidentified speaker (UNKNOWN): that's coming.
55:59Unidentified speaker (speaker_2): But here's the problem. But my problem with that is that if rates become problematic or if they go, if they burst out above this technical range that they're in, tenure yield getting resistance at 5.3%, if that happens and then there's a um a spiraling effect in terms of financing, which is, you know, which is the beginning of a problem, then Anthropic has a bit of an issue. This is something that nobody really wants to talk about. right now. They don't have 10 years of cash on their balance sheet.
56:30Unidentified speaker (speaker_1): I don't think it's a problem for Anthropic. And I don't think it's a problem for Alphabet or Amazon. Maybe it's a problem for Core Weave or
56:37Unidentified speaker (UNKNOWN): some of
56:37Unidentified speaker (speaker_1): those that are further down the stack.
56:39Unidentified speaker (speaker_0): Agree.
56:39Unidentified speaker (UNKNOWN): I
56:39Unidentified speaker (speaker_1): don't th I mean I don't know. Corey, sorry. I I stepped on you.
56:43Unidentified speaker (speaker_3): No, no, no. I I'd rather hear you than hear my own voice. Uh look, Anthropic at this moment has at least $230 billion on the balance sheet net of debt. They have the best balance sheet of any company in technology. I think of any company period. If you took all the cash that all the hyperscalers ever had at the all-time peak, which is about 2020, and added it all up, Anthropic has more today after this IPO. And the company is generating billions of dollars of cash every quarter. So they have more than enough money to get them through the next few years, not the next twenty or thirty. But
57:21Unidentified speaker (UNKNOWN): what do you count as they don't need
57:22Unidentified speaker (speaker_3): enough money right now to get through the next
57:23Unidentified speaker (UNKNOWN): one? They need to
57:25Unidentified speaker (speaker_3): get through now.
57:25Unidentified speaker (speaker_2): But my question is you're talking about net cash, but what do you count as debt?
57:29Unidentified speaker (speaker_3): That's after debt. Yeah,
57:30Unidentified speaker (UNKNOWN): but there is no debt.
57:31Unidentified speaker (speaker_2): Financial debt or like all the lease obligations.
57:33Unidentified speaker (UNKNOWN): You mean
57:33Unidentified speaker (speaker_3): commitments?
57:34Unidentified speaker (UNKNOWN): Yeah.
57:34Unidentified speaker (speaker_3): Yeah. No, so I don't the the lease the leases aren't debt, so that I don't I
57:38Unidentified speaker (UNKNOWN): don't
57:38Unidentified speaker (speaker_3): count. them as debt. Um
57:40Unidentified speaker (UNKNOWN): they have made
57:40Unidentified speaker (speaker_3): some commitments, some of which so they have some obligations, some of which they're gonna have to fulfill. So they've got less than uh $70 billion in obligations right now with $230 billion on the from the next year on the on the balance sheet right now. They have more than enough cash. If they shut down operations today, they could honor all of their commitments for the next two years plus.
58:00Unidentified speaker (UNKNOWN): And
58:00Unidentified speaker (speaker_3): they're not gonna going to shut
58:00Unidentified speaker (UNKNOWN): down an operations. operating. I don't think what I'm
58:03Unidentified speaker (speaker_2): saying is going to come to fruition. I don't, but I think that it's a risk that where there will be moments where it gets priced in and it will be part of a volatile ride beyond the first few
58:12Unidentified speaker (UNKNOWN): weeks of the year.
58:13Unidentified speaker (speaker_3): It is unquestionably the biggest risk. Are they going to be able to continue to grow the business and generate enough cash to fulfill these obligations to continue to grow the business and generate cash? That is the biggest question here. But what we know today and we didn't know yesterday, we could probably presume it yesterday.
58:39Unidentified speaker (UNKNOWN): I
58:39Unidentified speaker (speaker_2): don't feel that way about the balance sheet because the you know and and well I can't r uh couldn't really hear either you here hear you for a second, but the the lease obligations bother me. And again, I want to be really clear. This is not, I do not think that Anthropic is gonna have a massive problem within the next one or two years or anything like that. But I think the way you have to look at it as an equity investor is what's my potential return versus what's my volatility? And when we get past a little period of volatility from the IPO, the the first few weeks, whatever, I actually think that this picture is going to create continued volatility. and it may and it may mean that anthropic is uh not a great bet relative to other bets in tech that you might have, especially since we have higher rates and we're definitely later cycle, not earlier cycle.
59:27Unidentified speaker (speaker_0): What's gonna be less volatile
59:29Unidentified speaker (speaker_2): than anthropic?
59:30Unidentified speaker (speaker_0): I I mean I you know the volatility of most of these names is relatively high at this point.
59:35Unidentified speaker (speaker_2): What will be less volatile than anthropic in my opinion will be Microsoft, Meta, probably Alphabet, but I like Meta a little bit better than Alphabet because Meta has Muse, and that's that looks, you know, in terms of forecast.
59:47Unidentified speaker (speaker_0): Well, Meta looks like that until Apple comes out with an agent or until
59:51Unidentified speaker (speaker_2): Alphabet
59:52Unidentified speaker (speaker_0): comes out with an approach.
59:53Unidentified speaker (speaker_2): But
59:53Unidentified speaker (UNKNOWN): maybe you buy,
59:53Unidentified speaker (speaker_2): but maybe
59:54Unidentified speaker (UNKNOWN): And it
59:54Unidentified speaker (speaker_0): looks super
59:54Unidentified speaker (UNKNOWN): volatile.
59:55Unidentified speaker (speaker_2): But maybe
59:55Unidentified speaker (UNKNOWN): I
59:56Unidentified speaker (speaker_0): I just think I think that
59:58Unidentified speaker (UNKNOWN): I
59:58Unidentified speaker (speaker_0): I I mean I said this earlier on the show. Like if you tell me what's gonna happen a year from now, you're wrong. Mm-hmm. And so is everybody else. If you tell me what's gonna happen from five years from now, you're really wrong, and so is everybody else.
1:00:09Unidentified speaker (speaker_1): About ten years.
1:00:10Unidentified speaker (speaker_0): Like nobody knows.
1:00:11Unidentified speaker (speaker_2): We
1:00:11Unidentified speaker (UNKNOWN): don't know.
1:00:13Unidentified speaker (speaker_2): Which is why, which is why I focus on risk management, which is why I focus on return versus volatility, because I agree. We have no idea what's gonna happen.
1:00:22Unidentified speaker (UNKNOWN): Well, let
1:00:22Unidentified speaker (speaker_1): me let me let me
1:00:24Unidentified speaker (UNKNOWN): be
1:00:24Unidentified speaker (speaker_1): the a spoiler for that, because I do. know what's gonna happen. What's gonna happen is that the chips that are still the basis of all of this will continue to sell and they will continue to get better and they will continue to bring down the cost of compute. And that's been true for decades, for sixty-some years. And that will be true regardless of who wins the Frontier AI lab. So I would say what's gonna be going to not entirely without volatility, but less volatile will be the semiconductors.
1:00:56Unidentified speaker (UNKNOWN): Okay,
1:00:56Unidentified speaker (speaker_0): so here I have a question for you. Five years from now, will NVIDIA still be the leader?
1:01:01Unidentified speaker (speaker_1): No.
1:01:02Unidentified speaker (UNKNOWN): So
1:01:02Unidentified speaker (speaker_0): like that that's where the uncertainty comes.
1:01:04Unidentified speaker (speaker_1): Because it
1:01:04Unidentified speaker (UNKNOWN): will
1:01:04Unidentified speaker (speaker_1): broaden
1:01:04Unidentified speaker (UNKNOWN): out to
1:01:05Unidentified speaker (speaker_1): more.
1:01:05Unidentified speaker (UNKNOWN): But
1:01:05Unidentified speaker (speaker_0): there will be demand. And maybe there's a or maybe there's a startup chip chipmaker Maker that we don't even we're not even talking about.
1:01:11Unidentified speaker (UNKNOWN): Well, so
1:01:11Unidentified speaker (speaker_1): for example, Cerebra Cerebrus Systems, which is a company that's losing
1:01:14Unidentified speaker (UNKNOWN): money and
1:01:14Unidentified speaker (speaker_1): debt. down since it's IPO, could be in a better position in six months to a year from now. Yeah. And they run inference the fastest, so they say, and they have economies of scale as well. I you know, Corey was teasing me for quoting Chairman Mao, but uh many more things bloom in an environment where uh NVIDIA is not the sole property dominating the reference architecture for all these data centers and holding the key. keys and you have to beg NVIDIA for allocation. If if everyone can buy from many vendors of chips, including the Google TPU, which is now becoming a merchant solution, it will broaden out the opportunities and it will dis it will it will reduce volatility by not having these crazy swings of who's getting allocation, who can go to TSMC, only Jensen Huang and get the most allocation.
1:02:16Unidentified speaker (speaker_0): Oh,
1:02:16Unidentified speaker (UNKNOWN): I
1:02:16Unidentified speaker (speaker_0): think Corey,
1:02:17Unidentified speaker (UNKNOWN): we can hear you for some reason.
1:02:18Unidentified speaker (speaker_3): Oh, sorry. The same of what Tierna said about NVIDIA not being in the dominant pole position that they had been in in the last few years, except that their growth rates actually picked up. We saw a year-over-year sales growth of over 100% in the last quarter. Then that had fallen down to about 70 percent, only 70 percent. But the notion that NVIDIA sales are twice what they were a year ago when our jaws were on the ground with their sales a year ago, um, suggests a re-acceleration Acceleration there. So I, you know, I kind of hope for the sake of of of competition that there is more competition. and prices come down. Um, but we are not seeing that. These
1:02:55Unidentified speaker (UNKNOWN): are the things that are going to be a good thing. Right, I
1:02:56Unidentified speaker (speaker_0): guess that's why I asked
1:02:57Unidentified speaker (speaker_2): five years
1:02:57Unidentified speaker (speaker_0): and not one year out, right? Because these things do have long lead time.
1:03:02Unidentified speaker (speaker_2): But is it really boring of me to say that like instead of trying to pick, let's just it pick winners like we're talking about anthropic and OpenAI, but let's talk about chips for a second, for example. Instead of trying to pick winners, is NVIDIA gonna be back, right? Or is it really gonna lose
1:03:14Unidentified speaker (UNKNOWN): the channel?
1:03:15Unidentified speaker (speaker_2): share to AMD and Broadcom? Instead of trying to pick winners,
1:03:19Unidentified speaker (UNKNOWN): I
1:03:19Unidentified speaker (speaker_2): there's great news. They all went down together at the same time. You saw crisscross with AMD up and Avgo up and NVIDIA down, and then they got correlated. AFCO and the great Avgo. And it's just so easier. And then
1:03:33Unidentified speaker (speaker_1): Avocado.
1:03:33Unidentified speaker (speaker_0): Broadcom. Broadcom
1:03:34Unidentified speaker (UNKNOWN): is the colour. same. Broadcom, sorry.
1:03:35Unidentified speaker (speaker_2): I'm sorry, everybody, Broadcom, but they got correlated again. for a few days. They all went down. And if you're if if you think that there's a good enough chance of more runway for data center growth over the next couple years, why am I being too boring to say buy an equal weighted basket of the especially TSM?
1:03:55Unidentified speaker (UNKNOWN): I
1:03:55Unidentified speaker (speaker_0): mean, look, that that's certainly been my approach, right? It hasn't
1:03:58Unidentified speaker (UNKNOWN): it doesn't,
1:03:58Unidentified speaker (speaker_0): you don't have to pick a winner, right? All these these are all great companies, right? So you can can't
1:04:02Unidentified speaker (UNKNOWN): more fun one
1:04:02Unidentified speaker (speaker_3): to when do. you do though. Yeah,
1:04:03Unidentified speaker (speaker_1): but if you choose the best one.
1:04:04Unidentified speaker (UNKNOWN): I
1:04:04Unidentified speaker (speaker_0): mean yeah that's it there's there's an argument that says too too much of that is not good. But the the truth is they're all they're all very solid. They all have they all have really good I mean the thesis hasn't changed for any of these, for me at least uh and the fundamentals are certainly there. But you know going more on your point, like I think it I think it's easier to say, you know, what's happening in five years, right? Because you can s you we can I think I could say with with uh with confidence that the same the the the same pressure that you're talking about in terms of the chips uh in terms of the price uh prices coming down because of the competition, I think we could say the same thing about Anthropic, right, and its competition, right? So clearly there's pressure on companies like Anthropic and the other labs as they as they come public, right, as other new ones emerge. uh that you know, those price pressures are gonna apply here as well. You
1:04:54Unidentified speaker (UNKNOWN): know
1:04:54Unidentified speaker (speaker_3): what price is coming down, which is kind of interesting. We haven't talked about it is sort of in the weeds about Anthropic, but I think it's really important. They say that the cost of developing their latest large language model, which is their best, and it was the last one before that was the best. They say the cost of developing those is coming way down because they are were able to use their large language model, their last one, to develop the next one. So it's happening more quickly, more cheaply, which it would say they're developing it uh using less computational power. And so that the training of their large language models now, even though they're so much better than ever before, is so much less than it used to be. Which says to me, and it says in my Excel spreadsheets about the my model, says that their cost of compute will continue to fall, not just because they're going from training to inference. but because the training itself is getting so much cheaper. And I I think that that's really bullish for the giant LLMs right now, of which of course Anthropic and OpenAI are the are n are, you know, uh number one and number one A. And I think that that's really positive for them because it suggests that once they get just like Metcalf's law was true for networking, that once you get a lead, you keep a lead. in a world of AI.
1:06:04Unidentified speaker (UNKNOWN): Kiernan, are you not able to hear
1:06:06Unidentified speaker (speaker_1): Corey? He's
1:06:07Unidentified speaker (UNKNOWN): saying that the cost for them to develop their most advanced models are going down.
1:06:11Unidentified speaker (speaker_1): Bless your heart. I was sitting here like the grandpa who gets excluded from the conversation. Which I am kind of rapidly cruising
1:06:19Unidentified speaker (UNKNOWN): toward
1:06:19Unidentified speaker (speaker_1): it.
1:06:20Unidentified speaker (UNKNOWN): Thank
1:06:20Unidentified speaker (speaker_1): you, Dick.
1:06:22Unidentified speaker (speaker_0): Um I'm also curious about when like we had a wave earlier this year. where Mike Wilson of Morion Stanley came out and said, now is the time to own the beneficiaries, right? We have been so laser focused on the supply chain, the hyperscalers,
1:06:40Unidentified speaker (UNKNOWN): the
1:06:41Unidentified speaker (speaker_0): frontier labs. Um eventually all of this should help everything.
1:06:46Unidentified speaker (speaker_1): The beneficiaries. Right. So who are those two? All
1:06:49Unidentified speaker (UNKNOWN): of us.
1:06:50Unidentified speaker (speaker_0): Healthcare companies,
1:06:51Unidentified speaker (speaker_1): industrial. Normal companies. Everything.
1:06:54Unidentified speaker (speaker_2): I have
1:06:54Unidentified speaker (UNKNOWN): really good
1:06:55Unidentified speaker (speaker_2): news about that. I have
1:06:56Unidentified speaker (UNKNOWN): great news about that.
1:06:57Unidentified speaker (speaker_0): So
1:06:57Unidentified speaker (speaker_2): I would posit
1:06:59Unidentified speaker (speaker_0): Do you want to buy an equal weight basket of the semiconductors, or do you just want to buy SPW and like buy the whole SP equal weight at some point, at least as it I I don't know. I'm just I'm just putting it out. Like at so s when do you is it t that maybe that Mike Wilson called? call was too early, I don't know. But like at some point that's gonna be a place where we're
1:07:19Unidentified speaker (speaker_2): But a lot of those trades I think from that Wilson has recommended, a lot of those trades have worked. Um like because th the the RSP equal weighted S P five hundred really came back and really came back relative to the market cap weighted S P five hundred. Chips got sold. Yeah. Um maybe there's more runway. I think s some of these non non-tech tech names are good, but the the point that I think is is really important is that number one, I do think you're seeing incremental operating and margin expansion outside of tech because companies are using AI. And number two, when you see some people say, I'm not seeing it in the margins, I'm not worried about that. There might be other cyclical issues weighing on somebody's margins. Yeah. But when you look at what the companies say about what they're getting out of AI, I'm gonna give you an example. Citizens Financial, which is a stock pick of mine. They explained in their in their earnings um presentation a quarter ago, they explained that they're going to save hundreds of millions of uh opex op ex annually, which is
1:08:23fairly material to Citizens Financial, materially enough. And it's because of AI. So if if they have a bad quarter, something happens to revenue, margins don't look great, whatever, you get operating deleverage, it's not because of AI. The AI from an incremental perspective is working. And I think that's going to be a
1:08:41Unidentified speaker (UNKNOWN): lot of people. I question whether it's working.
1:08:43Unidentified speaker (speaker_1): And the concern I have is the positive examples I've heard have now been repeated for two to three years, which is um help desk has been aided by AI automation. So some of this stuff can be done without a live operator, which is true. Every programmer is now co-coding with Claude Code or OpenAI codex. So we're spending as a company $12 million a year because we're getting the work of 24 entry-level programmers, and so it's justified for our a coding. And all of the things that are supposed to happen with enterprise inside of airlines, retailers, consumer packaged goods, uh, hospitality, travel, and leisure companies, normal companies are, as far as I can tell from everyone who does these surveys, not showing the kind of ROI that they're supposed to show because many of the projects are not actually either not working in production or not. getting into production because people are finding out it it's complex. It it takes a lot
1:09:43Unidentified speaker (UNKNOWN): of work to make it work.
1:09:44Unidentified speaker (speaker_2): Are you looking at ROI per project, like incrementally, or like consolidated company ROI?
1:09:51Unidentified speaker (speaker_1): I don't do the surveys, but the people who do the surveys say we talk to these CIOs, and what they say is sometimes there is a a productivity increase, sometimes there is a cost savings, but it's at a low level, and the majority of these companies are not seeing either, and they're certainly not seeing a boost to revenue, which is kind of what you'd like to see. And when you quote citizens, my concern for a citizens would be: some of these things come from the boardroom and the CEO, and they say, We're going to save such and such hundreds of millions, or we're going to produce such and such hundreds of millions in extra revenue, and that that's a placeholder and their hopefully.
1:10:28Unidentified speaker (UNKNOWN): I
1:10:29Unidentified speaker (speaker_2): ignore the extra revenue. AI doesn't expand GDP.
1:10:33Unidentified speaker (speaker_1): It might, but my point is these things are made in it as predictions in advance of the actual results, and I'm not convinced any of them will come dependent on the
1:10:40Unidentified speaker (UNKNOWN): money. Corey,
1:10:41Unidentified speaker (speaker_0): what'd you say?
1:10:44Unidentified speaker (UNKNOWN): Oh,
1:10:44Unidentified speaker (speaker_1): we can't hear
1:10:45Unidentified speaker (UNKNOWN): you
1:10:45Unidentified speaker (speaker_2): again,
1:10:45Unidentified speaker (speaker_1): Corey. Oh, there you are.
1:10:47Unidentified speaker (speaker_3): Um AI is absolutely expanding GDP. I can look around my neighborhood and see the prices of the houses because the AI executives
1:10:56Unidentified speaker (UNKNOWN): are able to
1:10:57Unidentified speaker (speaker_3): buy bigger houses.
1:10:58Unidentified speaker (UNKNOWN): I can see
1:10:58Unidentified speaker (speaker_3): the Porsches driving down my street when they were Hondas 10 years ago.
1:11:03Unidentified speaker (UNKNOWN): That's
1:11:03Unidentified speaker (speaker_1): a different re that's a different type
1:11:04Unidentified speaker (UNKNOWN): of
1:11:05Unidentified speaker (speaker_1): different type of GDP extreme.
1:11:07Unidentified speaker (UNKNOWN): No,
1:11:07Unidentified speaker (speaker_0): there's
1:11:07Unidentified speaker (UNKNOWN): different productivity increases. But I
1:11:10Unidentified speaker (speaker_2): feel
1:11:10Unidentified speaker (UNKNOWN): like that's
1:11:10Unidentified speaker (speaker_2): near to
1:11:11Unidentified speaker (UNKNOWN): the case. Trevor Burrus. It's
1:11:11Unidentified speaker (speaker_3): happening. Yeah. Like
1:11:12Unidentified speaker (speaker_2): it doesn't
1:11:12Unidentified speaker (UNKNOWN): matter. There's also
1:11:13Unidentified speaker (speaker_3): Corey's Corey,
1:11:14Unidentified speaker (UNKNOWN): about productivity, right? that's
1:11:15Unidentified speaker (speaker_3): called froth.
1:11:16Unidentified speaker (speaker_2): That's
1:11:17Unidentified speaker (UNKNOWN): kind of the same thing.
1:11:18Unidentified speaker (speaker_0): That's a
1:11:19Unidentified speaker (UNKNOWN): thing. That's
1:11:19Unidentified speaker (speaker_3): wealth
1:11:20Unidentified speaker (speaker_0): creation.
1:11:23Unidentified speaker (speaker_3): And we'll see about productivity.
1:11:24Unidentified speaker (UNKNOWN): It doesn't increase the number
1:11:26Unidentified speaker (speaker_2): of dollars per money tree.
1:11:29Unidentified speaker (speaker_3): Tiernan had uh uh the you know the the there was an MIT T study that came out initially that showed that AI wasn't delivering results, and the study's been largely debunked since then. I can just tell you that I've been interviewing dozens of anthropic partners who, as I mentioned earlier in the program, are the evangelists for those companies that have been coming to San Francisco to get trained up on how to spread anthropic throughout their organization. And what we see in the anthropic numbers is that that's working. That's
1:11:55Unidentified speaker (UNKNOWN): my problem.
1:11:57Unidentified speaker (speaker_3): A company come in and they start using it and they put it to work and they're changing their businesses and getting greater
1:12:01Unidentified speaker (UNKNOWN): productivity at the time. That's my problem. And
1:12:02Unidentified speaker (speaker_3): Anthropic's getting paid for
1:12:03Unidentified speaker (UNKNOWN): that.
1:12:04Unidentified speaker (speaker_1): That's my problem, is there are still entry-level programmers whose work is either being augmented or replaced, but the entire economy doesn't work on supplementing or replacing entry-level programmers. At some point, it has to come to what is supposedly the automation promise, which is cash to quote. Cash to quote is important for large companies that have processes that are heavy processes. that are now supposed to be magically made more efficient by all of this. It can't just be based on we use Claude Codecs, Claude Code, and we replaced or we augmented a lot of productivity of programmers. That's not going to support an entire economy.
1:12:45Unidentified speaker (UNKNOWN): And that takes a lot of work. That
1:12:46Unidentified speaker (speaker_1): takes a lot of work.
1:12:49Unidentified speaker (UNKNOWN): That takes a lot of work.
1:12:50Unidentified speaker (speaker_1): We haven't seen that yet.
1:12:51Unidentified speaker (speaker_3): Yes. Indeed. And these changes are happening fast. And I think that that's that is one of the big risks to this stock, right? Is the the pace of change that is as now uh focused the wrath of voters on data centers could very quickly turn to AI writ large because it is changing employment. It's making things harder for some people in some professions, like entry-level programmers, to keep their work. And and the our ability
1:13:14Unidentified speaker (UNKNOWN): to be able to do that.
1:13:15Unidentified speaker (speaker_3): as a society to to make those changes uh palatable, uh our ability to retrain workers, which is something our country hasn't really put much investment in ever, um, is is gonna be is going to matter a lot as the pace of change increases. Aaron
1:13:28Unidentified speaker (UNKNOWN): Ross Powell
1:13:28Unidentified speaker (speaker_2): Corey, I think we're already there in terms of sentiment. In terms of negative sentiment.
1:13:33Unidentified speaker (UNKNOWN): Uh
1:13:33Unidentified speaker (speaker_0): I agree.
1:13:34Unidentified speaker (speaker_2): It's not just it's not just really about data centers. I don't think it's about that. I mean like just anecdote. anecdotally, like my 14 year old doesn't like AI, and when I try and ask why, I don't really get a very clear answer. But there is this deep uneasiness with AI because of the existential like feeling that, oh, it's gonna replace me. Um, but I think it's it's broader than that too. It's just this sort of feeling like, why are we doing this?
1:14:01Unidentified speaker (UNKNOWN): it?
1:14:01Unidentified speaker (speaker_1): What do you see?
1:14:02Unidentified speaker (UNKNOWN): I Yeah, mean, you know, you know, I've
1:14:03Unidentified speaker (speaker_3): got a I've got
1:14:04Unidentified speaker (UNKNOWN): kids in college. I have three kids in college and
1:14:05Unidentified speaker (speaker_3): one that just came out, and he's changing his career goals to try to respond to AI, to try to bec work for AI companies, because the kinds of jobs that he thought he'd be applying for while he was in college are harder to get now because they're being some of the dumb work that was out there for for you know 22-year-olds 22 year olds is is not available as much as I was. want.
1:14:23Unidentified speaker (UNKNOWN): You have
1:14:24Unidentified speaker (speaker_2): to challenge yourself not to rely on AI too much because you will become stupid.
1:14:29Unidentified speaker (UNKNOWN): Yeah.
1:14:29Unidentified speaker (speaker_2): Like there there are times where I'm like, I'm going to actually manually crunch this data in Excel because if I don't, I will forget how things like correlation are being. Well,
1:14:39Unidentified speaker (UNKNOWN): you know
1:14:39Unidentified speaker (speaker_0): what's very funny about that? You just said manually crunch the data in Excel.
1:14:43Unidentified speaker (speaker_2): Yeah.
1:14:44Unidentified speaker (UNKNOWN): Not
1:14:44Unidentified speaker (speaker_0): manually crunch. If Excel is crunching,
1:14:47Unidentified speaker (speaker_2): I'm not using a pencil.
1:14:48Unidentified speaker (UNKNOWN): If
1:14:48Unidentified speaker (speaker_1): you're going down and writing, then you're manually crunching. I
1:14:50Unidentified speaker (UNKNOWN): mean,
1:14:50Unidentified speaker (speaker_1): we've already outsourced some of our brains to all of this stuff.
1:14:54Unidentified speaker (UNKNOWN): It may not look obvious.
1:14:56Unidentified speaker (speaker_0): It may not look obvious, but when I first started on Wall Street, you know, we didn't have spreadsheets, right? And uh we we did manually crunch data. You
1:15:04Unidentified speaker (UNKNOWN): didn't even have pencils.
1:15:07Unidentified speaker (speaker_0): And uh we only use pencils.
1:15:08Unidentified speaker (UNKNOWN): Oh my dad
1:15:08Unidentified speaker (speaker_2): tells us
1:15:09Unidentified speaker (UNKNOWN): stone
1:15:10Unidentified speaker (speaker_1): stone tablets,
1:15:10Unidentified speaker (UNKNOWN): too. much. We
1:15:12Unidentified speaker (speaker_0): all thought that
1:15:13Unidentified speaker (UNKNOWN): we
1:15:13Unidentified speaker (speaker_0): all thought that Excel was gonna like going to just destroy entire industries.
1:15:16Unidentified speaker (UNKNOWN): In the 90s,
1:15:17Unidentified speaker (speaker_2): people were like Excel is making us stupider. Trevor
1:15:19Unidentified speaker (UNKNOWN): Burrus, Jr.
1:15:19Unidentified speaker (speaker_0): And let's think about what Excel has done in terms of productivity increase for GDP over all these years. Since we actually did all this.
1:15:29Unidentified speaker (speaker_2): Can I ask you about that the productivity? Like, isn't that near term because you can't expand the number of dollars that are circulating in the gun? guy.
1:15:35Unidentified speaker (speaker_0): It's limited. No,
1:15:36Unidentified speaker (UNKNOWN): it
1:15:36Unidentified speaker (speaker_0): no. It's like you can
1:15:37Unidentified speaker (UNKNOWN): do more dollars with one on. dollar. Think about
1:15:39Unidentified speaker (speaker_0): it, you could do more with one dollar. Like you know, you have people digging in the sand uh and then you give them a shovel, and so one worker can now, you know, dig up more whatever you're doing.
1:15:49Unidentified speaker (UNKNOWN): Right, but it doesn't
1:15:50Unidentified speaker (speaker_2): magically increase demand.
1:15:52Unidentified speaker (speaker_0): No, but it increases per you get more out of one worker, right?
1:15:57Unidentified speaker (UNKNOWN): So that's why I'm focusing
1:15:58Unidentified speaker (speaker_2): on margins. Yeah.
1:15:59Unidentified speaker (speaker_0): On
1:15:59Unidentified speaker (UNKNOWN): companies' company's margins.
1:16:00Unidentified speaker (speaker_0): Absolutely.
1:16:00Unidentified speaker (speaker_2): Yeah.
1:16:01Unidentified speaker (speaker_0): Absolutely. But it definitely has a possibility. positive tailwind, it has definitely a tailwind on productivity. You don't
1:16:06Unidentified speaker (UNKNOWN): agree with that.
1:16:07Unidentified speaker (speaker_2): No, no, ma no, I I'm not an economist. Maybe I I don't know. Yeah,
1:16:10Unidentified speaker (speaker_0): yeah.
1:16:10Unidentified speaker (UNKNOWN): I'm just
1:16:11Unidentified speaker (speaker_2): saying it doesn't magically increase demand. No, but
1:16:13Unidentified speaker (speaker_1): it has his
1:16:14Unidentified speaker (UNKNOWN): it doesn't
1:16:14Unidentified speaker (speaker_0): magically increase
1:16:15Unidentified speaker (UNKNOWN): demand, but
1:16:16Unidentified speaker (speaker_0): historically it has increased demand.
1:16:17Unidentified speaker (speaker_1): Okay. It has.
1:16:18Unidentified speaker (speaker_2): Jevins
1:16:18Unidentified speaker (speaker_1): paradox.
1:16:19Unidentified speaker (UNKNOWN): If you look at what's happened with
1:16:21Unidentified speaker (speaker_2): these various
1:16:21Unidentified speaker (UNKNOWN): uh
1:16:22Unidentified speaker (speaker_0): technological revolutions
1:16:24Unidentified speaker (UNKNOWN): that
1:16:24Unidentified speaker (speaker_0): supposedly were replacing things.
1:16:27Unidentified speaker (speaker_2): you
1:16:27Unidentified speaker (speaker_1): ended
1:16:27Unidentified speaker (UNKNOWN): up
1:16:27Unidentified speaker (speaker_0): increasing aggregate
1:16:29Unidentified speaker (UNKNOWN): demand in almost
1:16:32Unidentified speaker (speaker_2): for some period of time you do, for sure. I I
1:16:35Unidentified speaker (UNKNOWN): I
1:16:35Unidentified speaker (speaker_2): know that.
1:16:36Unidentified speaker (UNKNOWN): So
1:16:36Unidentified speaker (speaker_3): Always.
1:16:37Unidentified speaker (UNKNOWN): Yeah.
1:16:37Unidentified speaker (speaker_3): Yeah. It keeps going. We're not fighting over a limited amount of uh you know, the the the the pool that gets bigger. As as we have economic success, the econom the economy grows, that's how it's
1:16:54Unidentified speaker (speaker_1): We're not there yet for the mainstream of companies that need
1:16:58Unidentified speaker (UNKNOWN): to be.
1:16:59Unidentified speaker (speaker_0): But we generally like AI. I think that's the the thing. We generally like
1:17:03Unidentified speaker (UNKNOWN): AI.
1:17:04Unidentified speaker (speaker_0): We generally like anthropic. We may not love it right here, uh, but we like it. And maybe if we go back to the time where
1:17:10Unidentified speaker (UNKNOWN): it has
1:17:11Unidentified speaker (speaker_3): $96?
1:17:12Unidentified speaker (speaker_0): I asked uh $95.70, yeah. When I asked my second,
1:17:14Unidentified speaker (UNKNOWN): I think I'm not going to be
1:17:15Unidentified speaker (speaker_0): favorite employee, OpenAI, uh, what to do today, right? Chat GPT. And ChatGPT said no pass, right? I think that we all kind of agree.
1:17:24Unidentified speaker (speaker_1): Julie wasn't here then. I told uh Mark that I don't address my assistants by name or in a like a friendly way. I don't anthropomorphize the bot.
1:17:34Unidentified speaker (speaker_0): All right. Let's do a little. Guys, this was an amazing conversation with
1:17:38Unidentified speaker (UNKNOWN): the first time. Thank you, Mark.
1:17:38Unidentified speaker (speaker_0): There's so much we could talk about here. We touched so much today. Thank you so much for being here. What a great day.
1:17:43Unidentified speaker (UNKNOWN): Good to see you guys. Thank you. Thanks.