VC10X - conversations with the people who allocate capital

Rob Li argues AI reduces market efficiency by creating data deserts, enabling human investors to find alpha in zero-to-one innovations.

1h 1m Listen to the original Episode page
Venture CapitalPrivate EquityAI InfrastructureSemiconductorsConsumer DiscretionaryMarket EfficiencyInvestment Philosophy

Summary

Rob Li, Managing Partner at Amont Partners, discusses his concentrated, long-term investment philosophy, emphasizing the advantages of quasi-permanent capital over short-term mandates. He highlights specific positions such as a long in Micron, driven by the structural shift to High Bandwidth Memory (HBM), and a short in Nike, based on apparel product cycle risks. Li contends that AI and passive investing reduce market efficiency by creating 'data deserts' and 'data icebergs,' where machines lack context or historical data for new innovations. He illustrates how human judgment can exploit these inefficiencies through cross-regional arbitrage and deep industry knowledge. Additionally, Li outlines his '3M filter' for investing—Moat, Management, and Mispricing—and discusses the dynamic nature of disruption in the AI sector.

Key takeaways

  1. 1
    Li's '3M filter' for investing consists of Moat (price power/compounding), Management (ROIC mindset and aligned incentives), and Mispricing (market ignorance or patience for opportunity).

    Unidentified speaker (speaker_1) · 48:58

Episode sections

  1. 2:05 – 5:36

    Investment Philosophy and Fund Structure

    Rob Li describes Amont Partners as a concentrated, Buffett-style equity portfolio focusing on tech, consumer, and industrials, with a long-term horizon. He explains that outside LPs were added to scale the strategy and provide diverse macro/credit insights, while maintaining quasi-permanent capital to avoid short-term liquidity pressures.

  2. 5:36 – 8:58

    Micron and the High Bandwidth Memory Thesis

    Li discusses holding Micron since 2022, citing the emergence of High Bandwidth Memory (HBM) as a structural change to the memory cycle. He argues that HBM creates a massive, growing market from zero, providing downside protection through Micron's cash generation and low valuation multiples compared to other AI names.

  3. 8:58 – 18:32

    Nike Short and Product Cycle Risks

    Li explains his long-term short position in Nike, attributing it to the long product cycle in apparel (approx. two years) rather than brand weakness. He notes that missing a product cycle causes prolonged financial suffering, contrasting this with the long-term holding of assets like Micron.

  4. 18:32 – 24:02

    Cross-Regional Arbitrage and Insights

    Li highlights the advantage of investing globally, noting that specific industries like Nordic betting software or European power semiconductors do not exist in the US. He cites using insights from a Chinese logistics company (Baozun) to validate Shopify's transition to an asset-heavy model.

  5. 24:02 – 32:10

    Why Machines Make Markets Less Efficiency

    Li argues that AI and passive money reduce efficiency due to 'data deserts' (no data for zero-to-one innovations) and 'data icebergs' (machines misinterpreting data without context). He provides examples of Qualcomm and Intel to illustrate how human judgment can find alpha where machines fail.

  6. 32:10 – 35:41

    AI Edge and Data Deserts

    Rob Li argues that AI creates more opportunity for human investors because machines cannot access 'data deserts' (undisclosed information) or understand context. He cites Fortress Aviation, where AI lacks client data, allowing Li to verify operations and buy after a short seller report caused a 50% drop.

  7. 35:41 – 38:19

    Cross-Regional Arbitrage and Data Galapagos

    Li describes 'data Galapagos' where regional silos create arbitrage opportunities, such as using Korean/Chinese market performance to predict results for US, European, and Australian direct sellers. He believes AI will eventually capture these cross-regional correlations, eliminating this edge.

  8. 38:19 – 43:49

    AI CapEx Funding and Corporate ROI

    Li views debt funding for AI CapEx as a 'bridge solution' rather than a fundamental fix. He argues that sustainable investment requires Fortune 500 customers to see real ROI on AI spending. He tracks Snowflake's accelerating AI revenue growth as an early indicator of corporate spending health.

  9. 43:49 – 47:01

    SME AI Adoption and Productivity

    Li provides an example of a small auto parts dealership using an AI voice agent ('Carline') to capture after-hours revenue, demonstrating tangible ROI without buying enterprise software. He notes that even small businesses are finding creative ways to use AI models to generate sales.

  10. 47:01 – 48:53

    Patience and Thesis-Driven Investing

    Li emphasizes patience, citing Shopify's thesis playing out over three quarters despite stock volatility. He contrasts this with a failed activist investment in a uniform rental company, where a trade war disrupted the cost structure and destroyed the investment thesis, forcing a 35% loss.

  11. 48:53 – 51:48

    The 3M Filter: Moat, Management, Mispricing

    Li defines his '3M filter': Moat (price power/compounding), Management (ROIC mindset and aligned incentives via free cash flow/stock performance bonuses), and Mispricing (market ignorance or patience for opportunity). He states mispricing is where most ideas are killed or found.

  12. 51:48 – 54:49

    Dynamic Disruption: Disruptor vs. Disrupted

    Li explains that disruption is dynamic; today's disruptor (Nvidia) can become the disrupted entity as clients build custom ASICs/TPUs. He notes that AI agents require more CPU for orchestration, shifting the CPU-to-GPU ratio from 4:1 to 2:1, potentially boosting CPU growth relative to GPUs.

  13. 54:49 – 1:00:35

    Private Equity Lessons for Public Markets

    Li applies PE principles to public markets: patience (7-10 year horizons) and a focus on free cash flow over EPS/EBITDA. He uses supplier networks (e.g., sportswear manufacturers) to identify emerging trends like 'athleisure,' which helped him short Under Armour for failing to adapt.

Assets discussed

AssetKindStanceContext
AppleAAPLequityneutralApple made a threat. They're going to move to Intel, right?
NVIDIANVDAequityneutralI'll give you a classic class example, which is everybody know knows Nvidia. GPU was a new thing.
MicrosoftMSFTequityneutralMicrosoft's doing their own thing.
AmazonAMZNequityneutralAmazon doing their own ASIC program.

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

Claims worth verifying

Rob Li is the Managing Partner at Amont Partners and second generation at his family's single family office.

Unidentified speaker (UNKNOWN) ·0:28

How to verify: Company website or LinkedIn profile of Rob Li. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.

Micron can buy back its entire business in less than 32 months using accumulated cash and cash flow.

Unidentified speaker (UNKNOWN) ·16:14

How to verify: Micron's Q3 2026 earnings report or investor presentation detailing cash balances and free cash flow. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.

Intel stock dropped from $120 to $80 (35% drawdown) following recent quarterly earnings.

Unidentified speaker (UNKNOWN) ·30:00

How to verify: Intel (INTC) stock price history around the Q3 2026 earnings release date. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.

AI models cannot identify Fortress Aviation's clients or operational details because the company never disclosed this information.

Unidentified speaker (speaker_1) ·33:50

How to verify: Check Fortress Aviation's public filings or website for client lists; compare with Li's claim of ~14 clients. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.

Primary source

Fortress Aviation stock dropped 50% after a short seller report by Model Waters (likely 'Model Water' or similar) stating they were not fixing engines.

Unidentified speaker (UNKNOWN) ·34:34

How to verify: Search SEC filings or financial news archives for short seller reports on Fortress Aviation around 2026-2026; verify stock price movement. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.

Primary source

Fortress Aviation has roughly 14 clients.

Unidentified speaker (speaker_1) ·34:11

How to verify: Contact Fortress Aviation investor relations or check their website for client references. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.

Modern Water (likely 'Model Waters') published a short report on Fortress Aviation almost two years ago.

Unidentified speaker (speaker_1) ·51:01

How to verify: Search for short seller reports on Fortress Aviation dated approximately 2026-2026. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.

Andy Jassy (Amazon) stated payback on AI infrastructure is less than two years.

Unidentified speaker (speaker_1) ·39:57

How to verify: Check Amazon's Q3 2026 earnings call transcript or investor presentation for comments on AI infrastructure payback periods. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.

Snowflake AI-related revenue growth accelerated from mid-20s to approaching 40% by end of 2026.

Unidentified speaker (UNKNOWN) ·43:13

How to verify: Check Snowflake's Q3 2026 earnings report for AI revenue growth rates and guidance. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.

Shopify stock rose from $35 to near $1,000 over three quarters.

Unidentified speaker (UNKNOWN) ·47:20

How to verify: Check Shopify (SHOP) stock price history for the period leading up to late 2026. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.

Fortress Aviation sources 95% of its cloth materials from China.

Unidentified speaker (speaker_1) ·47:57

How to verify: Check the uniform rental company's (likely 'UniFirst' or similar) 10-K filing for supply chain disclosures; note: Li may have confused the company, as Fortress Aviation is an engine repair firm, not a uniform rental company. The 95% China claim likely refers to the uniform rental company he was activist in. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.

Primary source

Do your own research

  • Verify Micron's HBM production capacity and customer adoption rates against competitors Samsung and SK Hynix using earnings call transcripts and industry reports.
  • Check Nike's inventory levels and product launch calendars to assess the impact of product cycle misses on financial performance.
  • Analyze Snowflake's AI-related revenue growth trends and customer acquisition metrics to evaluate corporate AI spending health.
  • Review Amont Partners' 13F filings to confirm current holdings and position sizes in Micron, Nike, and other mentioned equities.
  • Assess the ROI of AI implementations in small businesses, such as the 'Carline' voice agent example, to validate claims about SME adoption.
  • Investigate the shift in CPU-to-GPU ratios in data centers to evaluate the impact of AI agents on CPU demand versus GPU demand.
  • Examine supplier network data for sportswear manufacturers to identify emerging trends like 'athleisure' and assess Under Armour's strategic response.
  • Compare valuation multiples of traditional memory companies versus AI-focused firms to verify claims about downside protection and margin of safety.
Full transcriptexpand
0:00
Prashanthan: Most people argue the opposite. That more machines mean more efficient markets. Uh why are they wrong?
0:06
Unidentified speaker (speaker_1): So we actually completely disagree. The first bucket will be called the data dieser. So literally there's no data for machine to see.
0:14
Prashanthan: With AI coming into the picture,
0:16
Unidentified speaker (UNKNOWN): do
0:16
Prashanthan: you think your edge still remains
0:18
Unidentified speaker (speaker_1): if you ask AI right now, you type in JAGT or uh Claw and say, Hey, who are their clients? Give me their client list.
0:25
Unidentified speaker (UNKNOWN): AI Hey,
0:26
Unidentified speaker (speaker_1): doesn't it even know.
0:28
Unidentified speaker (UNKNOWN): Hey
0:28
Prashanthan: everyone, this is Prashanthan. I'll be host of VC10X Podcast, and today we have Rob Lee with us. Rob is a managing partner at Amont Partners and the second generation at his family's single family office. He started in private equity at Boggan Stanley and he now runs a concentrated Buffett-style public equity portfolio.
0:46
Unidentified speaker (UNKNOWN): In
0:46
Prashanthan: this episode, we talk about why more machines are making markets less efficient. his view on the memory super cycle, and whether AI will erase the human edge in investing. So without wasting any time, let's dive straight in. This episode is brought to you by Podcast Enix, the team behind the show you're listening
1:03
Unidentified speaker (UNKNOWN): to. As
1:04
Prashanthan: an investor, you're already talking with founders, allocators, and operators. We just turn those conversations into a branded podcast, and each recording becomes a YouTube video, an audio,
1:14
Unidentified speaker (UNKNOWN): and a video.
1:15
Prashanthan: episode, short clips, LinkedIn posts, and a newsletter, activating all your marketing channels at once. This keeps you top of mind with LPs and founders without taking away much time from your week. You can book a free strategy call with me at podcasttenx.com. I'll add the link in the show notes below. Now enjoy the episode.
1:33
Unidentified speaker (UNKNOWN): Hey
1:34
Prashanthan: Rob, so good to have you on the VC VC10X TenX podcast. How do are you doing?
1:37
Unidentified speaker (speaker_1): Good. Thanks for uh the invitation. Happy to be here.
1:40
Prashanthan: Absolutely. It's a pleasure having you on. And uh to start things off, let's talk about uh you being a second generation at a single family office. Uh, walk us through how the family thinks about allocation, uh, what gets the core capital, how you split between public and private markets, and what horizon you invest for. And why do you run at Amont amount as a separate
2:00
Prashanthan: fund with outside LPs rather than just deploying? the strategy inside family capital?
2:05
Unidentified speaker (UNKNOWN): Well,
2:05
Unidentified speaker (speaker_1): that's a lot of questions to unpass. So let's go one by one. I would say the first, like just a broad philosophy, you know, how we invest. Um, you know, I I came from a private equity background, uh really the school thought from Warren Buffett. So really try to imitate or follow the Berkshire Highway way of
2:25
Unidentified speaker (speaker_1): investing. Uh so in terms of time horizon, I mean the ideal answer is forever, right? You we want to hold good assets that we trust the management, we believe in the business model that can generate cash flow and compound that free cash flow to reinvest with attractive return. Uh so for that type of business, you know, obviously we'll hold for ideally for
2:45
Unidentified speaker (speaker_1): many years to come. Uh, but you know, market change, things change, valuation change. So some things something become less. attractive from a risk reward perspective, you know, we're happy to access well.
2:55
Unidentified speaker (UNKNOWN): In
2:55
Unidentified speaker (speaker_1): terms of what we look at, you know, my background already came from the equity side, both public and private. So we put we pretty much stay within our circle competence, uh stay in the equity side. We don't do commodities or macros or rarely venture into equity or super early stage VC type investments. We stay in our
3:15
Unidentified speaker (speaker_1): lanes, the three sectors I personally know quite well: tech, consumer, industrials. So we stay, we stay in those lanes, really staying with our circle competence. In terms like, you know, the reason we decided to open it up for outside capital or allocators is really
3:35
Unidentified speaker (speaker_1): twofold. I would say number one is
3:38
Unidentified speaker (UNKNOWN): uh
3:38
Unidentified speaker (speaker_1): a we got a lot of inbound. You know, I've been in the in this industry for a number of years. Uh, you know, people read our letters, people understand our philosophy, people agree with philosophy. So they want to partner with us to really s help scale the strategy and enjoy return along the way. And we're happy to let them in, you know, as long as they
3:58
Unidentified speaker (speaker_1): fully on board with sort of a longer term and very fundamental perspective. when it comes to investing. The second I would say, you know, we're very intentional in terms of like curating a group of uh LPs or partners that share the same view or philosophy in terms of investing.
4:18
Unidentified speaker (speaker_1): So it's very nice to have this group of people that would kind of swap ideas on the market and also on opportunity set. Um And that's very helpful or useful to us. Even for some asset classes, you know, some folks we invest in macro or credit land. Uh those are areas we typically don't touch. But the
4:38
Unidentified speaker (speaker_1): insights they generate, whether they're looking at some at from a different perspective or angle, uh, that's very useful to us. Especially right now, you think about the all the AI CopEx question, right? The current investors are very different view versus equity investors. So that will help us to really gauge how the market thinks about the overall theme,
4:58
Unidentified speaker (speaker_1): uh, different points of view, different angles to understand the same question. And that will help, you know, make our investing better.
5:05
Unidentified speaker (UNKNOWN): Yeah,
5:05
Prashanthan: absolutely. That's quite interesting. And I did actually read some of your letters that uh you were gracious enough to forward me. And they were really good and really loved the philosophy behind. the fund. Uh so that's quite amazing. And uh, you know, when you say that with a family capital, you're more uh focused on holding it forever, like you or is
5:25
Prashanthan: that more to the ability that you can potentially hold it for forever if you wish to or uh or do you uh invest with the intention to hold forever? Which of the two is it?
5:36
Unidentified speaker (speaker_1): Right. So this burse both now we don't. definitely come with intention. And you read our letter, I mean you read the R our New York Stock Change interview on YouTube. You'll know, you know, we've been holding some of the positions like such as Micron since 2022 when it's below 40 bucks, right? Now up like whatever 25 times since in last call like four or five years.
5:57
Unidentified speaker (speaker_1): So let's guy you read our letter, we've been there since twenty twenty-three, twenty three, right? When was nobody cares, a So training eight times through earnings and, you know, thirty-five bucks. Now obviously the company's market cap has grown ten X since since been doing the new TPU business for Google and that business went from zero to a huge amount.
6:17
Unidentified speaker (speaker_1): Uh so when we invest, we definitely intend if everything goes on track, we wanna want to keep just staying there. And obviously from a tax efficiency perspective, that's very beneficial. From a structural advantage perspective, that's also very important because we do, we don't have like permanent permanent capital like Warren Buffett.
6:37
Unidentified speaker (speaker_1): I would say call like a quasi permanent capital. And all our outside capital partners understand that. So they're signing Running up off this for longer term. So that really allows to invest and think like a truly long-term investor, right? And that's really not an easy setup for most of the think about
6:57
Unidentified speaker (speaker_1): hedge funds out there. Most people have more short-term mandate. They have very strict liquidity terms. So even if they agree with our thesis on a business on a company, structurally, because our mandate, they cannot do what we do. When there is a market,
7:15
Unidentified speaker (speaker_1): correction or pullback, and you know, we've been through that many times since like three, four years ago, um, they cannot hold the stock. Like, period. So um so that gave us really an unfair advantage in terms of believing in our conviction as you know, as a fundamental So continue to move
7:35
Unidentified speaker (speaker_1): better, and that we can be there for for a longer period of time.
7:39
Prashanthan: Absolutely. That's quite interesting. And from the most latest uh letter that I read from you, uh you still hold a very sizable position in Micron. And that is uh in in a very uh interesting theme. It tells memory and uh historically it's been very cyclical. And there are investors out there who argue that you know
7:59
Prashanthan: when it's at the peak of the cycle, they these the stocks look the cheapest and they argue that we are we might just be there right now. And we have recently had about twenty percent drawdowns in these memory stocks, the big ones. How are you looking at that? And w when you think that okay, we are gonna hold this long term, and cyclicals can go directly opposite to that theme
8:19
Prashanthan: of you know, holding these things long long-term term because you have to track the cycles, you know, and also you know, s an investor maybe tend 10 to 15 years back would have thought that companies like Nike they have such strong brand value that they're gonna hold their value for a long, long time, right? So, an investor like you, maybe 10-15 years back, would have put their money in Nike and forgot about it
8:39
Prashanthan: that okay, we are long-term investors, the brand is not going away anytime soon. Uh so how do you think about that? Because both of these are contradicting. Nike has also had a very significant drawdown, and what you're investing in. Micron, also is a very cyclical kind of play, uh, from what I understand. So we'll love we'll love your view into that.
8:58
Unidentified speaker (UNKNOWN): Yeah,
8:58
Unidentified speaker (speaker_1): no, yeah. I think you asked a very good question. Sort of nothing stays in the same place, right? Everything's dynamic, and obviously, as fundamental change, you need to update your personal view. It's interesting you mentioned Nike. If you read our latest like three Q letter, you actually read in the short section, right? We actually talk about we've been shorting the stock for like over a
9:18
Unidentified speaker (speaker_1): year. And obviously that has generated a lot of return for our short book. And we talk about it. The brand is one part of the story. But the other thing, you know, for anything apparel or sportswear related, it's really it's a product story. You know, like, you
9:33
Unidentified speaker (UNKNOWN): know,
9:34
Unidentified speaker (speaker_1): iPhone, you know, like any c kind of consumer electronic products, sportsware functions. fundamentally is a product side of story. Now, historically, oh, even until now, the product cycle very long, typically it's like two years. So let's say you and I design a shoe today and we sign off, the shoe, the the
9:53
Unidentified speaker (speaker_1): actual product would not hit the shelf that consumer can buy two years down the road. Right. So we're now in October 2026. You and I agreed agree to design this great fantastic shoe. The shoe will not be there for people to buy to use until October of fall season of 2028. So it's a very long product cycle. Now if
10:13
Unidentified speaker (speaker_1): you miss a product cycle, you know, the brand could be still there. The brand image is still very strong. But if you miss the product cycle, you'll suffer For for at least a year and a half to two years. Because you want to change course, you want to get a new product there, but it takes two years for that to show up. So that's the primary reason we're bidding against that. It's
10:33
Unidentified speaker (speaker_1): necessarily we have a strong view on the brand image itself. You know, we can debate that. We don't have a strong view on that, but the product cycle, we're very clear on that. Now, going back to the micron, so if you uh watch the the New York Stock Change interview online, you see people at the ask same time. question that was the interview almost
10:53
Unidentified speaker (speaker_1): like three years ago, right? So people were asking the same question back then is the peak of the cycle, and people asked that question last year, and people asking the question right now. And it's interesting that people never ask the question whether stock is moving up. People only ask the question with stock coming down. Uh So it's a little bit difficult to tell what is
11:13
Unidentified speaker (speaker_1): the cause cost and what is the effect. Is it because the stock price is down? So people trying to find excuse to justify why the stock is not. Now, we invest in stock not just like most last five years, right? My investing career, we've been in the sector, I would say, the DRM sector, three times. And we've been through multiple
11:32
Unidentified speaker (speaker_1): cycles. So we're very aware that it was a very cyclical cynical industry, right? You always buy. That's yeah. you know, at the cycle bottom. When the three majors start cutting topics, that's the time you buy. And when you sell them when they, you know, well, things are good, but when they're adding capacity a lot, that's the time you sell. Historically, that worked every single
11:52
Unidentified speaker (speaker_1): time. We've been through three cycles from 2012 12 all the way through COVID, so 2022. So for 10 years, three cycles that worked a beautifully. up every single time. I think people really build in that second nature. That's how they think about that. But I would say since AI that's really
12:12
Unidentified speaker (speaker_1): dramatically changed the entire businessline model. And we I talked about in that video interview with New York Stock Change three years ago. One of the key reasons we decided to size up the position to our biggest position in the last couple of years is really the new product called high bandwidth memory, which,
12:32
Unidentified speaker (speaker_1): you know, in simple terms, is a huge size like a blackboard, right? So when you do calculation pre-AI, it's very simple. It's like two plus two equals four. You just need like small blackboard, like hand view or the size of your hand. You can just write two plus two equals four, duh, right? You don't need a huge blackboard. So they they've had the product have high
12:52
Unidentified speaker (speaker_1): balance memory for many years. It's just there's no use of it. There's no market of it, right? Because nobody needs a huge blackboard. Um, but when I came, then they suddenly realized hey, you know what? There's a very interesting use case because AIM uh calculations are very complex and they require multiple steps, right?
13:12
Unidentified speaker (speaker_1): So if you let's say you need to solve this mathematical problem. it's like hundred steps, you need to write, you know, step one, wipe it, right? Then you write step two and wipe it. You can use a small blackboard to do it, but every side every step you need to wipe it. So it takes a lot of time. Versus now you have this like huge size of like blackboard, the size of the
13:32
Unidentified speaker (speaker_1): wall, you can literally write 100 hundred steps in one shot, right? Get to the result and then wipe it and then do the second uh calculation. So that becomes a lot more efficient and saves you time. So all of a sudden, this this you know, this wall-sized blackboard, which has no use case before, now has a massive use case. So
13:52
Unidentified speaker (speaker_1): so that's really a story from going from zero to one. Um, you know, back in 2022, 2023, we do don't we did not know how big the market was, right? Because there's like one. client in BBS, and now they got like, you know, twenty clients right now. And the revenue was zero because the lure just started. Like we didn't know it's so this
14:12
Unidentified speaker (speaker_1): by the end of this year, this market will become roughly a hundred billion, right? So going from zero to a hundred billion in a couple of years. So it's a massive growth. But back then we was like, we didn't know it's gonna be 10 billion market, 30, 50, or whatever number. But we think the risk reward becoming becomes very attractive because that's totally not price down. Now, fast forward today, if
14:32
Unidentified speaker (speaker_1): you ask an average investor, they probably know about the you know this this business. And people know, okay, this is this is a good business. So now the question is coming back to us like, well, why are you still there? Right? It's still attractive. Now, from a risk reward war perspective, right now, it is attractive because if you look at the entire, not just
14:52
Unidentified speaker (speaker_1): tech, if you just think about the whole AI complex, if you judge them by absolute valuation metric, this sector, I'm not just talking about micron, but the whole DREM sector, they trace like three times, three and a half times like earnings. This is like absolutely cheapest versus everything
15:11
Unidentified speaker (speaker_1): else. Now, people say, well, but the earnings you cannot trust. the earnings, right? The earnings might come basically cycle B, earnings might come down 50%, earnings might come down 60%. Okay, so let's let's give it that. Let's say the earnings does come down 50 50%, because we're a cycle peak. They'll be trading six times earnings. Let's say the earnings
15:32
Unidentified speaker (speaker_1): come down 70%, come down two-thirds. They'll be trading 10x earnings, which is still lower than the vast majority of the AI names out there trace 50 times, 60 times earnings. And even like Nvidia, which trades like low teams earnings right now. So that gives you a lot of a downside protection. And number two is that remember, these
15:51
Unidentified speaker (speaker_1): companies such as Micron, they have so much cash and they generate so much cash flow, right? So unlike OpenAI and Thropic, Thrapic, actually burning cash,
16:01
Unidentified speaker (UNKNOWN): they
16:02
Unidentified speaker (speaker_1): are actually micro. micron is actually generating a lot of cash. And we talk about in our three Q letter, they have enough cash and cash flow that they can buy back the entire business, call like the 32 months.
16:14
Unidentified speaker (UNKNOWN): So
16:14
Unidentified speaker (speaker_1): less than three years. And that's like a massive amount of margin of safety for you. Because think about it, assuming there's another not a COVID or another huge macro shock, there's no revenue anymore. right? Or not even talk of earnings coming down 50, 60, 70%. We're saying they don't even have any income anymore. Just with the
16:34
Unidentified speaker (speaker_1): cash they accumulated over the last couple of years, they can buy back the entire business right now in less than three years. So if I were Warren Buffett, I would do it because This you room know there are only three guys in the world who can do this right now. And in the Western world, those are there's only one. The other two are in Korea. In the
16:54
Unidentified speaker (speaker_1): entire Western world, there's only one company. It's a strategic asset for the United States and for the Western world.
17:01
Unidentified speaker (UNKNOWN): Um,
17:01
Unidentified speaker (speaker_1): so that's a lot of downside production. And they'll start um giving back the cash to the shareholders starting in this December. Um so you know, like we think there's a lot of downside data prediction, production and we're happy to be there from all. longer term.
17:16
Prashanthan: Absolutely. And one of the things that you look at while making these bets is the, you know, cross-region arbitrage and sort of relates to how you mentioned about Micron, that it is based in the United States versus the other two who are based in South Korea. Can you double tap on that? That how much of an advantage is that? Because there are talks of even SK Hinex trying to
17:35
Prashanthan: open uh and partner with Intel, although not uh confirmed. as much as I'm aware, but they're also trying to come inside the US and try to uh play that same same game. So we'd love to understand how big of an advantage is being inside the US and uh uh for Micron?
17:52
Unidentified speaker (UNKNOWN): Uh
17:52
Unidentified speaker (speaker_1): well so for Micron, number one, they're a used company, right? So Hinex and uh Samsung, they can come build some build labs here, but they're not US-based. US based. It's you know, it's different. definitely not the same, you know, playing field. It's like TSMC is building the lab in here in the US, but you know, obviously Intel got preferential treatment from the
18:12
Unidentified speaker (speaker_1): US government because it is a US business, unlike TSMC. TSMC one day they wake up and say, Hey, look, we don't like the market anymore. What can pull out? They can pull out anytime they want. But Intel, these US companies based in here. here the US is a strategic asset for the United States. So it's very different. But
18:32
Unidentified speaker (speaker_1): going back to your broader question, uh in terms like, you know, as a global investor, because we do invest globally, uh, we probably right now we have like 30, a third our investments are actually outside US, across Europe and Latin America, a little bit like in Asia as well. Um I think the big
18:51
Unidentified speaker (speaker_1): advantage that we invest across different geos are really two. There's number one is the US is a big economy, but for some certain themes or segments uh you know that we want to express a view on, that industry simply does not exist in the US. So, you know,
19:12
Unidentified speaker (speaker_1): we talk about
19:12
Unidentified speaker (UNKNOWN): the
19:13
Unidentified speaker (speaker_1): DRAM D RAM sector, obviously also one of the three companies here in the US, but there's like lobby industries, you know, think about power semi companies that are mostly in Europe. You want to express a view on that sector. You just need to be in Europe. Uh think about, you know, like uh this way during COVID, we made investments in the
19:32
Unidentified speaker (speaker_1): software uh sector that caters to the casino or sports bidding uh uh industry. All the software businesses that are catering to online bidding or casino, they happen to be in Europe. They happen to be in northern like Nordic Europe. That's where the all the companies concentrate are. So if you say, hey
19:52
Unidentified speaker (speaker_1): look, you know, like you know it's COVID, people are staying home, people are bored, people want to do more casino bidding, online bidding, I'll but we express a view on that, you cannot do that in the US. We actually need to invest into these companies in Nordic Europe because that industry happened to be there. So having that flexibility to look beyond the
20:12
Unidentified speaker (speaker_1): US, it makes a lot of sense for us. And we mentioned Celeska, right? It's like IBM spun off that does a TP uh TPU server for Google, great business, but they're located in Canada.
20:24
Unidentified speaker (UNKNOWN): So
20:24
Unidentified speaker (speaker_1): if you say, hey, look, I can only invest in the US. business, uh, then you're gonna miss out that opportunity. Um the other advantage I would say is that when you invest across different regions, sometimes uh you know, companies in different regions, they share very similar fundamental drivers. And you can apply the insights you
20:45
Unidentified speaker (speaker_1): learn from region A to region B. And I'll give you an example. So this is a couple of years ago, it was almost 10 years ago.
20:52
Unidentified speaker (UNKNOWN): Uh
20:52
Unidentified speaker (speaker_1): we're looking at Shopify. They're doing the transformation from pure
20:57
Unidentified speaker (speaker_1): asset light model into asset heavy, meaning building their own logistics and payment and infrastructure so they can serve their, you know, their SME clients better. At that time, there's a lot of debate around this management decision. There are a lot of shareholders that's here in the US that only invest in the US say, hey, look, you are an internet company.
21:17
Unidentified speaker (speaker_1): We're investing in you because you're asset light, right? You internet company supposed to be asset light. And you're one of trying to be asset heavy. That's bad,
21:27
Unidentified speaker (UNKNOWN): right?
21:27
Unidentified speaker (speaker_1): Let's lower your RIC. That's bad. We don't like it. We're gonna sell. Right. So stock under a lot of pressure. Year. We become became very interested because we actually invested a similar company called Bao Zhun, it's actually based in China. They're doing similar business like Shopify, but in China. And they went through the same transition, going from pure
21:47
Unidentified speaker (speaker_1): asset light to asset heavy, building logistics or warehousing for the clients, and they achieved a good result. So that was success story. Now, China's market is different from the North American market. We understand that. But we thought, hey, at least number one. we saw a success story in a different region. So they're not the first one, right?
22:08
Unidentified speaker (speaker_1): Someone already tried it and succeeded. Number two, I understand why they want to do that, especially for SMEs, right? SMEs say, look, I want to deal with just a little warehouse. They'll have a different stuff. If you can take care of everything for us, that's better for us. Maybe what we want to spend more money with you. Um, And we understand that. will take some
22:27
Unidentified speaker (speaker_1): period of time for that to show up,
22:30
Unidentified speaker (UNKNOWN): right?
22:30
Unidentified speaker (speaker_1): In their numbers, right? Because you need to spend the money first, and then wait for eventually they wait for three, four quarters for them to show up. Now, I came from private equity, that's a sweet spot for me, because when we acquire a company, we typically do a lot of internal reward, real work, you know, like get a new management team, sell some
22:50
Unidentified speaker (speaker_1): units. we don't want, do some attack law MA for the stuff that we do want. So we're doing a lot of stuff internally, but for the outside world. So we're the same boring company growing 3% per year, nothing changed. But we know from the inside, something big changed. So it was like we're very patient, you know, less Shopify for that to to to
23:10
Unidentified speaker (speaker_1): flow through and see if it works out. And it did work out. We made actually quite a bit of money.
23:15
Unidentified speaker (speaker_1): on that position. This was before COVID. So that's a good example of applying insights and knowledge. We nerve from a different region back to Shopify's market and have more conviction than the risk of the market. We think the we did not know if the probability is
23:35
Unidentified speaker (speaker_1): 100% a hundred percent that this will work for the Record market. But we think the risk war is very attractive.
23:41
Prashanthan: Absolutely. That's quite interesting. And your core claim at amount is that cons and passive uh now drive most trading volume. And that this leaves a growing pool of mispricing in things machine can't measure. Uh most people argue the opposite that more machines mean more efficient markets. Uh why via are their they wrong? own.
24:01
Unidentified speaker (UNKNOWN): Yeah.
24:02
Unidentified speaker (speaker_1): So, you know. uh we've heard the view, people say, oh, you know, it was passive, was more machine quant trading, market will become more efficient. So we actually completely disagree. And here's our logic. I would love to hear your thoughts and audience thoughts. Now, if you go back to first principle thinking, you know, like when when machines
24:22
Unidentified speaker (speaker_1): make a decision to buy or to sell, what what do they base the decision on? It based on data. They need a lot of like high frequency, good quality, short-term data to tell them to buy or to sell. Now there's two situations. Number one, if there's no data to tell them,
24:42
Unidentified speaker (UNKNOWN): you
24:42
Unidentified speaker (speaker_1): know, about what's going on. And by default, there's no there's there's really uh they don't have any capability to decide. to buy or to sell, right? In the second situation where is yes, they see the data that's out there, but maybe don't understand the context around those data.
25:02
Unidentified speaker (speaker_1): We talk about a little about the data iceberg, right? Um, so that will give you a misunderstanding of what's really going on and they will make the wrong decision. That will become an opportunity for us. So I'll give you an example on both sides to illustrate how we think about this. So the first bucket, what we call the data
25:22
Unidentified speaker (speaker_1): desert. So literally there's no data for mission to see. So we talk about Micron a little bit, we talk about Celeste a little bit. Any situation where going from zero to one, right? Back in 2022, twenty twenty-two, when Micron hired high the new product called high balance memory, but it has no client no market, no revenue. There's no data for mission to
25:42
Unidentified speaker (speaker_1): see to really evaluate, well, how much is the business worth, right? And that became a great opportunity for us if we can make some judgment call there, even though it cannot be precise. Or Celistica. Celistica has been working with Google TPU since 2015. But for 10 years, there's really no business. You know, Google doesn't know what to do with it, right?
26:02
Unidentified speaker (speaker_1): And that's why the TPU team, uh, Google actually left to form a new company, Grok, which is now acquired by JSON Home and Nvidia. The reason they left, because they could not see a future inside the company, right? Because there's no use case for it until AI came out and then suddenly they found a use case in, and
26:22
Unidentified speaker (UNKNOWN): you know,
26:23
Unidentified speaker (speaker_1): Celestica, which has been working on this product for the longest time for Google. suddenly realized, oh great, okay. You know, our clients like product actually took off. We're gonna make a lot of money off that, right? But when
26:36
Unidentified speaker (speaker_1): um, you know, the business was starting to take off, remember there's only one client, which is Google, Google not disclosing because like this is too small. Um and it's really no data out there because the one-to-one B2B, there's no credit card data formation for mission to track. Um so there's nothing you can do about it. So these type of situations are fantastic for us
26:56
Unidentified speaker (speaker_1): because they're going from zero to one, and with the zero, there's really no data uh available for mission to see. So that's bucket number one. Bucket number two, we talk about okay, there's a bunch of data. Machine can read it, I can read it, and maybe faster. How do I still have an edge on that? So I'll give you,
27:13
Unidentified speaker (UNKNOWN): I'll
27:14
Unidentified speaker (speaker_1): give you one historical. example and one current example. Historical example is uh uh this is like before COVID. Qualcomm and um Apple had a lawsuit. So Apple says, hey Qualcomm, you're charging us too much for the baseband business
27:27
Unidentified speaker (UNKNOWN): uh
27:28
Unidentified speaker (speaker_1): unless you give us good rate a good deal, we're gonna move your business to Intel, Intel's basement business. When that lawsuit came out, the stock, Qualcomm stock immediately reacted down 50%. right? Because everybody, machines are reading the same lawsuit. Machines are like, hey, look, we know, you know, because you disclose Apple's your top client. Apple represents back then
27:48
Unidentified speaker (speaker_1): roughly 50% of your total revenue. Apple made a threat. They're going to move to Intel, right? Like you should be down 50%, right? That makes sense. Um from our perspective, you know, we're reading the same thing, but we have a different perspective. Uh if you ask actually study the baseband product between Intel and Qualcomm, you
28:08
Unidentified speaker (speaker_1): realize there's a huge performance difference between these two guys. Fast forward today, Apple eventually acquired the basement business from Intel, right? So now it's like Apple's like in-house product. But you and I know it's
28:21
Unidentified speaker (UNKNOWN): Is
28:21
Unidentified speaker (speaker_1): really start
28:22
Unidentified speaker (UNKNOWN): from
28:22
Unidentified speaker (speaker_1): the last generation of iPhone, Apple started using their own baseband, which is kind coming from Intel. For the longest period of time, they continue to use Qualcomm because Intel's product is simply not good enough. So we understood there's no chance that Tim Cook is going to move away actually from Qualcomm to Intel. He's doing this to
28:42
Unidentified speaker (speaker_1): bully Qualcomm using this lawsuit to say, hey, look, I can destroy your stock price. So please play ball. Please give me a good deal. Now,
28:50
Unidentified speaker (UNKNOWN): we
28:50
Unidentified speaker (speaker_1): were sure. assured they will get a good deal. The deal can be a discount, 5%, 7%, maybe 10%. It's not going to be 50%. So that became a very attractive risk award for us to make investment in Qualcomm. Now, eventually the two companies sell the lawsuit, right? So the second price came out and Apple said, hey,
29:10
Unidentified speaker (speaker_1): look, you know, we got a deal from Qualcomm. So all the baseback baseball business stay with Qualcomm. We're not moving away to Intel. And Qualcomm stock closed that gap like that on that day. Because that became the sweet spot for machines, right? Because they're so fast. Now if you say, hey, look, you know, I came from the Warren Buffett company investors. Um,
29:30
Unidentified speaker (speaker_1): very fundamental driven. I'm human. I'm not a human machine. I don't want to capture that spread. You gotta do that, position yourself before. the second priest lease come out. That's the only way to capture that, right? Because you gotta wait for the second priest lease to come out. You never have a chance because machines are so fast. I'll give you a current example,
29:50
Unidentified speaker (speaker_1): and we'll see in six months or three quarters if that's true or
29:55
Unidentified speaker (UNKNOWN): not. So
29:56
Unidentified speaker (speaker_1): Intel, the most recent quarterly earnings,
29:59
Unidentified speaker (speaker_1): um
30:00
Unidentified speaker (UNKNOWN): the
30:01
Unidentified speaker (speaker_1): stock came down a lot from 120. to 80, it's like a 35% drawdown. Because on the earnings call, the CEO, Lipu Time, he said, hey, I'm going to spend a lot, CapEx, for my foundry business, which mixed chips.
30:16
Unidentified speaker (speaker_1): But I'm not announcing any customers. So the market, especially machines, that got very, very angry or upset or panicked.
30:24
Unidentified speaker (UNKNOWN): Because
30:24
Unidentified speaker (speaker_1): from a machines perspective, one guy's like,
30:26
Unidentified speaker (UNKNOWN): you
30:27
Unidentified speaker (speaker_1): have no client because you didn't announce any clients, but you're spending a lot of money. That's so dumb, right? Because you're spending money, there's no return. Super bad. Stock came down 30%. Now we have a different view, even though we're reading the same, listening to the same call.
30:44
Unidentified speaker (UNKNOWN): Li
30:44
Unidentified speaker (speaker_1): Boutan was of the CEO for Cadence, which we were a shareholder of he's like a long time ago. Now this guy's style, because we're a shareholder of Cadence, we understand he's this guy that is very conservative, right? We call he's the type of guy who now releases Falcon to chase the the rabbit unless he actually
31:04
Unidentified speaker (speaker_1): sees the rabbit. So he's that type of guy. So the fact that he's actually willing to commit so much capital to the foundry business, that tells us that he actually already secured customer commitments. He already got some customers in the pocket. But for some other reason, you know, non-economic reasons, maybe customers
31:24
Unidentified speaker (speaker_1): like, well, we're still giving most war to TSMC. Please don't tell the market. Otherwise, TSMC will get pissed at us, right? It's like just like make our life a little bit easier, please. So he's not announcing the names of the customers, but we actually think he already got a customer. customers. So maybe in three quarters we'll know whether he's lying to us or maybe we're right. We don't know
31:44
Unidentified speaker (speaker_1): yet, but we think the brisk whistle war is very attractive for us. So these are some of the examples to show you. Either there's no data from mission to see, which will have a huge advantage, or yeah, maybe we're reading the same thing, we're reading the same data, but we have very different or sometimes opposite interpretation. of the same
32:04
Unidentified speaker (speaker_1): data or same fact. I will get different conclusions. And that could be another source of alpha.
32:10
Prashanthan: Absolutely. That's quite interesting how you're deriving alpha from one is believing in the person that you probably already know or have tracked for a long time. That's one. Other being that uh information which is not already there, but you assume that it's there. Right. And ways in that you're making the calls today
32:30
Prashanthan: that more in in your thinking, more likely it is going to come true in future. So that's how you're trying to derive alpha and differentiate from the machines who are just not going to make those assumptions and will solely depend on the data that's out there. Right. So one more thing I wanted to ask you is uh with AI coming into the picture, do you think your edge
32:49
Prashanthan: still remains what we just talked about? Or and and you think that can further uh make the problem worse for for what the machines were doing because all the AIs might just think the same way and derive the same conclusion. So that amplifies your edge even further. What do you say?
33:06
Unidentified speaker (speaker_1): We strongly prefer they think the same way because if they same the same think and behave the same way, now it's just l obviously that's a lot more off opportunity for us. We can take the other side if if we think they're wrong. Uh but going back to your broader question in the AI age, like what the are sort of opportunities that might go away and where the opportunity will
33:26
Unidentified speaker (speaker_1): stay. Um I'll talk about the opportunity will stay first. Um, especially in the first bucket. Uh if there's no data, it doesn't really m matter if you're AI, low low-grade grade AI, frontier AI. or just like traditional machine learning, there's nothing you can do about it. I'll give another example. So we made an investment in this
33:45
Unidentified speaker (speaker_1): company that repairs engines for airlines called
33:50
Unidentified speaker (speaker_1): Fortress Aviation.
33:51
Unidentified speaker (UNKNOWN): If
33:51
Unidentified speaker (speaker_1): you ask AI right now, you type in ChatGPT or uh Claw and say, hey, who are their clients? Give me their client list. AI doesn't even know, right? Because the company never disclosed or told. talked about who are our clients, let alone what exactly do we do for our clients. But you know, coming, I you know, I was in the bioworld, we understand the
34:11
Unidentified speaker (speaker_1): aerospace industry, we actually know who the clients are. They got like roughly 14 clients right now. Um, so there's a huge advantage. And we know who to talk to, right?
34:21
Unidentified speaker (UNKNOWN): Uh,
34:21
Unidentified speaker (speaker_1): if you don't even know the clients, how do you even go investigate? Are they fixing engines for airlines or not? Right. Because there was a short seller report about a year and a half ago and say they're not actually fixing engines for the clients.
34:34
Unidentified speaker (UNKNOWN): Now
34:34
Unidentified speaker (speaker_1): stop, you know, Model Waters push a report, stock came down 50%. And we can immediately start buying. Because we know how do we know? Number one, we know the clients. Number two, we understand how the industry works. Each jet engine, modern jet engine typically has three modules. Each module has a serial number. It's like your iPhone, like a computer. If the
34:54
Unidentified speaker (speaker_1): three modules like serial numbers match, that means they've never been tempered with, right? The state, this whole unit, entire time, which means nobody has touched the engine. They did not fix the engine. If the three modules, numbers, the serial numbers don't match, then you know some components have been replaced. If they're replaced, that means the company
35:15
Unidentified speaker (speaker_1): uh, the engine has been fixed. That's how you got to know. You you need to know to ask them the right questions, right? So that way how we build an edge, understanding that this company actually fixed the engines for airlines, and thus we had a convention by and generated alpha. Um
35:34
Unidentified speaker (speaker_1): the some of the some of the opportunities will go. away, I would say, mostly in the in
35:41
Unidentified speaker (UNKNOWN): the
35:41
Unidentified speaker (speaker_1): middle bucket, we'll call it the data Galapagos, right? We talked about the cross-reaching opportunities.
35:46
Unidentified speaker (UNKNOWN): Um
35:46
Unidentified speaker (speaker_1): I would say pre-AI, and we've done a few of these before, is that because we're living in a world where the wall should become so silo, right? So some funds only to European consumers, some funds only to US, some funds only to Asia. But maybe it's the same industry, let's say direct sell
36:06
Unidentified speaker (speaker_1): industry. US will have USAN, USkin, we'll have some private players such as M Oneway. Uh we have Blackmore in Australia, we have Oriflam in Europe. They for a period of time, they all share the same driver. There's really two markets, drive all their growth, growths, Korea and China. These country, these two markets do well. They
36:26
Unidentified speaker (speaker_1): all do well. These two countries don't do well, they don't do
36:29
Unidentified speaker (UNKNOWN): well.
36:30
Unidentified speaker (speaker_1): Now, because they're in different markets, Australia and Europe and US, they report results at different times. If you're smart, then you can basically follow the first one to report, let's say Blackboard reports, they say, oh, Korea and China are not doing well. We're going to gonna get guide down our numbers. You should immediately apply that knowledge to the
36:50
Unidentified speaker (speaker_1): European players and the US players because they're driven by the same two market drivers. And you can make a lot of like free arbitrage money. Now, for some reason,
37:02
Unidentified speaker (UNKNOWN): you
37:02
Unidentified speaker (speaker_1): know, till now, that opportunity, that arbitrage has not been captured by the machines. Maybe because, you know, as again, we're we living live in such a silent world in and the we Wall Street through right now. But going forward, I do think these opportunities can be easily arbitraged away. Because I don't see any reason of why AI cannot capture
37:22
Unidentified speaker (speaker_1): those cross regional arbitrage if they're driven by the same stuff. So those might go away. But you know, the first bucket and the last bucket we talk about the data deserts, where there's no data available for from machine to C see, or the data iceberg or machine C or A. AI can see the the ten percent above the water. They
37:42
Unidentified speaker (speaker_1): don't understand the context, the management style, the his track record, or like, you know, what's going on through Tim Cook's head. Because Tim Cook's not gonna tell you why he wanna bully Qualcomm. Um and those will give you a lot of the edge if we can understand the context.
37:59
Unidentified speaker (UNKNOWN): Yeah,
37:59
Prashanthan: absolutely. That's quite interesting. And uh we'd would love to understand how you're uh looking at the current rising uh debt in the industry because of the massive CapEx. Uh and previously in the conversation you mentioned that there are companies that are kind of still very safe given the amount of cash flow they're generating and will not be so much
38:19
Prashanthan: affected by that. But a big chunk of companies uh that are in the AI space uh sort of depend on uh raising get debt uh to you know satisfy that capex. So how are you looking at that? Are you trying to stay away from companies that have that dependence on uh debt or or and leaning more towards you know the safer bets uh
38:39
Prashanthan: who have more cash flow and can sail through these times?
38:43
Unidentified speaker (UNKNOWN): Yeah
38:43
Unidentified speaker (speaker_1): so you know I first want to clarify we're we're not necessarily looking for anything safe or not safe. As investor, investors, we're always looking for the good risk for, right? Which is really about if we're right, how much return can we make? If we're wrong, how much money we're gonna lose, right? We're gonna avoid permanent capital loss. That's the key point. Now, going back to your question, I think it's
39:03
Unidentified speaker (speaker_1): a very topical question. I was a credit, a massive Wall Street credit conference just last week, even though we don't really do a lot of credit. But you know, we like to go and hear how the credit investment investors think about all the issues that we think about and just hear the view. That's very useful. From our our humble view, and
39:23
Unidentified speaker (speaker_1): we talk about a little bit in our three Q later, the debt market funding the AI CapEx, we think that's really a bridge solution. Uh it's Is really not a fundamental solution to the entire thing. The ultimate solution. to this whole funding question needs to come from real
39:43
Unidentified speaker (speaker_1): downstream corporate customers. Are they willing to spend more money? Right. So think about we talk about Micron top of the last guy, the upstream infrastructure guys already making a lot of money, right? It's been pretty money for the last couple of years.
39:56
Unidentified speaker (UNKNOWN): And
39:57
Unidentified speaker (speaker_1): you think about Millar, Millet, which is the cloud service providers, your Amazon, your Microsoft, your Google, they're making a lot of money too, right? Andy Jassy from Amazon, just like things this past weekend, they talk about, oh, our look at our payback. It's less than two years is fantastic payback. Therefore, if you give me an extra
40:17
Unidentified speaker (speaker_1): dollar, I'm going to invest into this, right? Because we're leasing out compute. Most corporate customers, they don't buy token tokens directly from OpenAI or Anthropic. right? They buy these compute, these tokens through Amazon, through Microsoft,
40:32
Unidentified speaker (UNKNOWN): you
40:33
Unidentified speaker (speaker_1): know, through Google, right? So
40:34
Unidentified speaker (UNKNOWN): I'm
40:35
Unidentified speaker (speaker_1): Andy Jassy, I run Amazon. It's great. I get very good payback. So let me invest more money, right? Because I'm making a lot of money right now. They're very happy.
40:43
Unidentified speaker (UNKNOWN): So
40:43
Unidentified speaker (speaker_1): this really comes down to Amazon's clients, right? The Fortune 500s, your Starbucks, your Doc Chemical, your ExxonMobil, your JetBlue, like these guys, Walmart, are they seeing enough ROI to justify spending more money on AI, burning more tokens, right? If you
41:03
Unidentified speaker (speaker_1): say, oh, you know what, it is generating good return. I'm spending $100 last year. I'm gonna And we're spend spending $200 or $500 next year, then that's great, right? So the gross, as long as the gross curve. eventually match
41:16
Unidentified speaker (UNKNOWN): the
41:16
Unidentified speaker (speaker_1): capex curve, then this whole equation, this mathematical equation, will work. But if the fortune 500 is eventually say, well, you know, it's useful, but it's just okay, return just okay. So last year we spent $100. Maybe we'll spend $150 this year. And you know, the grocery groceries is not there, then
41:36
Unidentified speaker (speaker_1): these this become a bigger question. We will. w we're gonna run eventually gonna run into overbuild and there's not enough um money coming in to support the massive build out and we have to rely on the equity market and the market, right? But the bridge is not is not itself a long term solution. The bridge is to get you somewhere,
41:57
Unidentified speaker (speaker_1): right? That we cannot rely re-own on the bridge for the next like fifty 50 years. So that's something we track very closely. Do we have a good answer to that? We don't. And we don't think this if you ask like 500, Fortune 500, right? 500 COs, if you ask
42:11
Unidentified speaker (UNKNOWN): them,
42:12
Unidentified speaker (speaker_1): do you know your RI using AI if you burn, you know, 100 million dollars?
42:18
Unidentified speaker (UNKNOWN): I
42:19
Unidentified speaker (speaker_1): bet like not all of them are gonna tell you is like we have a different answer. People are still watching, people are still experimenting. Um, but one I would say one good indicator that we're tracking um is Snowflake or Databric. We're a show that are Snowflake for for a few years now, by now. Because now if Fortune
42:39
Unidentified speaker (speaker_1): 500,
42:40
Unidentified speaker (speaker_1): these guys have been doing data like Cal's data warehouse for you for the longest period of time. So a lot of Fortune 500 are not burning through their token through Amazon directly. They're going through Snowflake and use that to route their request or prompt to Amazon to bring that token. So Snowflake
43:00
Unidentified speaker (speaker_1): actually have a very real-time view of the consumption, a token consumption of their Fortune 500, but their corporate clients. Now, Snowflake, we know from their perspective,
43:13
Unidentified speaker (UNKNOWN): their
43:13
Unidentified speaker (speaker_1): AI related. revenue growth accelerated from mid-20s, like about two quarters ago, to now approaching 40s, 40%, exiting 2026. So we clearly see there's a big acceleration
43:28
Unidentified speaker (UNKNOWN): of
43:29
Unidentified speaker (speaker_1): AI spending from their corporate client side. That's a good sign. That's an early sign. You know, they represent a slice of the corporate Cooper rural, and we're even in very early stage. But so we get some like very strong, healthy indicator, but we'll keep watching that into 2027. And I want to give you another example.
43:49
Unidentified speaker (speaker_1): Like if it's like SME, right? So we're looking at Fortune Highlands. I just have a call with like a a big aut Auto Parts dealership. And they're actually talking about they're implementing a voice agent. They call that Carline. car line. Um, so they they're basically a voice agent after 5 p.m., everybody left work. A customer wants to call me
44:09
Unidentified speaker (speaker_1): in to repair his car. And this car line will do a lot of slightly, you know, can make a booking, ask the customer what auto parts, like, do you need a break? Do you need an oil change? And they see they're uh capturing a new revenue, jump jumped to a lot.
44:25
Unidentified speaker (UNKNOWN): So
44:25
Unidentified speaker (speaker_1): they actually were to capture a lot of missed revenue opportunity in the after hours. And most, they realized most customers would go in, they cannot tell if this Caroline, this agent, is a real person, a real woman, or a machine. It's an AI agent. But they could not tell. So they definitely see a lot of ROI, and now they're just rolling this
44:45
Unidentified speaker (speaker_1): up to all their franchise stores nationwide. So that's another good example. They're actually seeing good RLI.
44:51
Unidentified speaker (speaker_1): And they didn't even buy it from any software vendors. One of their guys vibe coded it based on OpenS model. That's about a year ago. So,
45:01
Unidentified speaker (UNKNOWN): you
45:01
Unidentified speaker (speaker_1): know, even in SME land, you definitely see some corporates being creative and figure out the right solutions to already not return on burning that OpenI open eye token to create this, like, you know, after
45:14
Unidentified speaker (UNKNOWN): 5
45:14
Unidentified speaker (speaker_1): five p.m. voice agent car call i generating new sales opportun opportunity for them. So you know there's a lot of potential there. Uh also we're still watching. Uh but um you know we think um
45:25
Unidentified speaker (UNKNOWN): you know
45:26
Unidentified speaker (speaker_1): as we go into 2027 we're gonna see more indicators.
45:29
Prashanthan: Absolutely that's quite interesting. And while it's you know always helpful to hear the outside validation of a thesis, you know, I I think one uh that I personally follow is that to apply it to myself and my own work and uh the companies and and see if AI is actually helping and increase my productivity and I'm willing to spend on it.
45:49
Prashanthan: Uh right. And the answer for me
45:51
Unidentified speaker (UNKNOWN): was
45:52
Unidentified speaker (speaker_1): you could do like a video like an AI agent, right, to do the easy interview and do the podcast and save you more time. Who knows?
46:00
Prashanthan: Yeah, absolutely.
46:01
Unidentified speaker (UNKNOWN): Maybe in the
46:01
Prashanthan: future. Right. But I I do feel the opposite of this is also true that I think when AI takes over everything, I think that personalization and that human angle will will get more and more valuable. Uh and these conversations, like these one on one real conversations, uh I think will be more valuable. S along the same uh lines of what you said about uh,
46:21
Prashanthan: you know, uh the quants quant and AI giving the same answers and crowding, I think this uh this will happen with these conversations as well. If agents start interviewing and agents start answering from the other side, they might uh have the same kind of answers and not as or not as authentic. And I think this human uh kind of conversation will still hold value,
46:41
Prashanthan: I think, in decades to come. So that's my view into that. Uh But yeah, interesting. And you know, you start every uh idea by asking why the mispricing ex exists and how the market will be persuaded, right? How do you tell a real variant view from just being early or wrong? And how long will you wait for the market to come around?
47:01
Unidentified speaker (speaker_1): So I would say on the long side, we're very patient. Um, you know, like Shopify, the thesis played out, you know, after three quarters. Now for three quarters, they're Sort of spending money, but not seen seeing it resolved. You know, obviously my prom problem went from thirty-five 35 bucks to whatever close to a thousand right now. But that did not go through a straight line, right? There's a lot of pullback,
47:20
Unidentified speaker (UNKNOWN): you
47:20
Unidentified speaker (speaker_1): know, during those like last like four years.
47:22
Unidentified speaker (UNKNOWN): So
47:23
Unidentified speaker (speaker_1): we're very patient. Um, you know, as long as the thesis did not change, we're happy to be there. Uh, obviously the thesis change, right? We're involved in uh an activist Activists play on uh the second largest uniform rental companies in the US.
47:38
Unidentified speaker (UNKNOWN): Uh
47:38
Unidentified speaker (speaker_1): They got some management issues. We're trying to get them to be solved. Everything was like going well until Mr. Trump last year he started a trade war. And this company happened to source 95% of their cocks materials from China. So,
47:56
Unidentified speaker (UNKNOWN): like
47:57
Unidentified speaker (speaker_1): was in for 24 hours, the deal is off because the buyer are not willing to commit and buy. buying this company and take on the trade war risk. Right. So obviously that's a factor outside of control. That's a factor we did not consider beforehand. And that factor destroyed our thesis because, you know, the trade war red
48:16
Unidentified speaker (speaker_1): disrupt their cost structure. So we have to, we lost 35% of the money, we have to move up because the thesis no longer valley. So it's really about do we believe the thesis. that we invest in originally still there? It's still there, we're happy to stay for longer. If it's not there, we'll leave.
48:33
Prashanthan: Absolutely. That's quite interesting. And ideas from the blind spot teams uh then have to pass through your 3M filter with this moat management mispricing. Uh most ideas must die there. Which M kills the most ideas? And one of your moat criteria is limited technology. disruption risk. Yet one of your blind spot
48:53
Prashanthan: themes is secular disruption. How do you hold both at once?
48:58
Unidentified speaker (UNKNOWN): Yeah,
48:58
Unidentified speaker (speaker_1): so what two separate questions? Let me answer one by one. The three M we talk about moat management and misprices. First, I want to explain that. So moat is easy to understand, meaning the company have the ability to compound their cash flow with the same organ or even better. return on capital. That means typically they have price and power. If
49:18
Unidentified speaker (speaker_1): they raise price, customer got to pay up. Customer have no choice. That's what we define really price power and remote. The second end, management. When we talk about management, there's really been two things. Number one, the management needs to have an ROIC See mindset when they're investing a dollar into a new factory, a new fab, a
49:38
Unidentified speaker (speaker_1): new market.
49:40
Unidentified speaker (speaker_1): They need to have some thinking around what's the return on that. Now, we don't need them to know the answer because nobody knows. But we need we want them to have that exercise done, at least in their head. That's number one. Number two is that we want to make sure the management incentive are aligned with our incentive, right? So we tip looking at the the pro
49:59
Unidentified speaker (speaker_1): proxy or how they get paid. We want to see the management CEO, CFOs got paid by free cash flow generation, stock return, margin expansion. We don't want simply to see revenue growth, which you see a lot of companies, like a third of their bonus comes from revenue growth. Because we we
50:19
Unidentified speaker (speaker_1): don't want the CEO
50:21
Unidentified speaker (UNKNOWN): to
50:21
Unidentified speaker (speaker_1): spend the money purely for, you know, to get bigger, bigger, and bigger and not earning a good return. on that revenue. So those are the two criteria when talking about management. What kills most ideas really in the third bucket, right? What we call the mispricing, we need to see something that the market's not seeing. Or
50:41
Unidentified speaker (speaker_1): we need to be very patient waiting for an opportunity. Going back to the you Yeah, know we we talked talk about force aviation, right? The we know the companies is where the private, the one public, but there's really no good opportunity because the evaluation with things is not attracting enough. So we've been like watching, watching, and watching very patiently until,
51:01
Unidentified speaker (speaker_1): you know, it's like almost two years ago. Modern water publishes a short sort of report, cannot fix engines, stock dropped 50%, back to seven, eight times, you know, EPITA. EBITDA. We think, okay, that become a very attractive and we know they are indeed. in the effects engines. So that become, you know, that triggered this mispricing. We're happy to get in. So I would say we're
51:21
Unidentified speaker (speaker_1): watching a lot of the companies right now. Uh, good business, we trust the management, management are honest and incentives are lying. But, you know, there's really from an evaluation perspective, it's really not super attractive. So we just keep watching. There's something might come up, opportunity might come up, and we'll we're happy to seize that.
51:39
Unidentified speaker (UNKNOWN): So
51:39
Unidentified speaker (speaker_1): that's to your first question. And second question, in terms of, you know, disruptor or being disrupted, it's really,
51:48
Unidentified speaker (speaker_1): you know, I would say
51:49
Unidentified speaker (UNKNOWN): two
51:50
Unidentified speaker (speaker_1): sub the same coin. Because, you know, investing is a dynamic process, right? It's not a static. So if you think about um, you
51:59
Unidentified speaker (UNKNOWN): know,
51:59
Unidentified speaker (speaker_1): a company can disrupt uh you have an existing player. players, that's fantastic. Uh because that'll be typically that's like a situation going from zero to one, and we're happy to ride that train for that disruption. But and then you go into the second phase, the company that's dis disrupted everybody else,
52:19
Unidentified speaker (speaker_1): they might become disrupted by themselves. I'll give you a classic class example, which is everybody know knows Nvidia. GPU was a new thing. And that's fantastic, right? That challenge is CPU, because GPU, when it comes to AA computing, is much better than CPU. So that's great. NVIDIA obviously has done extremely well at the stock. But then
52:39
Unidentified speaker (speaker_1): this disruptor today is becoming disrupted, or a lot of people trying to disrupt NVIDIA, right? So you have the ASIC ASAC guys, obviously, the most important one being. Google's the GPU, uh Google TPU, which we're talking about, they're trying to challenge NVIDIA GPU because they say we're we're we're
52:59
Unidentified speaker (speaker_1): cheaper, we're more efficient, let's do
53:01
Unidentified speaker (UNKNOWN): that.
53:02
Unidentified speaker (speaker_1): Google's been doing that, and not everybody else, every major client of NVIDIA are trying to do that. OpenAI have a holo panel, which is based off TPU. Um Microsoft's doing their own thing. Amazon doing their own ASIC program. All the major clients are doing that. So the disruptor, Jinsen Huang,
53:22
Unidentified speaker (speaker_1): is facing the challenges like potentially being disrupted by the ASIC world. So and think about CPU, which is interesting. CPU is not having come back because the fine, you know, you do the computing, you're great, better, TPU GPU is better than CPU, but now it's with the agentex AI, you need a lot of CPU to do
53:42
Unidentified speaker (speaker_1): the orchestration, to tell GPU what to do, to allocate the workloads, right?
53:48
Unidentified speaker (UNKNOWN): To
53:48
Unidentified speaker (speaker_1): do the orchestration. And GPU suddenly, the ratio went from four to one, meaning, you know, uh one CPU can just handle four GPU in the agenda world. Now the ratio has moved to two to one to one. to one. So parity, which means now CPU's growth is gonna be much higher than
54:08
Unidentified speaker (speaker_1): GPU growth. So everything, especially in in the world we live in today that changed so much, everything dynamic. We need to watch like, you know, keep watching them. You know, sometimes the disruptor become disrupted. And sometimes the the one that's that been being disrupted suddenly they can come back and become be a disruptor, which makes our our our investing job
54:28
Unidentified speaker (speaker_1): very interesting.
54:30
Prashanthan: Absolutely, completely agree. Uh you call this a private equity like public portfolio, concentrated 18 plus month holds with management incentives as a core core criteria. What did Morgan Stanley PE teach you that most public market investors never learn? And how do you get PE level access to management when you own a minority
54:49
Prashanthan: stake?
54:51
Unidentified speaker (speaker_1): So two questions really about, you know, like what's the manager having spend years in private equity, what does it give you? And versus the public private equity, private market, thinking how to apply it to that to the public. So I will I'll address the second question first, which is the the private equity mentality
55:11
Unidentified speaker (speaker_1): and how you apply that to public market. I think the key is really to be patient, right? Because a private equity investor, you always host that for easily seven to ten years. Right. That's your time horizon by default. You can never say, hey, let's buy this company and sell the next month. That just doesn't happen. Most of the time you're
55:31
Unidentified speaker (speaker_1): buying a mature business, you need to make a lot of the changes internally and be patiently wait for the that change to show up. And that could be a few years down a row. So that gives you a very patient approach, which I think, you know, works well with you know our our more on buffet way of thinking, since we don't use leverage. We just need to be very, very patient
55:51
Unidentified speaker (speaker_1): for that earnest power to show
55:53
Unidentified speaker (UNKNOWN): up
55:53
Unidentified speaker (speaker_1): and compound and translate to free free cash flow, which you can eat at the end of the day. So that's number one. And number two is which I mentioned a little bit earlier. the focus on free cash flow. In private equity land, multiple matters, but at the end of the day, it's really the free cash flow that payoffs your debt
56:13
Unidentified speaker (speaker_1): that you can pay off as a pay out as a dividend to recoup your investment. So that's really the key for free cash flow, right? It's not about EPS, it's not about EBITA, EBITDA. it's not about adjusted EBITDA, Ebita, which is really effective a fake number. We really care about the free cash flow. Like how much cash can this business generate that comes
56:33
Unidentified speaker (speaker_1): into my pocket, right? At the end of the day, that's what we care about. Um so I think that really shaped how we think about investing in the in the public market as well. We really care about that free cash flow generation, right? We we typically don't like companies that burn a lot of money, uh uh companies that have a very
56:51
Unidentified speaker (speaker_1): weak or highly leveraged balance sheet. And that's what been, you know, we talked a little bit about in our letter, we've been sure the neo clouds for quite a while because, you know, we don't think it's really a a tech business per se. It's really more like a real estate business and particularly significant leverage. average. When things are going fine, they're
57:11
Unidentified speaker (speaker_1): fine. But when things are not going great,
57:14
Unidentified speaker (UNKNOWN): that's
57:14
Unidentified speaker (speaker_1): that can spell a lot of risk if you have such a weak overlevel balance sheet.
57:21
Unidentified speaker (UNKNOWN): So
57:21
Unidentified speaker (speaker_1): that's to your that's to the the first question there. So you know to your second question around, you know, what
57:27
Unidentified speaker (UNKNOWN): the
57:27
Unidentified speaker (speaker_1): what sort of knowledge or network you build in product equity days that can be helpful for public market investing. It's really about that accumulating knowledge and network that can really help guide your thinking. So it could be aerospace, a lot of private companies there, or it could be upstream, let's say sportsware. And we're talking, let's say, you know, we're talking to
57:47
Unidentified speaker (speaker_1): one of the key suppliers to a lot of the sports brand, including Lululemon and I keep.
57:57
Unidentified speaker (speaker_1): The idea is not necessarily to ask them like, hey, look, tell me whether your company is good long or short, or tell me which of your brand
58:04
Unidentified speaker (UNKNOWN): that
58:04
Unidentified speaker (speaker_1): you work with, your client is a long or short. The idea is more about tell me something emerging trends that in your space that you think are very interesting, but the outside world
58:16
Unidentified speaker (UNKNOWN): are
58:17
Unidentified speaker (speaker_1): not really paying attention to.
58:21
Unidentified speaker (speaker_1): Find those are the most interesting insights we can get from them and then apply that to looking for, you know, uncover opportunities. Maybe, you know, people ignore. A good example could be like 10 years ago when there's new emergent trends called athleisure
58:37
Unidentified speaker (UNKNOWN): in
58:37
Unidentified speaker (speaker_1): the sportswear space, right? Moving away from performance to athleisure. Performance meaning you're wearing sneakers to play tennis or go for running. After Athleisure leisure, meaning you can wear, you're allowed to wear sneakers to the office. to a date, to a bar, to a form where you've been.
58:54
Unidentified speaker (speaker_1): So when we talk to this lane, you know, like Shinjoi International, which works with a lot of the sports brands, and they tell us, hey, look, we definitely see that, not just here in the US, we see that in Europe, we see it in Asia. That's a massive trend.
59:07
Unidentified speaker (UNKNOWN): So
59:07
Unidentified speaker (speaker_1): we had conviction that's a real thing, that's not a fa, fact or expressive express a view on that.
59:13
Unidentified speaker (UNKNOWN): And
59:13
Unidentified speaker (speaker_1): so we started doing more work and remote. realize, hey,
59:15
Unidentified speaker (UNKNOWN): a
59:16
Unidentified speaker (speaker_1): lot of brands are already doing that. There's like one brand who refused to participate. And that turned out to be Under Armor, a US brand. If you go back and read the COA interviews, you'll see him talking about, yeah, we we see competition doing athleisure, but we think that's a fad, it's gonna weigh, we're gonna not waste our time doing athleisure. So we think they made a
59:36
Unidentified speaker (speaker_1): mistake and we put a us short on
59:38
Unidentified speaker (UNKNOWN): that
59:38
Unidentified speaker (speaker_1): and work balance. worked out well. Because even though they're at their performance line products continue to find doing okay. But simply there's like no bigger pie, and everybody's eating that pie, and that pie is growing bigger and bigger and bigger.
59:51
Unidentified speaker (UNKNOWN): And
59:52
Unidentified speaker (speaker_1): our armor just refused to eat that pie, and that's why they underperform all their competition. So we're able to really use that insight to generate that return there.
1:00:01
Prashanthan: Absolutely. That's quite interesting. Ram, it's been an incredible conversation and we come towards the end of it. Do let our listeners know where they can get in touch with you or learn more about what you're doing at Amont.
1:00:13
Unidentified speaker (speaker_1): Oh, sure. I think you have our false contact formation. So, you know, like uh feel free to share with listeners. Um, and you know, we're we're looking forward to continuing the conversation, and learn earn continue continued learning from. the community.
1:00:28
Prashanthan: Absolutely. Thanks so much, Rob, and I wish you happy investing.
1:00:31
Unidentified speaker (speaker_1): Thank you very much. Take care.
1:00:33
Unidentified speaker (UNKNOWN): Bye.
This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights