← Enterprising Investor
Enterprising Investor
Ken Laudan: Investing in the Age of AI
00:25:18
Join Mike Wallberg, CFA, and Ken Laudan, portfolio manager of Buffalo Fund's Blue Chip Growth Fund, for a thought-provoking discussion on the future of AI in investing. Ken shares his contrary view on AI's transformational potential, drawing on his...
Loading summary
Transcript39 lines
- [00:00]
Mike Wahlberg
Get ready for cfa Institute Live 2025 with free power packed webinars designed to inspire and inform. Led by industry experts, these concise sessions tackle game changing topics like how AI is transforming investment strategies in emerging markets.
- [00:16]
Ken Louden
Don't just show up, show up ready.
- [00:17]
Mike Wahlberg
Gain the insights you need to drive.
- [00:19]
Ken Louden
Meaningful conversations this May in Chicago.
- [00:22]
Mike Wahlberg
Claim your spot now@cfainstitute.org Foreign welcome to the Enterprising Investor, the flagship investment podcast for CFA Institute. I'm Mike Wahlberg and I'm really looking forward to my chat today with Ken Loudon because, well, we've had many people on the show over the last year who work in many different corners of the capital markets. We've had CEOs and global risk managers, fund managers and finance researchers and LLM providers, and on and on. And AI is so pervasive and figuring out its applications is such an urgent item on their desk. And listeners want to know as well, like how this new new thing is going to transform their lives and when. So it's natural. It's taken up lots of mind share with people and audio time. Honestly, on this show and countless others, well, Ken is a bit more circumspect about the whole thing. He has a contrary view, one he shared in the Wall Street Journal a few weeks back, and as a value investor myself, I'm keen to hear his thoughts. Ken has over 30 years of experience as both the sell side analyst and buy side portfolio manager, and he is now the portfolio manager of Buffalo Fund's Blue Chip Growth Fund, which Morningstar rates as four stars across three, five, 10 and 15 years. Welcome to the show, Ken.
- [01:41]
Ken Louden
It's true to be here Mike. Let's do this.
- [01:44]
Mike Wahlberg
So Ken, before we get deep into the Nvidia weeds, tell me where we are in the markets more generally. It's been a pretty crazy year with lots driving it.
- [01:51]
Ken Louden
Yes it has, and I would just start off by saying that the market anytime is both hard and humbling. But I think it's a very tricky time for the market today for two reasons. Number one is there's a shifting narrative that's at least starting to emerge regarding a possible regime shift towards the other 493 stocks in the S&P 500 and away from the so called cool kids crowd, also known as the Magnificent Seven. And investors generally within the large cap categories and certainly within the Mag seven up until three weeks ago had become, in my opinion, overly complacent about how this massive AI infrastructure and platform investment that started in earnest about 18 months ago was going to be monetized. Investors really just assuming if you build it they will come. Maybe that's starting to change now some 18 months into the AI infrastructure build. The second aspect that makes this tricky is that we sit on the precipice of argu, arguably the biggest earnings week of the year. With Microsoft, Meta, Amazon, Apple and AMD all reporting this week, there's big questions and uncertainties on the AI CapEx spend going forward into 2H24 and certainly into 25, as well as getting a better handle on the macro backdrop drop to see if earnings for the broad S and P match the shifting narrative of accelerating earnings into the back half of the year. And I'm a bit dubious on the acceleration, but we'll see.
- [03:26]
Mike Wahlberg
More generally though, I mean we're operating today against a backdrop that's had one point in the rate cutting cycle, had other darlings I guess that have come and gone since sort of November of last year. You've got the OZ epic trade that's been on and off. AI is sort of one of many and we've talked before a bit about the equity risk premium as well and what that looks like. So through a macro perspective, how are you seeing conditions for the performance of equities in general before we get sort of into the AI specific?
- [03:56]
Ken Louden
Yeah, so if you look at the Max 7, generally speaking, despite the 8% pullback we've had over the last couple of weeks, they're still up a eye popping 36% year to date and collectively traded around 27 times the the calendar 25 estimates translates into about a 3% free cash flow yield. But you did touch on it and so I have to address it is this whole thing about equity risk premium. The Magnificent Seven currently have an equity risk premium of about negative 70 basis points even after the pullback, albeit that's down From a negative 130 basis points Equity risk premium prior to the sell off. And there's different ways to calculate equity risk premium. We can talk about that too. But regardless of whatever method you use, the risk reward implied by the equity risk premium is just provocative, I guess I would say. And I think that's one of the reasons why we're starting to see a little bit of this regime change occur with a thrust toward these other 493 stocks that just have a better risk reward. I think I don't have it in front of me, but I'm pointing off a road from last week. I think the equity risk premium, for example of the equal weight S&P 500 is somewhere around 210, 220 basis points. Still below, by the way, what equity risk premiums have historically run at, but certainly at least offering right, something that's more traditional relative to the hype cycle within AI.
- [05:30]
Mike Wahlberg
And the implication being then when you've got a negative equity risk premium like that, that it's unlikely to outperform on a go forward basis specifically against bonds, but probably also against the other 493.
- [05:42]
Ken Louden
You bring up a great point I think if you. Well I know I went back and looked at this about three months ago. Some 220 plus years of equity markets, there have only been 59 times when the S and P index of large companies had a negative equity risk premium. And virtually in every instance of that, the bonds outperformed equities in the subsequent year. And I know the Max 7 is not the S&P 500. It really has been the last year and a half because without the Max 7, I mean the S and P would, would be negative. So anyway, this is not to say that AI won't be big and transformative and disruptive. It will be, but I'll address that maybe when we start to cover AI. It just means that, you know, and, and I do think the market's going to continue to ride the broad shoulders of Nvidia, at least for the time being. There are just some accumulating data points that the important technology phase transition from AI enablers, which are the infrastructure and platform companies, to the AI adopters are going to be more protracted than what either investors or perhaps even the AI enablers like Nvidia and Google, Amazon and Microsoft have believed up to now.
- [06:56]
Mike Wahlberg
So to set the table a bit for listeners, I just had a look over the weekend. We're recording on July 29th in the morning. So this will come out in a couple of days. But from the peak they were all sort of July 5 to July 10 for Nvidia, Meta, Alphabet and Microsoft. They're down 16, 13, 12 and 9% respectively from those peaks, but still up healthily from, you know, on a year to date basis. Nvidia specifically is up 135% through last, the end of last week. The others up kind of mid teens to mid-30s. So definitely a material check back. But despite that, you know, we, we've come a long way, we've had significant growth and strength in those names. So let's talk about the underpinnings of that growth and how sustainable that that is. And you said to me that AI is neither transformative nor revolutionary today, far from it. So tell me how you arrived at that conclusion and what it means for the various players and the market in general. Yeah.
- [07:55]
Ken Louden
Can I pose a couple of caveats? Here is number one, as I started to say, I think AI is eventually going to be transformative and revolutionary and live up to the hype. It's just really the flying. The ointment for me is just AI today is most. The architecture of AI today is mostly large language models or LLMs is what people have adopted. And those are really, really good for automating, coding, enhancing language based pain points such as call centers, customer service call centers, and maybe automating obviously certain steps outside of the IT area and maybe marketing and sales. There just hasn't been a killer app on the horizon that has been based off of large language models. And for us to really see the inflection of demand, I think the AI architecture is going to really have to evolve into something beyond large language models, which as the technology would tell you, it's really good in solving language problems and it doesn't really, it does answer questions, but it doesn't have really action oriented steps to it. And I think that's where we, we need to go before AI demand really inflects.
- [09:07]
Mike Wahlberg
So Ken, do you think it's basically folks got romance to a certain extent? Like the excitement that got behind the AI revolution was more not totally based on this. Obviously it's doing a lot of actual calculating and computational heavy lifting, problem solving in certain ways. But the fact that it could write and you know, to a certain extent speak in a way that was replicate of human speech, that people felt, okay, we're there, we've arrived, it figured out how to write. And so therefore, and this is coming from a writer, you feel like, okay, wow, we've made leaps and bounds ahead and perhaps we're not quite as far ahead as we thought we were.
- [09:46]
Ken Louden
Yeah, I think that's typical in any sort of transformative potential technology. I know I was romanced by the ability to sort of get quick answers to rather complex questions. I think what's happened since then, Right. We've had this thrust of multiple general purpose AI models that have come out and they all seem to have sort of the same features or bugs, and that's with something called hallucination toxicity. And that's basically making up an answer to a question that is invariably wrong. They have questionable cost benefit, they possess key uncertainties around data security and privacy. And we're starting to hear Anecdotally, at least, where while we saw a 50% increase in coding and generating large enterprise software, that some of that ROI is closer to maybe 20 or 30% and not 50%. All this will evolve, of course, but these are sort of the fits and starts of any new technology. And for it to be adopted broadly by the enterprise and for consumers for that matter, you've got to get rid of these bugs and have more consistency in maybe generating the type of productivity savings. I just guess I'm now a little bit more skeptical that large language models are going to be able to scale like we thought they were and be able to continue to drive strong return on invested capital from these investments. And I think that's one of the reasons why we have seen adoption be much more muted than what either investors or the hyperscalers probably thought six months ago.
- [11:19]
Mike Wahlberg
Yeah, well, I feel like part of it is as a feature of the models themselves, that they've been sort of coded to fill in the gaps when they don't know the answer. And part of it as well is just some issues with the data itself. And I guess on the data question, actually I wanted to ask you about, you know, we heard there's some dodgy stuff that went down with training the original chatgpt in terms of scraping data that they weren't supposed to. And now that the light of day is on, is on the model and everybody's looking at it, everyone wants to make sure that they're being paid for their copyrighted materials and so on and so forth. So how does the oversight on what data is scraped and what is paid for that data in the future? Like, how does that affect the quality of the data1 and the cost ultimately to users to access these models? Because the data is no longer free.
- [12:05]
Ken Louden
Yeah. And it might get more. More expensive. Great question, probably a bit more technical that I'm able to answer. I would just tell you I had a large enterprise software company in our offices about 10 days ago and what they were seeing is that employers are very concerned about this hallucination rate, or what they call hallucination toxicity, is running as high as 35% in some cases. And where they thought the most traction is going to be was more on lower cost small language models rather than large language models. And this is sort of using a company's own data and training the algorithm on a limited amount of data but very important data so they can eliminate these hallucinations that seem to be, again, a feature rather than a bug of these large language models. I think it's just enterprises are being very thoughtful and methodical about how they're going to implement these AI workloads and they're doing it incrementally. So it's rather than revolutionary. And I think we all thought it might be a revolutionary sort of thing and we really haven't moved beyond. So we started with search, right? We've had search for 12 years, 20 years, I guess, maybe with Google, but certainly where Google was the dominant search engine. And then large language models came along and we had answer. And the next phase is going to be action. We're a ways away from the action component.
- [13:32]
Mike Wahlberg
Can you give us an example of that? What do you mean by action?
- [13:35]
Ken Louden
So think about a AI tool that knows everything about you, your likes, your dislikes, your taste, your preferences. And it can remember that, right? You don't have to constantly prompt it and remind it and it can actually do something like order your groceries at 10:30 on Monday morning and have it delivered by 5:00 that afternoon and saving you great time and effort. Great.
- [14:01]
Mike Wahlberg
So it's thinking a little bit more. It knows you, it prompts you on that stuff. So we've got. So what I'm hearing then is we've kind of gone from search to answer where the technology is kind of helping fill in the gaps for you. And I know when we've talked before, you, you said that there are sort of inflection points that you're looking for with AI in terms of user adoption for it. So this I would say sounds like this sort of a search to answer sort of as one inflection point. What are other inflection points that would need to happen for these companies to sort of grow into the earnings expectations that investors are building in for them?
- [14:36]
Ken Louden
I would suggest a major inflection will occur when AI has the ability to substantially augment human interactions through more cognitive reasoning. And we just talked about actions and that just doesn't exist today with LLMs.
- [14:51]
Mike Wahlberg
Maybe it would be helpful to step back for a second, just look at what we're talking about in terms of the scale of the investment by the, by the hyperscalers and how much folks are building into their valuations based on AI happening. So obviously for Nvidia and amd this is about customer demand and then obviously for the others it's turning that capex into something profitable. So what's been put in, what's coming up and how worried should investors be about that all getting pulled forward?
- [15:19]
Ken Louden
Yeah, my big AI concern is around monetization timing. There is something going around investment Circles called the $150 billion AI questions and there are other variations of this like the 600 billion dollar AI question. But focusing on the 100 billion, 150 billion as a starting point represents a reasonably good estimate as what's been spent on AI infrastructure since January of 23 via GPUs, networking equipment, servers, etc by the US hyperscalers. I guess the essence of the question attempts to answer when the four US hyperscalers you point out, Amazon, Microsoft, Google and Meta will start to base their capex AI spend on actual end user demand, not on what they project or hope it might be at some point in the future. I guess the takeaway to all this, and I could certainly be wrong, is that nobody, including the hyperscalers really know when end user demand adopter demand will inflect. That most likely will be determined by the pacing of new capabilities and features produced by the AI models themselves. But the cost of invest, not investing in AI may exceed the cost of investing even if it means a drawdown in gross margin earnings and free cash flow for the hyperscale.
- [16:32]
Mike Wahlberg
So they're just trying to stay ahead of the curve basically, and they're throwing money at it. And there must be, I mean those four must be some huge percentage of Nvidia's sales.
- [16:41]
Ken Louden
I would imagine about 45% of Nvidia sales in the last quarter. And I think they all view it as AI as existential. They realize being the third or fourth largest player in this sort of transformation, technology transformation is going to not accrue the level of profits that they should be able to capture. I mean Microsoft missed the mobile opportunity, right? And so they've missed out on 10 years of significant earnings and revenue growth and none of these companies want to do that. However, there is, I think, a limit to what investors will accept in terms of investing in front of the demand curve where you can start to monetize this. And you know, I don't think it's this quarter or next quarter, but I think as we get into 2025, if we don't see the sort of demand start to manifest, then I think you're going to see investors start to push back on the hyperscalers that they're going to have to be more measured with their AI investments until they can better quantify when they'll be able to start to see these drive return on invested capital higher and not be dilutive to.
- [17:50]
Mike Wahlberg
Margins makes your job that much harder. Hey, so you're. When you've got to sort of figure out what's going to happen over the next year or two or five. But you know, the stock's trading on sentiment and people's perceptions of who's winning and not winning. It's not something you can eat.
- [18:05]
Ken Louden
You know, it's interesting, we had a Google report last week and Sundar talked about this very issue about investing in front of the demand. And he mentioned that he thought that the risk were greater to not invest than to over invest. And it'll be interesting to see if this week, as we talked about earlier, such a big week for a number of reasons in earnings, whether Microsoft, Google and Meta say the exact same thing, Investors will be patient near term, but they're not going to wait two years for demand to inflect before they start to really question and discount I guess the multiples of the ramifications of over investing.
- [18:45]
Mike Wahlberg
I think a question is kind of a random one. I'm just thinking about power usage. I know within crypto you've got the debate went on and I guess it's continued still because not all of the cryptocurrencies are on proof of stake, still on proof of work. So the power needs are continue to be high in pockets of that segment. But all this computing power has to be putting a strain on electricity supplies to markets. How big a risk is this to AI sort of getting there.
- [19:11]
Ken Louden
You know, it's going to be a bottleneck. And despite the hallucination toxicity, despite the questions about cost benefit, the power consumption issue is unresolved. I believe the numbers coming out of Nvidia over the next five years is that compute power is going to grow at like 250% a year while the growth in the grid capacity is going to be 2 1/2 percent. It's unclear exactly how you reconcile that difference in being able to generate the type of power that you need to run all these, these, these AI workloads.
- [19:46]
Mike Wahlberg
This is on a grid that's already fairly strained. Right. Aging and facing new demands from EV chargers and all the rest of it. Right.
- [19:53]
Ken Louden
And next generation chips are going to be more, much more effective on a per unit basis. Right. And generating less absorption but just the overall use, projected use anyway of these GPUs is going to drive grid demand dramatically higher than what it's able to produce at two and a half percent. I think I saw Semiconductor Institute came out, this is maybe three months ago and indicated that they thought data center consumption was going to be up to 19% of grid usage. From 4% today. And that's got to come from somewhere, Mike. And I'm not sure where that, where that, where that comes from right now. Yeah, there's some band aids you can put on like nuclear power plants and you know, locating AI data centers behind the fence. But those are just what they are, they're band aids. They're not a sort of holistic solution to solving this power consumption issue.
- [20:46]
Mike Wahlberg
So other than being cautious around these valuations, like how should investors play this theme given, given your view?
- [20:53]
Ken Louden
Well, I, I think it depends on your time horizon. We tend to think about investments over a three to five year period of time. So I own all of the Max 7 names, right. And as I mentioned earlier, this is a game of kings and emperors. And so you need scale and capital, you need scale and distribution and you need scale in data. And the hyperscalers all have that. I think the critical question will be is what's that gap between the investment in AI infrastructure, the enablers which we've seen it's been robust. Nvidia's Exhibit A versus the adoption of that infrastructure and use of that infrastructure by large commercial enterprises. And I think at the end of the day, to simplify what may be sort of complex is that gap which we always knew was going to exist, the gap between the enablers, the AI enablers and the AI adopters is looking like it's going to be bigger than what we had anticipated it to be. And you know, we discussed sort of why the large language models, the hallucination rate, the cost benefit questions, privacy, security. And so I think it makes sense to sort of have more of a defensive portfolio. These know it names don't have to be the Magnificent 7 doesn't have to be 45% of your portfolio. It's. I think it's actually 55% of the Russell 1000 Growth Index and it doesn't need to be 35% of your portfolio, which is about what those Max 7 are as a weighting for the S&P 500. You can own these names, just own them in probably more pragmatic weights to where you're not overexposed. In lieu of a potential inventory correction that could occur with GPUs is demand sort of remains unclear and focus on defensive oriented names. I think in both Healthcare seems like it's a very interesting place to be. We're overweight. Healthcare, we're overweight. Industrials, particularly those would have, that are leveraged to the electrification of everything. Aerospace, water purification, we have a water problem. By the way, which is sort of a super macro trend that probably doesn't get enough attention. That's probably where I'd spend some time is I think this narrative of moving away from the max 7 at least for the next 3 to 5 months probably has some likes to it. And so we're sort of tweaking our portfolio to be better positioned to outperform in this market where we're under a regime change and could start to benefit some of the names that have not benefited heretofore. But that didn't mean you don't also want to own some exposure to AI super cycle names like Nvidia, like Microsoft. I think Microsoft's gonna be a long term winner, that Google probably wins long term. They got some challenges and Amazon sort of an interesting play for other reasons other than AI, just based off the margin leverage that they're getting on the E commerce side. So there's reasons to own those days but I think there's also reasons to move away from having a concentrated weighting within those names that have so dramatically outperformed over the last 18 months. And that's how we're, I mean it's more than what we say, it's, it's how we're trying to position the portfolio. Certainly roll out the door benchmark.
- [24:09]
Mike Wahlberg
Good pieces of advice there in general to outperform long term, have a long term view and be, be willing to see through the, through the noise and, and also to manage your weights portfolio recognized when you've got concentration risk and concentration of risks in one area. So those are a couple of great takeaways from this. So Ken, we're coming to the end of our, of our chat today. Going to hit you with the two part question here. What was your first job in the industry? And if you could go back and take yourself for coffee on your first day, what key piece of advice would you offer yourself?
- [24:39]
Ken Louden
So my first job in the institutional investment industry was a sell side analyst covering healthcare service stocks. And I think if I was to take myself to breakfast it would have been. I loved investing in healthcare for almost a 30 year period of time until I became a generalist four and a half years ago I didn't realize what I was missing and I probably would have been more focused as a generalist rather than a sector specialist. For a large number of years I've been a growth investor either way but I sort of would been very interesting to be able to invest in the next big thing from a generalist perspective over the last 25 years.
- [25:17]
Mike Wahlberg
I've been speaking today with Ken Louden, portfolio manager of the Buffalo Fund Blue Chip Growth Fund. Thanks for sharing your thoughts with us today. It was a great chat.
- [25:25]
Ken Louden
Appreciate it, Mike. All the best.
- [25:27]
Mike Wahlberg
I'm Mike Wahlberg, and this is B, the enterprising investor.