EP.45 - Palantir ($PLTR): Is the Leading Growth Rate Sustainable? With Gil Luria, D. A. Davidson

As their revenue growth accelerated, retail investors figured it out first, bid it up all the way to 200. Institutional investors were always caught a step behind, including most of the sell side. The bull case is that the stock's valuation is now, you know, 50 times forward cash flow, not 200, 2% yield, and they're growing 90%, 93%, up from 85 last quarter. Next question is the future growth rate expected and how reasonable is that? Can they continue at 90%, 50%, 20%? It's pretty straightforward to underwrite 50 to 70% CAGR for at least three years, probably five years, and you start getting to really big numbers. Last thing I'll say is that they're winners, and people like winners. Customers like winners.

Thank you, Doug. Love your podcast. I think you're bringing a really important voice to the conversation. I enjoy hearing you and I appreciate you bringing me back. I have to say, your "Is OpenAI in Trouble?" episode from last December is one of our most viewed ones. So thank you for coming on and sharing your opinions, whether they're controversial or not. I know the audience loves to hear them.

They accelerated at a time where nobody thought they could, because accelerating from 85% growth is pretty hard, and they did. They're now at 93% growth, which is basically saying like they're doubling over last year. And the strength was not just in the commercial, where they're going 150%. 90% growth in government tells you that both engines are working remarkably well. We thought that for a while now that the commercial business would become bigger than the government business, just mathematically, and it's not happening because the government business is growing so phenomenally well. You know, let's back up. Palantir, at the dawn of the AI era, was an interesting company that wasn't very well understood, was underappreciated, and because of who they are — and we could talk more about that — because of what they do and how they do it, they were perfectly positioned for AI. And so they had a massive run, right? The stock went from 20 to 200 as their revenue growth accelerated. Retail investors figured it out first, bid it up all the way to 200. Institutional investors were always caught a step behind, including most of the sell side and us included. We're all caught a step behind and we're like, "Well, yeah, no, they're doing well, but it's too expensive for who they are." And then there was a point about a year ago, as they reported the second and third quarters of last year, where they had such a remarkably good quarter that a lot of us said, "We've never seen anything like this. We need to do more work to understand what it is that Palantir does that makes them so special, because clearly we are all missing something here." And we did that. We started going to their AIP conferences. We started talking to their customers. We started talking to their partners. We spent a lot of time with the company to understand what it is that they're doing very well. And again, we can get into that. But as we realized what they're doing really well and gotten really enthusiastic and started saying that this is the best software company in the world, maybe the best company in the world, the stock was trading at 200 times cash flow. And so for a sell-side analyst like me, that's a non-starter. And so I said very positive things, but we stayed on the sideline because it was too expensive. Then two things happened since last year. The cash flow doubled and the stock went down 40% at the beginning of the year for a variety of reasons that again we could get into. And so we got into this situation. We started talking to institutional investors and like, yeah, we're dusting this thing off because now that Snowflake, Datadog, Shopify, CrowdStrike are trading at 50 times cash flow, all 50 times cash looks interesting because they're growing twice as fast with twice as much margin. And we did the work again and we came to this conclusion that wait a second, there's one big threat that we agree with that we really want to think through, which is are OpenAI and Anthropic going to start hiring forward deployed engineers, send them into companies and do what Palantir does? We worked through that threat. We actually believe it's overstated. It's not really that big of a threat, and the opportunity is very big, which is when we decided to say, okay, at 50 times cash flow, other high growth software was 50 times cash flow. Palantir growing twice as fast, twice as much margin, time to step in. Institutional investors started stepping in, and then they saw a quarter where nobody expected Palantir to accelerate again, and it did. So that's what we saw in the quarter.

That's one piece. So let's talk through the progression here. So the first thing is to realize this is a company that comes up with comprehensive solutions, end-to-end solutions, right? They don't show up and say, "Hey, subscribe to my software, and if you buy it, hire somebody else to implement it, maintain it internally, do the customization as you will, all I care about is the subscription," right? Almost all other software companies over the last decade have gone to a model where they didn't have any professional services. They just wanted the SaaS business, and why? For good reason, right? It's a business that scaled very fast at very high margin, and it added a tremendous amount of value during the big software boom through 2021. But because they left professional services, they didn't have that muscle. Palantir has built that muscle over 20 years, and so it can come in, send forward deployed engineers — which is a fancy word for consultant — figure out what the business needs, then bring the ontology, which is a very unique approach that only they have, to map the business to a data schema. They're very good at that. They have the software that does that. They've been doing it for a long time. And then — and this is why the AI era matters — then you can take that data that you brought from all these different places in a comprehensive way and have a very powerful decision-making engine, which AI has made so much better. And what that means is, because you're — and then by the way, Palantirians stay, some of these forward deployed engineers stay to make sure you're generating the value. And if you're generating a lot of value, Palantir captures part of that value, outcome-based pricing. So it's that whole story. And what that story means right now is that Palantir doesn't go to the IT department. Palantir does not sell to the CTO. Palantir sells to the CFO and the CEO. They go around IT because IT would rather do all this in house. But the reason the CEO and CFO want to engage in this conversation is that they're not doing a good job, certainly not in the era of AI. Everybody's trying to figure out AI. Nobody's finding themselves except for Palantir customers, which means that now you have word of mouth, which is why they have more inbound calls than they need to make outbound calls, which is why you get leverage on the sales line, which is why you can have 62% operating margins, which is more than 15 points better than any other software company. So this whole thing works together. And the last thing I'll say is that they're winners. And people like winners. Customers like winners. And they have a backbone. They have a mission. They want to protect Western civilization. They have receipts from the battlefront. And this gets people excited. Not everybody — we can talk about the pushback — but it gets a lot of people excited and wanting to work with Palantir. So they get to cherry-pick customers, deliver results, and win. That's what we learned over the last year.

That's exactly right.

It's hard. A few things. First of all, these forward deployed engineers are the best of the best. You know, I grew up in this business in an era where McKinsey hired the best of the best, and then when companies wanted to transform, they'd hire McKinsey or BCG. That's not where those people go now. They go to Palantir now. So Accenture could send you consultants, but those are Accenture consultants. Look, I worked at Deloitte. Deloitte can send you consultants. That's great. But if you want to transform your business, you go to Palantir now because McKinsey doesn't give you all the answers because they don't have that technology part. They don't have that software that they built over years with that ontology approach that can make that work. So that's one, is being able to hire those forward deployed engineers. Second is again, that ontology approach is a lot harder than it seems. Again, they've honed it over many, many years of being at the customer site and figuring out problems and seeing how to make that approach work. And then there's the brand. They've developed a great brand. All those things are hard to replicate.

Most of our computer infrastructure is based on databases with columns and rows, right? So we give things names and we assign them a serial number, and then we use a sea of terms to define all the things and we try to generalize it. Let's come up with the set of all the things that an industrial company has, and all the types of products they have, and all the types of customers they have, and all the types of inputs they have, and we have this massive, massive map, and then we put that in a database. And then companies, as they grow, they end up having many, many, many of these databases, but all the definitions are slightly different. And so big companies especially have their data in a lot of different places that don't interconnect, because all those terms are defined differently and defined in ways that are really generic — generic to an industry or generic even overall. And that's, you know, what the default is for companies is to dump it all into Snowflake or Databricks and then try to fish something out of there. That's the default. What Palantir does — and we can go, the historical story is really interesting of how they got here, we could get into that as well — but what they do is they go into the business, again, they send forward deployed engineers and they talk to them about that actual specific business and they say, what's important in your business? What are the types of products that you care about? What are the inputs that are most important? What does your customer look like? What is their profile? And then they can look at those specific terms that are specific to that company and weave all of the company's data into those terms. That's the term what ontology refers to, is mapping data to how the business actually works, to how that specific business actually works. When you do that, you now have the data in a meaningful way that you can apply AI to to get your important answers. That is very, very hard to do. There's really one — there's a good case study to tell you how hard this is. It's a company called C3.ai, the ticker is AI. Siebel, of Siebel fame, sold to Oracle many years ago, decided that he could do what Palantir does, and he hired great people as well, and he tried to replicate the model, and it's an absolute disaster. Remember how I said that Palantir's growing at 93% with 62% margins? C3.ai is not growing and loses money. So it's much harder to do than it seems. It's been built over many years. If you want, we could get into the story of why it was built.

9/11. So, Peter Thiel and Joe Lonsdale and Alex Karp were very frustrated with what they saw after 9/11, that all these different agencies had information about terrorists and they didn't share it because they wanted to protect their sources, their information. So they didn't share it. So we couldn't connect the dots. So we couldn't prevent that attack. And they were hellbent on building a system that would not let that ever happen again. And it took them several years. I think the story is they only got that first meeting with the CIA in 2008, but they got it done. And now they allow all these different agencies to share the most secret, confidential information without compromising it. So the FBI information gets shared without anybody knowing what the FBI knows or how they know it. The CIA, the NSA, all that information, local law enforcement, you can combine all that information in a way that you can connect the dots without sacrificing the confidentiality of any of the individual databases that these agencies maintain. And we've prevented numerous terrorist attacks all around the world and in the US because of that. But that sounds exactly what I just described for commercial, right? Taking disparate data from disparate sources, legacy sources, databases you got through acquisition, different types of databases, and being able to combine the data in a meaningful way so you could make decisions. So again, that's why it's so hard to replicate what they're doing because they've been doing it for 20 years. So they've perfected the ontology layer, and I assume some of that as you go from company to company, you don't have to build it from scratch. You just kind of change it a little bit because you use the same databases, and then there's the customization part that they added in that makes it sticky and never removed. The most recent net retention rate was 157%, which is to say they're still getting most of their growth from their existing customers in spite of the fact that they have many nine-figure customers, and they're still growing with those customers because when you solve the really big problems, you add so much value that even if you just capture part of that value, you grow with your customer. And again, that's another unique aspect that they can now do. And this outcome-based pricing is relatively unique to Palantir because if you're any other software package software company, your Salesforce, your SAP, your Workday, you're just selling package software, even if you are helping companies save money or make more money, there's no way you can do that, and there's very little way for the company to keep track of that. If you're Palantir and you have people throughout that are helping you get the results and then helping you measure the results, you can say, "I delivered $100 million worth of value this year. I want to get paid 20," and they can say that, and they can put that in a contract, and they can get paid 20.

Their chief revenue officer was general counsel before. He's a lawyer, right? Because the key skill here isn't a marketing skill, it's a contract skill. How do I make contracts with a customer that when they win, I win? Right now, the bull case is that the stock's valuation is now, you know, 50 times forward cash flow, not 200, 2% yield, and they're growing 90%, 93%, up from 85 last quarter, so it accelerated, and now its valuation is in line with the comps. So you can say, hey, 2% free cash flow yield is now by market standards reasonable. The next question is the future growth rate expected and how reasonable is that? Can they continue at 90%, 50%, 20%? Where do you expect growth in the back half, next year, the year after, and what do you think the stock has discounted for long-term growth? It's pretty straightforward to write 50 to 70% CAGR for at least three years, probably five years, and you start getting to really big numbers. And so from a valuation perspective, to me — but whatever you give these, again, high valuations make me nervous in the first place, which is ironic since I'm a tech analyst. But if you just take that cluster of other successful winning software companies — CrowdStrike, Shopify, Datadog, Snowflake — and say, okay, which one of those is trading at the highest multiple, Palantir should trade at a premium to that. That's going to be my rule of thumb, because you know they move around as we speak. Datadog's down a little bit today. You know, Snowflake's out, CrowdStrike's out. But that cohort — Shopify's up a lot yesterday. If you take that cohort and you look at where they're trading, that tells you how much the market is valuing growth right now. And again, I would argue that Palantir should trade at a premium to the group and to the highest member of that group.

Yeah, let me make an even stronger statement because, you know, part of understanding Palantir is understanding Dr. Alex Karp, right? He is very outspoken, to say the least. Highly recommend watching Palantir earnings. They're very entertaining, and he talks about the fact that only his customers are doing well with AI. I'm not sure that's a big stretch. I think everybody is trying to implement AI. I think some companies are having some success, but it's pretty clear — and again, this is why we did so much diligence talking to their customers — is that their customers are feeling like they're getting a big benefit from AI. Because again, you can't just turn AI on internally and ask people to use it and it works. You actually do have to have somebody come in and answer the really important questions about what problems you have, what challenges do you have, what opportunities do you have, and then implement systems that can allow you to address those challenges and to capitalize on those opportunities. And only then can you get the benefit from AI, because AI is such a different paradigm, such a different framework than anything enterprise technology has ever dealt with, that you need a lot of help right now. And just asking your internal IT folks to figure it out isn't working. Asking Accenture and IBM to help you out is only working a little bit. Palantir's customers are talking about real benefits that they're getting from implementing Palantir. Again, that's what creates the word of mouth.

Conceptually, unstructured data is very different than structured data, and deterministic systems are very different than probabilistic systems. Right? I'll say that in a slightly different way. Our companies are used to dealing, again, with an entire technology stack that's just built the way we build computers, with files and folders and rows and columns. And if the system is built properly, the answer you get from multiplying one number by the other is always going to be the same. The query that you make to a database is always going to give you the same answer. And you could count on that, and you've never had to worry about that. In the world of AI, when we're trying to replace tasks that up until now only humans have been able to do, you have to factor for the fact that you're dealing with documents that all could look different than each other, that are unstructured, that are images and text, and the formatting and the structure of those documents is very different. You have to deal with that on the one hand, and then on the other hand you have to deal with the fact that when you make a query you get a different answer every time, right? It's probabilistic. The large language models are next-token predictors, and therefore they usually give you a correct answer, and that percentage is going up all the time, but it's not 100%, which creates a whole world of challenges. So just shifting that mindset from deterministic, structured data to a world where I have to do that as well as probabilistic, unstructured data takes a tremendous amount of work, and all companies are going through this journey. Palantir's customers are running faster than other companies. That's what's happening right now.

Well, let's start with the quarter. What happened in the quarter — and this is going to happen to a lot of other technology companies, it's already started to happen — is AI isn't replacing software. It's not replacing hardware. It's crowding it out. So because companies are spending so much on tokens right now, they have to cut costs everywhere else. So that's not just software, it's hardware, it's professional services, it's everything. And we're going to hear a lot more companies like BlackLine and Verint and Pega and NICE say that we're being crowded out. The spend is being crowded out. We can't sign new contracts because our customers just have a budget issue because they're spending so much on tokens. But what actually happened in the quarter with IBM, I think it got misinterpreted a little bit because the language was a little ambiguous in that pre-release. But what actually happened is they sold less mainframes and less mainframe software, right? It's amazing that we still have mainframes, but that's what they sold less of. And it's just a category that it's easier to crowd out. Like, okay, my mainframe is 20 years old. It's okay if it's 21 years old. I'll replace it next year. And that's what happened to IBM. But what I'm referring to in terms of the differentiation between Palantir and IBM and Accenture and etc. is that what I described for that huge transformation for technology, for AI, should have been the biggest opportunity ever for professional services. Right? Companies can't handle something internally, they have big unstructured questions to answer, they should be going to professional services and getting their questions answered. And yet the growth of those companies in their professional services was incredibly anemic. And again, it's because they have armies and armies of consultants doing low-level work. They all have very good high-end consultants somewhere in there, but the vast majority of their business is just an army of consultants doing low-end work and doing business process outsourcing. So they never got that real benefit from this huge opportunity in the market like Palantir did.

Yeah. So the mainframes are now a small niche industry, right? It's for really heavy-duty, consistent, mission-critical transaction processing. That's what mainframe computers are still good for. But again, they have a very small role. So if you walk into, say, a Visa data center, you're going to find rows and rows and rows of servers, and you're going to find a couple of mainframes in the corner that do some very, very specific things that mainframes are particularly efficient at. So it's a niche market, but surprisingly it's still been there for IBM, and it's even had low single-digit growth for some periods.

OpenAI and Anthropic are at about a $100 billion run rate. I don't think a lot of that is from consumers. I think most of it is from companies. If you think about technology spend in general, it's in the trillions. So it's not that it's going to displace all, but it's going to displace some growth. And then let's not forget that part of that token spend ultimately will address the labor market as well, right, which is tens of trillions of dollars. We're not there yet. I think Dario got ahead of himself saying, you know, 50% of people are going to be unemployed. But we are getting closer to a point where we're starting to measure AI's value as a cost per task. Right? That's where the conversation's going. First, we just rent the GPUs by the hour. Then we started paying per token, and now we're going to migrate to paying per task. And that's more analogous to labor, right? We're going to start talking about AI's contribution on an FTE level pretty soon. And which is not to say — we can get into this conversation — which is not

to say it's going to replace all employees and 50% of people are going to go unemployed like Dario says, but it is going to be that same market. The market for labor is going to be comprised of humans and agents and AI models, and that's where this conversation is headed. So for now, we're talking about, you know, a single trillion dollar market for technology, and you have a hundred billion of spend that's now being siphoned off to AI. Pretty soon we're going to be talking about tens of trillions of dollars of labor market where AI gets to participate in.