Palantir - Investment Thesis
The one who knows it all.
We’re going to talk about a complex subject today as we’ll discuss the famous Artificial Intelligence which everyone talks about for anything and nothing lately - it’s hard to know what those words even mean anymore. But I believe that Palantir could be one of the only real AI software on the market.
I went a long way on this company which I literally judged for being a pump & dump with ridiculous promises and unclear documentations. That was before pouring hours to read & listen to everything I could about it and ended up changing my mind completely.
I still believe AI is a hoax in 90% of the cases it is used. Not for Palantir.
[Airtificial Intelligence]
There are lots of definitions for AI let’s use this one going further in this write-up - so we agree on the terms we use.
“Artificial intelligence (AI) refers to computer systems capable of performing complex tasks that historically only a human could do, such as reasoning, making decisions, or solving problems.”
An AI is a system which needs to be fed with data, to analyze them and give a pragmatic answer to a specific problem. Some would say that translation is AI. No, translation as done by Google Translate is simply a database that knows a correspondence between words in different languages and we’ve all been tricked trying to translate sentences - which are often done wrongly.
An AI would be able to use the knowledge from the terabytes of documents it’s been fed to precisely understand the meaning of what we are asking and translate our idea, not just translate the words one by one. I hope I have made the difference clear between a database and an AI - an AI would understand the structure of your sentence and determine the best way to preserve your intent in the translation.
My point here is that a lot of confusion is done lately between AI and simple software based on databases without any model training or real analysis of the data. AI exists in our world today. We can think about Tesla’s FSD, ImageJourney or ChatGPT image creation, Meta’s advertising… And Palantir’s softwares.
[Company]
We have to talk about Alexander Karp who founded the company because he’s quite a phenomenon but before that, we need to talk about the company’s name which comes from a pretty famous artifact from a pretty famous book written by a pretty famous author and adapted to… Well, pretty famous movies.
The reference is pretty simple. The Palantìri are artifacts that allow their users to see anything, anywhere, but also to see the past or have a glimpse of possible futures. There’s no better introduction to the company as that is exactly what they aim to do - in a magic-less world.
The name of the company is what their software would be for your business: a tool capable of seeing everything, past, present, and future. Once installed it will be fed with its data and will know better than anyone everything about the business’s workflows. That is how it will be able to understand why anything happens, optimize what isn’t, anticipate what should happen and issue warnings or propose solutions before or when problems occur.
Nothing here is magic, only software. Let’s dig a bit more so this voodoo makes a bit more sense.
Foundry. It all starts here with what the company calls Foundry. We’re not in AI yet, we’re into something that has existed for a long time now, known as big data. Data collection from data pools, filtering, organizing, and using it.
But data is the start of everything.
Foundry is divided into three parts:
1. Ontology. The heart. Ontology will be the keystone of Palantir’s software, it is where data flows & is kept. Any institution or company has tons of data about everything and nothing and has to use it.
I’ll use a concrete and real example to make it easier to understand.
Palantir partnered with Airbus some years ago in a project called Skywise, supposed to help the manufacturing of aircraft - pretty complex products. You can easily imagine the number of different pieces, humans & interactions you need to build one of those and Airbus decided to digitalize it all, generating tons of data. Each component, its manufacturer, price & stock or information about their staff, capacities, schedule & performances passing by of course live data for the entire Airbus fleet with accurate & updated information about each component and much more…
The ontology is where this data is stored. It can be added manually or automatically thanks to the data generated by processes or sent by components. This is only the tip of the iceberg but it gives you an idea. Anything which happens in the process of building or flying a plane generates data, which is fed to the ontology.
But disorganised data is useless data.
2. Modeling. The structure. This is were data is organized to teach the software the relations between them. Another example.
Screws are used everywhere, even in aircraft, so you need to have some. The ontology knows how many you have, how much they cost, the supplier, the reduction you had by buying a lot, where they are stored, and more… but someone needs to do the links in between all those information and that’s where modeling happens - done manually by engineers, a tough job.
Visuals always help. This is where we stand now. Palantir will collect any data from anywhere and organize it with logical links, creating a structure who knows everything.
3. Operations. The Limbs. When everything is done, the only thing left to do is to play with our new toy: a software which now knows everything about your company.
Where you buy your components, how much each costs, how long to be delivered, what is pending and what is missing, the state of your actual stock, which plane is actually being built and what exact resources in terms of hardware and human hours you need to complete it, who is working on it, who has time off, who is responsible for what, which plane is flying, which components are new and which are failing, which should be replaced and we can start over with it all, as we know where to buy it, how much it costs, how long to receive it, who can change it, etc…
I think you start to see where this is all going. Now, this is great and those tools sure are good and very demanded but they’re also sold by Microsoft, Google & Amazon as long as the data comes from their own infrastructures - which is already a great advantage for Palantir as they’ll accept data from them all.
But they went further. Where no one else (to my knowledge) went.
Artificial Intelligence Platform - AIP. The brain. Many would call Foundry AI already but in my opinion, it isn’t: it “simply” is a big data software, knowledgeable but not intelligent, giving you an information that is available from its database, nothing more.
At least, not before adding AIP as this new software’s goal is to add LLMs to access the data stored & organized in Foundry. It will do more than “just” add LLMs over it; it will give companies the capacity to customize those LLMs to access the data present in Foundry.
AIP is your employee who knows everything about everything on your business, is never tired, and will give you the most pertinent answers solely based on a practical vision of things.
Foundry is knowledgeable, AIP is intelligent.
This video gives a very good overview of how AIP works. It isn’t just a huge collection of data anymore, it is an intelligent tool which knows your business better than yourself and can anticipate, resolve, and give logical & practical insights on every situation, may they be real or hypothetical.
Other. Just before moving on, I will talk about Gotham later on and provide a few quick words about Apollo now but won’t go too deeply into it as it is interesting but not much relevant to my investment thesis.
So, briefly, Apollo is a packaging solution that helps deploy software on any environment simply & quickly and is used internally to deploy Foundry and AIP. For those familiar with it and to make some very simple comparison, it is kinda like Docker.
[Market]
Every, or most companies are selling their products into a very specific market. But not every product fits into categories as when it comes to software, markets become pretty wide. But this is amplified in the case of Palantir in my opinion.
What is Palantir selling really? A software helping business optimization, planning & anticipation for each of their workflows. And what does it need to function? Data. Which company doesn’t want to optimize its workflows? And which company doesn’t have data?
None - or very few.
It’s very, very hard to define a TAM for this company as everyone could benefit from its product, in every sector - retailers, manufacturers, freight, tech, healthcare... And the best proof of that is feedback from Palantir’s clients which, as said, come from very different industries with very different businesses.
https://x.com/arny_trezzi/status/1709929346876821544
[Commercial vs governmental]
We will come back to the company rapidely and talk about its origin. Founded by Alexander Karp and four others - notably Peter Thiel who’s everywhere - and was originally built as an intelligence & counter-terrorism company following the 9/11 attacks, selling its services to U.S. intelligence departments. I won’t go into what the company did at the time because it isn’t very relevant to my point, but it is important to understand the roots of the company and the tight relationship it developed with the U.S. intelligence branches through the years.
It grew around softwares as a way to prevent events like 9/11 from happening again, to develop ressources which would see it coming and anticipate proper reactions to it, to be more resilient & efficient.
Gotham. The weapon. Hence, Gotham, a software alike Foundry & AIP but designed for military decision making and sold to governmental agencies, used at the moment in Ukraine & Israel to help evaluate situations and propose adequate responses to what is happening on the field.
Some disclaimer here: it is impossible to talk about Palantir without talking geopolitics because it is a big dimension of their company & business. Let’s just keep in mind that personal opinions do not matter here. I am simply talking about an interesting company, nothing else.
This post will be full of videos but they rally help to understand Palantir better.
This video is three years old and yet shows some pretty advanced reasoning from their tools already - which have been getting better & better the last years.
An important part of Palantir is working closely with western governments & institutions to provide this kind of service. Around 55% of their revenues came from governmental agencies FY-23.
But things are turning around and even if Palantir is very focused and implicated in its relationship with governments, more & more commercial companies are interested in their products and this is what interests me the most about Palantir because those companies are what gives Palantir its real potential.
The question for years was: How to reach them? How to present them with the potential of AIP? And how to integrate it into their infrastructures?
The answer came in 2023.
[Bootcamps]
With what Palantir calls bootcamps, and actually are... bootcamps. The goal is to reunite around a table technical teams from different companies and showcase a real Foundry & AIP example.
They’ll even go further and use corporate data from those companies to build an entire & functionnal foundry in less than 5 days.
You can find more details on here.
https://blog.palantir.com/deploying-full-spectrum-ai-in-days-how-aip-bootcamps-work-21829ec8d560
A lot of people called Palantir liars for this kind of publicity and I can understand the skepticism around it as this kind of softwares should take months to implement - again, more about this later - but the results speak for themselves… This comes from the company’s last earnings:
"two years ago, we did 92 pilots. And last year, really mostly the second half of the year, we did over 500 bootcamps."
Palantir ended 2023 with more than 560 bootcamps done for more than 465 companies. But demand for bootcamps is one thing; the real question is, are those companies becoming clients? Does Palantir convert those bootcamps into money?
Let’s use the most recent data we have from Palantir’s Q1-24 where Alex Karp talked in his letter about 660 bootcamps done for 450 companies. Those resulted in 136 deals with 87 of them being above $1M.
136 deals on 660 bootcamps means a conversion rate above 20%, the 87 deals above $1M mean a conversion rate above 13%. And this isn’t the only data pointing in a positive direction.
The number of commercial clients also logically rose and is growing very rapidly, around 20% QoQ.
Skeptics can be skeptics (and I used to be) but data never lies - commercial companies like & demand the product. And if this wasn’t enough, it was confirmed once more by management.
“We have a growing backlog of AIP Bootcamps due to the overwhelming demand.”
[Competitive Advantages]
If you’re still reading, you’re probably able to answer the question: What sets Palantir apart already, but I’ll still summarize it here.
A product everyone needs. I already used this argument with Olo saying that food franchises would become irrelevant or less competitive without Olo’s software. This is true again for Palantir although we’re not only talking about restaurants here, we’re talking about every company in the world.
Data is the most valuable resource nowadays and every company tries to collect, analyze and upgrade its business thanks to it. They need to know what they do, how they do it and how it is received to do so but building this kind of infrastructures internally is very complexe and costly.
“Palantir is the future Operating System for the companies of the future.”
There is some truth in this.
No defined market. As said earlier, as long as the business has data and can rely on them to optimize itself, Palantir is a good fit.
More than big data. This might not have been the case if the company stopped at Foundry but they didn’t and chose to innovate and propose AIP & LLMs on top on the collected data, making their software intelligent & easier to use, with deeper possibilities thanks to its capacity to reflect & analyze.
Every bit of data. I said earlier that Foundry wasn’t that original and that other companies proposed comparable solutions - Microsoft, Google, and Amazon, for example. But there is one important thing that sets Palantir apart.
They accept data from everywhere, while those competitors tend to only accept data from their own data solutions. Google’s big data software will accept data coming from Google cloud, and the same goes for Microsoft. They try to force companies to rely entirely on their products.
Palantir doesn’t do that. They built the ontology to accept every bit of data from anyone.
Selling software and not data. Another critical point often misunderstood about Palantir, the only thing the company sells is a software; what their clients do with it afterwards isn’t any of their concern.
It matters because those clients will feed Foundry with the most critical data of their business and anyone who can access it would have an explicit overview of their entire business. This is like getting the Coca-Cola recipe - although the secret ingredient is the brand, nothing else.
Not owning the data is critical. And Palantir even view it. Once the software is sold, it is the client’s responsibility to protect it - which raises another bullish case for any IT security companies but that is another topic.
Governmental relations. This has to be mentioned here as Palantir has been working with Western governments for years and has developed close relationships with them. It’s hard to detail how this will serve them but it’s more an asset than a liability.
[Risks]
But everything isn’t perfect and risks exist - some because of themselves, others due to external factors.
Political stance. I have talked about this a lot and it is understandable that the company takes public stances considering that their business is close to the U.S government, but it can become a liability in the commercial world.
Your or my personal take on things does not matter here, we’re talking about a company publicly taking sides and we have enough examples with Starbucks & Disney that companies shouldn’t do it - although Palantir has no relation to retail.
Besides their image, there also is a categorical refusal to work with “the enemy.” And those are Karp’s words, not mine. Palantir’s management isn’t entirely revenue-focused and they’d gladly pass up opportunities if they were to include “enemies”.
Controversity. The above point is about how Palantir’s management sees things. This one is about how Palantir is seen by the world. Not as good as they’d wish, even in the West, and the contract they won with the National Health Service in England showed this clearly.
Some context first: There was an offer to renew the system for the entire UK health system and Palantir won it - which is an amazing achievement with positive first results as everything is working better already.
"The progress was down to NHS staff delivering more than 1.63 million treatments in November, the highest monthly activity on record and around 150,000 more than the same month before the pandemic."
But it didn’t prevent lots of manifestations against Palantir for different reasons. Some because of what Palantir stands for - Israel, in this specific case.
Their image didn’t hurt them in this specific case because they won the contract, but they might lose some because of it.
Complexity. Others because of missunderstandings on what Palantir really does - although the subject isn’t easy. The NHS contract is about treating very sensitive data about… Almost every English. And many didn’t want to have an American company to handle those data - which they of course won’t but that is what a “misunderstanding” is.
Complexity shouldn’t be a risk, but when you try to win public service contracts, the population ends up screaming a lot - even though it often is to say nothing.
AI imposter, Consulting Company & Monetization. We might not consider them as risks but there are criticisms on various fronts, accusations of fraud and other concerns about how Palantir will monetize its products over the next few years.
This last subject should be addressed but we do not have a lot of information about it. Karp has said many times that Palantir's objective is to build the greatest product that everyone wants, and that monetization was secondary.
[Controversial Alexander Karp]
I will have a small but focused part on Karp because he kinda deserves it as he’s a personage, closer to a controversial Elon Musk than a rock star Jensen Huang though.
A strong personality often makes a strong leader and as a founder, Karp considers Palantir as his children and has been and still is giving it all his time, energy & dedication, and no one can take this away from him. He is a strong founder and a very good & efficient CEO, pushing his teams to excellence every day.
But his temper & strong convictions are an asset as much as a flaw as it often makes him insensitive and say things that shouldn’t be said by a CEO publicly - although some would disagree on this take. We are not even talking about (geo)political stances here but about straightforward aggressive takes on lots of different subjects or people.
On another hand, and to continue with his temperament, he’s also more focused on the world and the difference Palantir can make to it than shareholders - many would agree with me on this I think. Once more, personal opinion doesn’t matter; some think this is a good thing, others think this is a bad thing. But when you’re a publicly traded company, your focus should be to your shareholders, no one else.
And I’m not so sure that’s what is happening on Palantir, not all day, every days at least.
[Finances]
The company’s finances give us a pretty straightforward story.
The story of a company that has succeeded in becoming profitable while maintaining a proper growth, as they are still planning to grow more than 20% this year - after growing at a 30% CAGR over the last five years. Financing itself through share programs even now that the company is profitable - which is good and bad. I am all in to reward workers with this kind of incentives but it shouldn’t dilute other shareholders too much over a long period. It isn’t concerning yet, let’s see how this moves during the next quarters.
There won’t be much more to say here except to talk about this strong pile of cash with a $3.6B net debt with most of it slowly generating other income through bonds - who says no to more?
[Opportunity]
As usual, this section isn’t meant to give a future price target but to have an idea about the company’s potential.
As I have no definite markets and no concrete idea on how governmental contracts will develop, this will be very incomplete & hard to model. So I’ll simply talk about the commercial part and the impact of bootcamps on the company’s growth.
As said earlier, Palantir did roughly 560 bootcamps in 2023, most of them during the second quarter while it did 660 in Q1-24 only - useless to express how impressive that is and how the company has focused on it lately. It’s also very hard to know if demand will hold this rhythm but let’s make some assumptions.
1. Palantir can keep up around 500 bootcamps per quarter and convert 10% of them.
2. Those new clients will be divided into two categories, 70% of them as $1M clients & the rest as $5M contracts - conservative compared to what is happening.
3. Net margins will increase with time as Palantir continues to work on profitability.
Now, this is only the commercial part of Palantir, we’d need to add governmental revenues to those results. It might seem irrelevant but those numbers would have Palantir trading at 35x sales considering only half of its business while the software industry trading at 11x sales.
If we were to add the government branch and estimate a 20% growth rate - many analysts expect it to grow more rapidly than this, so another conservative number, we could have something like this.
A company tradding at x17 FY-24 sales - guidance is a bit lower than this projection for FY-24 but not that far, with a low estimate at $2.67B revenues.
If you have time and are interested, I will put here Emir’s valuation model which is much more complete & interesting than what I’ve done here. The big difference between him & I being that I find modeling those growth companies hardly relatable as things can change so fast… I like to rapidly see the potential based on rapid & plausible estimations, Emir goes into things much deeper.
We do not necessarily reach the same conclusion as I’d consider Palantir fairly priced around the $22 - $20 mark but it’s always interesting to have different opinions, especially on growth stocks.
[Conclusion]
You now have a more comprehensive view on what Palantir is & does. A wonderful company and a wonderful product to my opinion which plays perfectly in what could be the most important narrative over the next years: Optimizing existing business through a better knowledge of themselves thanks to their own data.
“The Operating System for the companies of the future.”
Thank you for reading it all! If you like it, please consider subscribing to receive it all directly in your inbox and not miss a thing!
Everything I share here is free but if you found the content valuable enough, you can always leave a tip!