I've used Palantir as my go-to example of an overvalued company for years. Michael Burry recently said it isn't worth a dollar, and The Economist called it possibly the most overvalued firm of all time.

However, I must admit that things have changed. A year ago, Palantir traded at twice the multiple with half the growth. So I felt it was time to figure out exactly what was going on and what Palantir looks like in 2026.

It’s safe to say I was very impressed by what I saw, and it might be the first time ever I've defended a stock trading at 60+ times sales. Obviously, it’s not a value play, but everyone who loves high-quality businesses and is generally curious should keep reading, because I think you'll be positively surprised.

Let’s dive in!

— Daniel

The only reason I can comfortably cover dozens of companies a year is that I get to talk through pitches with members of our Intrinsic Value Mastermind. Palantir was well outside my circle of competence, but I spoke with a handful of software engineers and founders in the community, and by the end of it, I understood the company and the industry dynamics far better than I could have on my own in half the time.

If you want to move more quickly and thoroughly through companies, I'd genuinely recommend applying. Besides the 1-on-1 calls with members, we host weekly calls with Kyle, Shawn, and me, along with outside fund managers, famous authors, and acclaimed individual investors.

And I should note that you don't need to be an expert investor to join. The best conversations always come from people with different backgrounds and industries – this pitch is a good example of why.

One more thing: our Intrinsic Value Conference in New York City on Saturday the 19th is free to attend for Community members. So this is a particularly good moment to join.

We’ll get to more detailed explanations later, but I think I need to start by briefly explaining what Palantir does before I give more background on how it was founded.

Think of any large organization – that could be an airport, a global car manufacturer, a large hospital, or even the national army. All of these organizations have an enormous amount of critical information that they need to, well, organize. But the bigger the organization, the more information is scattered across dozens of separate systems and departments that barely talk to each other. And when they do, it’s rarely efficient.

In the world of computer science, the term Ontology can be roughly defined as: A structured framework that formally defines concepts, properties, and relationships within a specific domain (it also has ancient philosophical roots as a term).

To illustrate the concept of ontology, which Palantir identifies closely with, let’s stick with an airport example for now. Assume an aircraft runs into mechanical problems in Frankfurt – I use Frankfurt as an example because of my PTSD from being stranded there for 50+ hours after multiple United flights to Omaha apparently had such mechanical problems… Anyway, airports are one of the busiest places you can imagine.

Dozens of teams have to coordinate which airplane is at which spot at any time of day, and whenever something doesn’t work, there’s a chain reaction of adjustments.

The maintenance system knows which part failed and how long the repair takes. A different system knows the crew and how many duty hours they have left. A third knows which passengers are on board. A fourth knows which of them have connecting flights. A fifth knows what the replacement seats are worth. And so on.

So you end up with many different systems and teams that need to coordinate as quickly and effectively as possible. The way this gets resolved today, in 2026, is people phoning each other. Meanwhile, the plane sits on the ground, costing money every minute.

What Palantir sells is a software layer that sits atop all five and builds a single shared model of the airport. You can think of this “ontology” as a map that covers everything that goes on at the airport. You can see the aircraft, have updated information on all of them, and see everything else you need – live information on the parts, the crew, possible connecting flights, etc. It’s like a control system.

You might think it just sounds like a database, but the major difference lies in the relationships and actions embedded in Palantir’s ontology. A database knows “facts” – the plane's tail number, what part failed, or a passenger's seat, for example.

What Palantir builds knows not only the facts or assets (an aircraft in this example) but also the relationships between them, what actions are possible, and how best to communicate adjustments. It knows passenger A is on aircraft B, which needs part C, which gets fitted by crew number D. And it knows that rebooking is a viable option if other planes and crew are available to take over. It also knows who's permitted to approve such a change, and what else has to happen.

That combination is what Palantir calls the Ontology. When I described it to Shawn, he said it reminded him of those action movies where the field agents are out fighting, and there's a control room that always knows exactly what's going on. And that’s pretty spot on, especially on the defense side. Correspondingly, you can see why many think Palantir, and its co-founder, Peter Thiel, give off super-villain vibes.

I hope the above gave you a slightly better understanding of what Palantir actually does. Now, getting to the founding story. Palantir was founded in 2003 by Peter Thiel, Alex Karp, Joe Lonsdale, Stephen Cohen, and Nathan Gettings. The idea actually traces back to PayPal.

Thiel had a persistent fraud problem at PayPal, and the fix he came up with was a mix of software to surface suspicious patterns across vast amounts of data, plus a human making the final call. Then 9/11 happened, and it became clear US agencies collectively had all the information needed to stop it but couldn't assemble it in time. That's when Thiel’s idea for Palantir was born – a system that connects fragmented, seemingly unrelated data points to map out hidden criminal behavior. Just like PayPal but with different data and at a much larger scale.

Thiel bankrolled an initial cost of $30 million. In-Q-Tel, the CIA's venture arm (yes, the CIA has a venture arm…), put in $2 million. Which seems trivial next to a company now worth somewhere between $350 and $400 billion depending on the day. But the money wasn't the point. In-Q-Tel provided important access and introduced Palantir to intelligence agencies. And for a good two years, Palantir’s only customers were those agencies.

In part, that’s also because it took Palantir a while to figure out how best to onboard customers. Historically, the onboarding process was long and complex. Palantir has two kinds of engineers. Product engineers build the software in-house, while forward-deployed engineers, so-called FDE’s, fly out to customers and work from their offices a couple of days a week to implement and adjust the software to their needs.

Most of their job is figuring out exactly how the corporation works to develop the ontology (essentially the map). The on-the-ground work is what consumed most of the time in the onboarding process. I went through a lot of accounts from ex-Palantir employees, and the recurring theme is that internal politics is one of the biggest obstacles.

Before Palantir arrives, everyone is doing their own thing, and between inertia and the fear that you won't be needed once the software exists, middle management frequently refuses to help or hand over important data.

Palantir’s onboarding struggles and overall complexity actually caused growth to decelerate steadily from 2019 through June 2023, bottoming at 12% year-over-year. For a supposed high-growth software company, that's close to an obituary. At that point, Palantir was even seen as a fake SaaS company – enterprise consultants with a software logo.

Then AI came and changed the game for Palantir.

But before we get into how AI transformed Palantir, let me tell you about Palantir’s four platforms. Simplified, you can think of these as Palantir’s products.

Given Palantir’s roots in defense, Gotham was the first platform Palantir created. Roughly five years of continuous building, starting in 2003, enabled intelligence agencies to create a digital twin of defense-relevant fields. 

I just recently came across a podcast discussing police work in Germany. A police officer said he worked on a project to figure out the structure and relationships of a crime syndicate operating in Germany. It took him months to create a mind map of the group. According to him, if he had been allowed to use Palantir, this could’ve been done in minutes.

Almost all the data already existed, and someone, somewhere, always knew the people and their relationships, but it took him months to gather and centralize it. This is exactly what Gotham can do in an instant. And it would update the mind map live as soon as new information comes in at any point in the chain. 

A bit more than a decade later, Palantir built its second Platform – Foundry. Foundry does the same thing for corporate operations. What we described at the beginning of this newsletter – the airport example – was basically a Foundry example. 

Both Gotham and Foundry sit on Apollo, Palantir’s deployment engine. You can think of it as if Gotham and Foundry were apps, and Apollo were the operating system that keeps them running. Apollo is built to update Gotham and Foundry across all sorts of environments. Of course, they can’t always sit in a standard cloud environment, which would make things too easy. They need to run anywhere from secure government servers to laptops in military vehicles on the front lines.

Perhaps the platforms give you a sense of how complex Palantir’s business actually is. Apparently, it was even too complex for many customers. Not only was it difficult to convince potential customers to onboard Palantir, but even when they did, they lacked the understanding needed to use it effectively. That slowed down growth tremendously in the early 2020s. But then Palantir introduced AIP – the AI Platform.

And while they won’t win a prize for the most creative name, AIP changed both how Palantir approaches customers and how efficiently the software runs. Now you can talk to Gotham and Foundry just like you talk to ChatGPT or Claude. You don’t need to know exactly what to do and where to find it; you can just have an AI navigate through it.

This made things so much simpler that Palantir became increasingly important to its existing customers. A significant chunk of Palantir’s growth came from existing customers in the last few years. Net dollar retention – how much last year's customers spend this year – went from about 100% in 2023 to almost 160% last quarter. So for every dollar a customer spent last year, they now spend $1.60.

Palantir also changed its strategy for onboarding new customers. They realized that explaining what Palantir does is way too complex (though I'm doing my best here!).

So they decided to start offering customer-facing bootcamps. Palantir invites the CEOs and CIOs of potential clients; they bring some of their own data, and over three to five days, Palantir shows them specifically how it would improve their business. And Palantir pays for them, so attendees take on no risk.

Those changes caused Palantir to significantly reaccelerate growth after the 2023 lows.

What’s striking is that none of this growth cost any margin. In 2023, with growth slowing, net profit margin was 3-4%. Today it's around 60%. That’s undoubtedly one of the biggest margin swings we have seen in any company we covered. 

A great way to illustrate this remarkable financial profile is to look at the so-called Rule of 40. The Rule of 40 is a software heuristic that combines growth rate and profit margin. 20% growth at a 20% margin gives you 40%, and that's considered a very healthy business.

If you add up Palantir’s 90% growth and 65% margin, though, you get to 155%, which was unheard of before.

That's possible because Palantir only pursues large customers with lots of upsell potential, which is also why the customer count looks relatively small for a SaaS company of this size (only about 1,000 customers). The average deal size runs into the millions. Last quarter alone, they closed 220 deals worth at least $1 million, close to 100 worth at least $5 million, and more than 70 worth at least $10 million. So while adding only 42 net new customers, they closed 220 million-dollar deals.

The most difficult question to answer about Palantir is when competition will arise. Whenever a company grows at these rates and margins, it can only be a matter of time until competition shows up. And based on my knowledge, the ontology itself can be built by other companies as well – think of Microsoft with its Fabric IQ, Google, or other tech giants.

However, Palantir has a major data advantage. LLMs tend to be about equally strong. Perhaps one is slightly stronger today, but a month later it could be out of the Top 5 again. What really matters is that you have a unique and specialized set of data to train the model. And that’s where Palantir is unbeatable. The data Palantir was able to work with over the past two decades makes catching up a very costly and time-intensive bet. 

And from the customer’s perspective, there’s little reason to switch either. Once you've been through that multi-month or year-long onboarding process, you'll think hard before doing it again somewhere else. 

There's also the “special” culture, which you can sense within two minutes of watching any Karp interview. Very little hierarchy, with people largely working on whatever they think creates the most value. That's rare in a company worth hundreds of billions, and such culture isn’t something that you can suddenly change or introduce.

And then there’s product philosophy, which, again, runs against the convention of most competitors. Enterprises historically bought best-in-class point solutions – Slack over Teams, Excel over Sheets, and Zoom over Meet. Companies build one or two excellent tools and sell them broadly. That’s lower value per transaction, but higher volume. Palantir basically builds the entire operating system for your company.

The obvious objection is that Microsoft and Google both bundle so many tools that they could level up the bundle, wire AI through it, and offer something similar. But going from "good enough" to best-in-class is much harder than it sounds, and we've argued exactly that in both our Google and Microsoft episodes. 

As many of you know, I usually avoid betting on companies competing with Google, Microsoft, or Amazon, but here I think they're too bloated and too inflexible to win on the ontology front. Also, I doubt they’ll be able to shift their focus to defeating Palantir while they literally spend hundreds of billions and deploy their brightest minds to solve other problems.

But what about OpenAI and Anthropic? Couldn’t they eventually build this themselves to justify their own valuations? I think Sam Altman essentially gave the reason why OpenAI won’t in one of his latest interviews with David Senra.

Besides claiming that GPT4 was already powerful enough to disrupt most of SaaS, and that inertia and the lack of good products are the reason it hadn’t, he also, somewhat contradictorily, said that OpenAI will remain focused on research and compute instead of building products.

Anthropic is more focused on building products and features for enterprises, and yet, I don’t see any reason for a Palantir client to ever switch or potential new clients to choose Anthropic over Palantir. 

Even assuming Anthropic would be able to build a similar ontology for clients, you’d essentially get the same product while being locked to a single model instead of staying agnostic with Palantir (Palantir offers ChatGPT, Claude, Mistral, etc.).

Karp also regularly attacks both on privacy concerns, warning companies not to hand their IP and business model to frontier labs that will eventually take their business. And while he has every incentive to stoke that fear, that doesn't make him wrong.

And last but not least, what are the odds that companies will start to build solutions in-house? Given how much AI has impacted Palantir’s products, it’s not totally unthinkable, in theory, that companies could build their own ontologies eventually. Naturally, they have all the data, and with AI available, it should be theoretically possible.

However, I highly doubt companies would choose to go this path over just hiring Palantir. First, why should a company stop focusing on its core business and redirect its smartest engineers and a ton of resources to build something that can simply be bought? It’s like saying we don’t use Zoom or Google Meet because we can just build our own tool. It doesn’t make much sense.

And second, remember one of Palantir’s worst struggles? Internal politics. Those dynamics still exist, even if it’s not an external company coming in. All in all, I don’t see internal builds becoming a serious problem at Palantir’s target customer for a while.

Palantir not only has outstanding growth and margins, but also an outstanding balance sheet. Palantir holds more than $9 billion in cash with no debt. Yet, there’s no plan to buy back stock, pay a dividend, or spend it on M&A. Personally, I think that’s the right decision. I would rather build a war chest now than see stock being bought back at prices that are at least on the more optimistic side.

Opportunistic buybacks at a later point or reinvestment in the company seem like much better options than buybacks or dividends just for their own sake. I’m generally not a fan of rigid buyback programs when the stock is not clearly undervalued.

That said, the cost of not buying back shares is dilution. Stock-based compensation still runs at 13% of revenue, so shareholders are being diluted at some real speed. Since revenue is growing rapidly, that number is going down, and yet it’s a meaningful headwind for anyone buying today. 

What I kept asking myself is why Palantir doesn't spend more of that cash on new talent, given demand exceeds what they can deliver.

Palantir’s answer to this has been a partnership with Accenture, in which Palantir effectively borrows Accenture's consultants as forward-deployed engineers. So Accenture trains its own people on Foundry and AIP, and those handle the deployment work on-site.

Palantir sells the license and keeps its own engineers on the platform. That’s higher margin and much faster than hiring people on your own, but I assume it also dilutes the quality of the work. Ultimately, Palantir is much more selective in its hiring. For perspective, Palantir has about 4,500 employees. Accenture has close to 800,000, and already more than 1,000 Foundry specialists.

We talk a bit more about Karp and Thiel in the episode Shawn and I recorded on Palantir, but I wanna briefly mention it here, too. Thiel has been a highly controversial figure for a while now, but Karp was little known for the longest time. After all, he has already run the company for more than 23 years.

In the last few years, though, Palantir has received more public attention, and it didn’t take long for that attention to turn a spotlight on Alex Karp. And boy did he use that attention. One of the first interviews I saw from him was on stage with Andrew Ross Sorkin. Alex Karp talked about how every human needs a higher and a lower purpose. Asked about his lower purpose, he said, quote: “I love the idea of getting a drone and having light fentanyl-laced urine spraying on analysts that tried to screw us.”

Again, we dived deeper into his thinking and philosophy in the podcast, but saying this publicly on stage might give you an idea of his thought process. Now, combined with his goal of preserving U.S. and Western dominance, you can also imagine how foreign governments stand on the idea of giving Palantir their most sensitive data.

Earlier, I mentioned the German police officer who said Palantir would have saved him months of work. The obvious solution would then be to give him access. The reason he hasn’t is comments from Karp like the above. Germany and France have already announced plans to develop a Palantir alternative, and European countries and major cities have canceled contracts.

For the record, I don’t necessarily think this will turn into a problem, given that I have little confidence European governments will come up with something that's even remotely as good as Palantir. And yet, it is a headwind.

While the size and growth of Palantir’s government and commercial businesses are similar, historically, U.S. business growth has far outpaced international growth, which is mostly in Europe. Because Europe is growing more slowly, here’s Karp’s take on things, in short: “Europe’s growth sucks!” (this is a real quote).

But again, I assume this is because there’s hesitation and resistance to working with Palantir in Europe.

Alright, time to talk valuation! I have to admit, when I started looking at Palantir and told Shawn this company is far less expensive than I thought, it was trading at $120 right before earnings. Now it trades at $185 – up a bit more than 50%. That’s a bummer.

And yet, there’s an argument to make that, should Karp’s unofficial guidance turn out true, there’s still room to go for Palantir. Karp claimed he can grow the whole company at the current rate of the US commercial business through the end of 2027. That would mean 150%+. If that’s the case, the 60x sales multiple I looked at would drop to 20x. The multiple is slightly higher today, so let’s say it drops to the mid-20s. And whatever you think about Karp, he has delivered on his promises so far.

And while price-to-sales (P/S) might not be our favorite metric to look at in evaluating a business, it makes the point here quite well since Palantir’s margins are best-in-class as well. When the average enterprise software company trades around 7x sales on roughly a 20% operating margin, that's 35x operating profit. Palantir on its 2027 numbers would be about 25x sales – but at a 60% margin, that's also ~35x operating profit.

If you put Karp’s estimates into a model and decay growth to 60% after 2027, then subtract 15-20 percentage points per year, ending at 28% in 2031, the revenue CAGR lands in the high 50s. With a stable 60% margin and a 30x exit earnings multiple, you get to a fair value of about $240 per share.

The picture changes, though, when we model Palantir using Analysts’ estimates. The 5-year CAGR then decreases to 30%, basically cut in half. In that case, the fair value of the stock is about $110 before applying a margin-of-safety discount.

I believe the fair value is higher than that, given Palantir's incredible growth and the fact that they are achieving it without any margin compression. Should we see a price in the $100-$120 range again, without a material change in the story, I would be inclined to establish a position.

That said, I need a large margin of safety on this to compensate for my lack of technical understanding. In our episode, I talked about a framework I use when buying a company whose returns depend entirely on the growth rate over the next few years.

I like to ask myself: If the company’s next earnings report lands and growth halves, could I realistically understand what happened? Would I actually know where things went wrong, beyond whatever management summons as an excuse? It sounds simple, but it works quite well for me. For example, if Lululemon reports slower growth, it may be because it sold fewer clothing items or sold them at a lower price. And the reason for that is either competition or macro. To figure that out, I can look at competition and macro data.

For a payments company, it's a bit harder to get the data, but it’s still competition or macro, and I can figure it out. Then there’s a company like The Trade Desk, where we genuinely can’t figure out why revenue keeps declining or where the bottom is supposed to be.

And being honest with myself, Palantir sits in that last category. While the business is growing, everyone has the illusion of understanding why. If that changed suddenly, I'm not sure how many of us could still claim to understand the business well. My best guess is that the narrative around the moat changes and that the ontology wasn’t as hard to copy after all.

Again, there are no signs of this yet, and there are good arguments that Palantir is very sticky, but I need to be honest with myself about my level of understanding. That served us very well with The Trade Desk.

To listen to our episode on Palantir, check out our podcast here.

Updates on our Intrinsic Value Portfolio below 👇

To discuss stocks daily with Shawn, Kyle, Daniel, and all Intrinsic Value Mastermind Members, apply here.

"The age of social media platforms and food delivery apps had arrived. Medical breakthroughs, education reform, and military advances would have to wait.”

— Alex Karp

📺 WATCH: Sam Altman’s Interview on David Senra’s Show

🎧 LISTEN: CATL Deep Dive by Stig Brodersen, Manish Karira, and Ralph Summerford

📖 READ: The Guardian: What Meta’s Settlement means for future Lawsuits

You can also read our archive of past Intrinsic Value breakdowns, in case you’ve missed any, here — we’ve covered companies ranging from Alphabet to FICO, Transdigm, Lululemon, PayPal, DoorDash, Crocs, LVMH, Uber, and more!

Enjoy reading this newsletter? Forward it to a friend.

Was this newsletter forwarded to you? Sign up here.

Apply to join our Intrinsic Value Mastermind Community of investors.

Learn how to join us in New York City for the 2026 Intrinsic Value Conference!

Read our full archive of Intrinsic Value Breakdowns here.

Keep an eye on your inbox for our newsletters on Sundays. If you have any feedback for us, simply respond to this email or message [email protected].