Palantir: 100X Potential. From Here.
Update
Edited by Brian Birnbaum and an update of my original Palantir deep dive.
Palantir is in the same fundamental situation Nvidia was in back in 2016.
Last week I posted on X that Palantir stock could 100X from current levels. The post reached over 600,000 impressions and confounded X’s financial community. With my Palantir investment up tenfold, consensus pegs the company as tremendously overvalued. It wasn’t terribly surprising to see most express disbelief. However, most investors think quarter to quarter. I think across decades.
The first question is, what is Palantir? Palantir sells digital twins to its customers, which are exhaustive real-time digital copies of a company’s operations. Digital twins turn a company’s real-time functions into tangible, manipulable data that can be plugged into AI models for rapid and easy automation. Digital twins will eventually be foundational to all operations–as required and assumed as laptops, internet, and electricity–turning massive quantities of data into actionable insights if not outright automated functions.
With the competition years behind, Palantir is positioned to take the whole market. Many ask why Microsoft or other companies can’t compete. Creating and operating a serviceable, let alone quality digital twin requires the one thing you can’t achieve faster than your competition: Time. Accelerated creation of a digital twin might be able to pull off some cool things–until someone sees data they’re not supposed to, or an LLM’s ever-so-subtle misinterpretation costs your company $300M in the blink of an eye. Creating a high-quality digital twin requires, without question, decades of exclusive focus, especially when working with companies in highly sensitive industries such as defense, finance, and healthcare.
Microsoft, Google, and the other megacap names must either tend to their core businesses or risk pole position. The best talent flows to these segments. When they do step outside their core competency, they often fail against smaller, more specialized firms whose top talent is hurled at said specialty. A great example is Spotify, which defeated both Amazon and Apple. While music is an afterthought for Amazon and Apple employees, it’s what Spotify lives and breathes.
Palantir has world-class talent density. Their culture is optimized for the task ahead. For Microsoft or Google to take on Palantir will require a drastic shift of focus towards digital twins. They would have to compromise if not outright abandon certain business goals, some of them critical, and risk their financial and operational standing. Google would need to poach many of their own engineers from search for a digital twins department that’s 20 years behind Palantir’s. Microsoft would need to strip the cream of their engineering crop from their suite of productivity tools. Any company looking to turn Palantir’s margin into their opportunity would first have to announce to Wall Street a CapEx program far more daring than the metaverse program that sent Meta’s stock down upwards of 70% in 2022. Beyond questions of feasibility lie those of whether any frontrunner in tech would be willing to risk their golden geese for such an adventure–one that would likely fail given Palantir’s lead, which, again, can only be overtaken with time and superior operations.
This dilemma has played out countless times across human history and is a powerful predictor of competitive outcomes in business (which underpinned my Spotify thesis). The most fascinating instance, however, came long before the streaming wars
The Spanish conquest of the Inca empire occurred in 1532, when Francisco Pizarro led a small force of 168 Spaniards into the heart of the vast Inca Empire. This ancient people wielded a population of about 10 to 12 million people and an army of nearly 80,000 warriors. Despite being vastly outnumbered, Pizarro took advantage of internal strife: a civil war between Inca Emperor Atahualpa and his brother Huáscar. The Spaniards also possessed advanced weaponry such as guns, cannons, and horses, which the Incas had never encountered. Confident in his power, Atahualpa agreed to meet Pizarro at the town of Cajamarca, believing the Europeans posed little threat.
Atahualpa arrived with thousands of unarmed attendants for what he thought was a peaceful meeting. During the encounter, a Spanish friar demanded that Atahualpa accept Christianity and submit to King Charles V of Spain. When Atahualpa rejected this and reportedly threw the Bible to the ground, Pizarro launched a surprise attack. The Spaniards slaughtered thousands of Incas in the ensuing chaos and captured Atahualpa. Pizarro held him hostage, demanding a massive ransom of gold and silver, which the Incas delivered. Despite the payment, Atahualpa was executed in 1533, sealing the fate of the Inca Empire.
Pizarro won because his smaller team was focused on war. The Incas assumed the vastly smaller Spanish force wouldn’t even consider such a bold move. Block co-founder Jim McKelvey termed this dynamic the The Innovation Stack: a small hyperfocused organization executes thousands of specialized actions and details that, over time, build a moat.
The Spaniards began plotting their war long before Cajamarca. Similarly, Palantir has built a moat from their 20-year obsession with digital twins that no one heeded until it was far too late. Now that digital twins are the only way to deploy AI at scale, everyone has to catch up.
Interestingly, the relevant mental model behind my long term view of Palantir’s 100X potential is the Nvidia Algorithm, which I teach to students of my Tech Stock Goldmine course. Nvidia’s FCF/share is growing exponentially, as you can see in the graph below. Stock prices track free cash flow per share over the long term, and Nvidia stock is up 90X from August 2016 alone.
Nvidia’s Algorithm achieved exponential FCF growth in such a manner that, in my view, reflects Palantir’s fundamental situation as well. By algorithm I refer to an operational pattern that serves to maximize shareholder returns over time.
Nvidia’s success is the result of cultivating GPU mastery for decades–well prior to the moment when LLMs brought AI into the mainstream consciousness. Far beyond designing them, the company owes its exponential growth to making their GPUs easy to use via CUDA. After its launch in 2007, CUDA exploded their GPU’s number of applications. Previously used only for gaming, CUDA enabled all kinds of computational tasks, exponentially increasing Nvidia’s TAM at a marginal cost that led to higher margins over time.
Now that AI is kicking off, this operational algorithm has created the perfect storm. To beat Nvidia at their GPU game, you must undergo the same multi-decade process. With every additional GPU they put out into the world, the moat also gets stronger; buying a GPU that does not support this framework makes less sense over time. CUDA creates a network effect that makes it even harder for competitors to take share.
The above operational pattern can be generalized, as you can see in the graphic below. When a company spends decades on a technology that fundamentally enhances humanity’s ability to process information and/or harvest energy, the catalyst–e.g. a sudden technological shift– incites exponential growth that appears to have come out of nowhere. The driver of the algorithm is the ever-rising demand for computation. Whatever we choose to call it–AI, data science, etc.–the demand for computation is rising exponentially.
The above generalization reflects Palantir’s fundamental situation. After two decades in the trenches, they’ve mastered digital twins. Recently, they launched AIP (artificial intelligence platform), via which they sell, deploy, and operate digital twins, thus exploding the number of end applications and margins. As you might’ve guessed by now, AIP is to Palantir what CUDA is to Nvidia: it’s turning Palantir’s digital twins from cryptic and clunky pieces of software into something anyone can use.
With every incremental improvement to productization, Palantir becomes that much more difficult to replicate. Competitors will somehow have to not only build digital twins that work across all industries, but also to productize and sell fast to ensure Palantir doesn’t implant itself inside every organization across the Western economy–as Nvidia has accomplished to date, installing both their chips and its CUDA software in practically every every imaginable scenario. And at the moment, Palantir has no real competition and productization is well on its way.
I’m often asked if I’ve modeled, quantitatively, Palantir’s 100X growth over the long term. Such models can show whatever numbers you like with the desired inputs. The only question is whether Palantir realizes such an explosion in free cash flow per share, which can only be predicted through qualitative factors–i.e. mental models–such as the history of The Innovation Stack and how it applies to Palantir today. Which is to say, my answer is a resounding no. The only way to have forecast their growth is by understanding that Palantir had a unique offering and that its culture is world-class.
The quantitatives–i.e. the financials–all follow. With productization likely, margins were more likely to rise. In early stages of the “Nvidia algorithm,” qualitative models are far more powerful predictors than any Excel sheet could ever be. It can’t predict AIP, digital twins, or the rise of AI.
Per the graph below, Palantir reached 321 commercial clients by the end of Q3 2024. Such rapid growth evidences the accelerating rate of productization. But the truth is, Palantir hasn’t even started to scale. 321 clients fails to represent even 0.1% of the entire market. By the size of the runway ahead, there’s plenty of upside for Palantir.
Much like CUDA, AIP will generate network effects that deliver ten times more value to customers, to ten times more customers, at a tenth of the cost. While 100X seems quite fantastical after this early runup, achieving these gains for customers would demand such growth. I expect plenty of corrections along the way. As evidenced by my Amazon and Meta theses, corrections are usually an opportunity if your qualitative model points to higher levels of FCF in the future.
Both stocks dropped over 50% in 2022, driven by the market’s myopia. But a simple qualitative analysis paired with simple numbers revealed that both organizations were simply gearing up to achieve new financial heights. Indeed, there is no substitute to a deep qualitative understanding of companies, because if you simply look at numbers with no context you are likely to cheat yourself out of powerful long term returns.
At the moment, digital twins are monolithic. Meaning, they’re simply a real-time digital copy of the organization’s operations that allows you to plug data into AI models. However, over the coming decade, I believe we will see digital twins interconnect. Once everyone in your supply chain has digital twins, it only makes sense to connect them. The amount of value that can be delivered to customers at a marginal cost via the resulting network effects is unfathomable today, but this leap would transform Palantir from a company that powers other companies to one that powers entire industries.
By connecting digital twins, Palantir jumps from the automation of functions within organizations to doing so across organizations. It’s similar to the idea behind my Spotify thesis: by adding new verticals, they’ve managed to increase ARPU (average revenue per user) at a marginal cost. They’ve leveraged their existing platform to deliver substantially more value without increasing costs. And with a bit of added focus on the cost structure, the result is spectacular growth in free cash flow per share levels, as you can see below:
The upside will likely continue to multiply as different layers of our civilization continue to adopt digital twins. In the coming decades, digital twins will come to guide every decision to optimize time and other scarce resources. In this manner, Palantir could become a key component of our civilizational fabric by not only powering industries, but goverments.
The probability of Palantir capturing this opportunity is meaningful so long as the corporate culture remains sharp. You may have noticed that CEO Alex Karp makes frequent mention of Palantir’s culture. He knows it broadly determines their fate.
Here’s a particularly insightful comment of his during the Q3 2024 earnings call, showing that he avoids thinking in quantitative terms as well. He understands Palantir’s competitive edge comes from qualitative and subtle elements:
That the financials of Palantir would flow from our products and our culture and our way of implementing that we would bring violence and death to our enemies, while making targeting and general issues of safety better for our allies and for Americans, that we would stand by our values in thick and thin, including that the west and America are superior ways of organizing […]
In Q1 2016 Nvidia’s free cash flow per share was ~$0.05. Four years later it came in at $0.18 and as of Q3 2024, it came in at $2.30. To summarize, the main drivers of performance have been:
Nvidia having no real competition during this period of time.
An exponential increase in ease of deployment.
An exponentially rising demand for computation.
If my assessment of Palantir’s competitive position is correct, I believe we will see Palantir’s free cash flow per share levels rise similarly. Palantir’s free cash flow per share in Q1 2021 came in at 0.076, which is roughly where Nvidia was at in Q1 2017. The two other conditions are being met at present and I believe the last one will be met over the long term, regardless of what AI turns out to be. If Palantir checks the above three boxes, I believe we are going to see a winner-takes-all scenario emerge, with Palantir adding ten times more value, to ten times more customers at a tenth of the current cost.
Again, this doesn’t mean that there won’t be corrections along the way. As I’ve explained previously, exponential trends tend to defy common sense. They make little sense at the beginning when progress is scarcely visible and even less sense down the line, when progress is staggering. This is a relatively new phenomenon, and both Nvidia’s and Palantir’s stocks are instances of the market trying to put a price on an exponential trend. However, because humans are linear thinkers, we can only expect a bumpy ride going forward.
Lastly, many ask me whether AI agents could at some point displace Palantir’s platform. My view is that agents will be able to very quickly and cheaply create high-performing digital twins if they have the adequate data. You can only train an AI model to automate a task that you have plenty of data on. By extension, you can train distinctive AI models if you have data that no one else has. This is why, as I teach in my Tech Stock Goldmine course, proprietary data is the moat of the 21st century.
As Palantir speeds up its distribution via increased productisation, it obtains a data advantage. If Palantir does evolve into a winner-takes-all, it will have more and better data that anyone else on how to create a high-performing digital twin. In turn, Palantir will be able to use this data to train AI models that make the creation, deployment, maintenance and operation of digital twins far easier. This will further reduce deployment times, which will accelerate distribution and so forth.
The long term implications of this flywheel are vast and present even more upside than discussed in this write up. In essence, the flywheel promises to position Palantir at the top of the cloud computing sales funnel. We saw in my recent Chegg deep dive what happens when someone conquers the top of your sales funnel.
Until next time!
⚡ If you enjoyed the post, please feel free to share with friends, drop a like and leave me a comment.
You can also reach me at:
Twitter: @alc2022
LinkedIn: antoniolinaresc
Ojala no tarde tanto como Nvidia. Como indicas, las correcciones llegarán, pero seguro que tú seguirás analizándola para no dejarnos llevar por las emociones. Un abrazo