원문 보기
So, Michael Bury made the bet of his life against Palantir. He built it up with options last year, and he still holds it , even after closing his own fund. Michael actually made one post that contributed to Palantir shares falling over 6% the next day. But there's a hole in his words, and frankly, the company he chose might be his worst target .
So his track record of making such predictions matters, and I want to discuss three points with you . Years ago, he warned that a collapse was coming. After that, the market grew by more than 50% over the next few years. In 2019, he compared index funds to the 2008 subprime mortgage bonds that exploded.
It turned out that everything was fine with index funds . And around March 2021, he stated that Bitcoin was like a bubble in the housing market. Then, in June , he warned of the biggest crash in history. The bear market did indeed arrive in 2022, and Bitcoin eventually fell by about 70%, but the housing market-style crash he described never happened.
So, if you acted on every warning, you would have missed the historic bull market. And to his credit, he knows it. Here is his own post this year. I became a meme because of the number of times I predicted a crash. I became the boy who cried wolf. So this is the man who is now betting against Palantir, and here's how he did it.
In the fall of 2025, Michael's fund showed a large short position in Palantir in bonds. The same reporting showed signs of weakness against Nvidia. So , a put option only makes a profit when the stock falls. He later said that these options are valid for four years. So, he's not betting on a bad week, he's predicting a multi-year decline.
And in February, he compared Nvidia to RCA during the radio boom of the 1920s. RCA fell about 98% from peak to trough, even as radio itself continued to grow for decades. So this comparison indicates what he thinks the future will be like. He then closed the fund and kept the Palantir stake for himself.
So in April, no longer having a fund, he publishes one line on the Internet. Anthropic, he wrote, is ahead of Palantir. And then he deleted it. So, he can't get out quietly, he can't get out cheaply, and the time on these options is running out all this time. So what is he really saying? Well, that post contained four arguments, and I'll give you a strong version of each of them.
First, growth. Anthropic has grown from a billion- dollar valuation to over thirty billion in a matter of months, and its annual recurring revenue is now estimated at around 60-70 billion. This is the current pace, the annual rate at the moment. Palantir, he said, took twenty years to reach five billion dollars.
The second is the product, and he says it's pretty simple. According to him , Anthropic is a simpler, cheaper, and more intuitive option for businesses. So, it seems more logical to people. Palantir, he says, is expensive and complicated. Third, and perhaps his strongest argument, he rejects what Palantir is best known for.
According to him, let Palantir deal with the public sector, where margins are low and volume is small. And at the heart of all three points lies his real argument. Companies are creating direct connections with AI model developers. So if you don't have your own top model, you are effectively a middleman, and middlemen are being squeezed out.
The fourth argument is a number, and it is the only statement in the entire publication that can be verified. Now there is evidence that he didn't mention, so I'll put all of that on the table as well. Anthropic began to branch out into specific industries. The company has already released tools for healthcare, then came finance and KYC— checks that the bank conducts for each client.
These agents work inside Excel, PowerPoint, Word, and Outlook. And after the financial package, a legal one appeared. It also signs agreements with the world's largest consulting firms. Accenture has created a separate group and plans to train about 30,000 specialists. TCS integrates Claude into claims processing for insurers, and Anthropic's share of corporate spending on the model has grown from 24% to 40%.
So, the pessimistic scenario here is not fictional. The cutting-edge lab purposefully builds tools for specific industries and pays consulting firms to implement them in the companies Palantir sells to. The only question is whether this describes what Palantir actually does. So, there are three points where this argument falls apart.
This is a number he misrepresented, a business he called small, and a level he doesn't seem to know exists. Michael wrote that Anthropic takes 73% of all new corporate spending , and cited data from the Ramp platform. Ramp data suggests otherwise. First-time buyers prefer Anthropic over OpenAI. So these are two model developers among those buying AI for the first time.
This doesn't measure all new corporate spending and says nothing about Palantir, which isn't even in this comparison, because Palantir isn't a chatbot subscription you can just buy with a corporate card. So, the statement with the big number describes something else, and the entire conclusion is actually based on that.
The growth trend that has been identified is actually quite impressive. Almost one in four businesses on the platform now pays for Anthropic. A year earlier, it was, just think about it, every 25th. So, the shift is significant, and it's getting bigger , bigger, and bigger. This is simply not an indicator that someone is taking anything away from Palantir.
Yes, and now the state statement, and, frankly, this is where I get uncomfortable. Michael called Palantir's public business low and small. Palantir's revenue from the US government was $809 million last quarter, up a modest 90% year-over- year. So, it's not small, and 90% of the growth is not a tedious, low-margin contracting business.
And as for size, Shyam Sankar, Chief Operating Officer , gave a figure worth hearing. Shyam said Palantir's 12-month share from the War Department is meager. He estimated it at less than a quarter of 1% of the Pentagon budget. So when you think about how much Palantir is only taking in from the entire budget, that's the business opportunity that Michael calls small.
But a bigger problem for his argument arose in March. Anthropic and the Pentagon had a falling out over security regulations. So the Pentagon banned the lab from working with the War Department, and contractors, including Palantir, had to remove Anthropic from those systems. Now compare this to the bears' arguments .
The argument is that Anthropic is displacing Palantir, and in Palantir's largest, most sensitive segment, Anthropic didn't even compete. It was completely removed from the building. Meanwhile, Palantir continued to receive contracts from the Pentagon. This involves the Maven program, and from what I can see, it's not luck.
They do this all the time. Sales to the defense and intelligence sectors require permits and official approval. You have to implement systems into secret networks, and procurement cycles take years. Frontier Labs is built for rapid, consumer-style implementation, so they are not suited for this kind of work.
So they are not as set up as Palantir to get that kind of work. Which brings us to the third point, “ roars,” which simply means things that keep competitors out . Palantir's key asset is something called ontology. This word scares many people, so here's what it means. Most apps simply show you data, like dashboards and numbers on a screen.
An ontology is a little different because it is a living map of how an organization works. He connects people, cars, cargo, money, and decisions, so he knows how it all connects. Imagine a bank. A normal system shows you a list of transactions. The ontology knows that this account belongs to this customer.
This client is associated with this device, and the device just logged in from a location that is not consistent with the user's habits. So fraud is detected at the moment of its commission. No need to wait for analysis and reports to notice it later. This is Palantir's secret ingredient. Add a language model on top of this.
The model itself is incredibly capable, but it knows nothing about your business. And if you ask her about her supply chain, you'll only get a general answer. Ask an ontology-based model — and bingo, it will give you an answer about your supply chain. She knows your real suppliers and your true limitations.
So, this card is the product. Ontology is a key element here, it cannot simply be bought or hastily created. This is accumulated knowledge about how a company actually works. So who actually built this map? Well, this is the second part of this story, and his name is, of course, Alex Karp. Alex runs Palantir, and he wasn't diplomatic when he called the shorting market manipulation on CNBC.
Whatever you think about it , he openly attacked it on television, and Alex has been advocating the opposite position for 20 years. Palantir has spent nearly 20 years learning how to build such maps in the most complex environments that exist. Alex described the early days on the latest earnings report, and honestly, it sounds pretty tough.
Back then, there was no AI to work with. Alex said that Shem and the others had Blackberrys strapped to their heads. They did this to make the code work in sensitive areas, and created the first rough version of everything that came after. They also did something that almost no software company would do .
They sent their own engineers to work directly at customer offices. They sat next to analysts, reworking the software for real-world problems. It's slow, it's expensive, and it doesn't scale the way investors like it, at least traditionally . But that's why Palantir is almost never removed once it's implemented.
You can cancel your chatbot subscription at the end of the month. You cannot undo the system by which your organization understands itself. And for me, that answers this whole vertical approach. The agent template that creates a presentation for investors is a tool for a specific task . It's useful, and it will sell well, but it only works with the documents you provide it with.
It doesn't know about your risks. It doesn't know who you owe money to or how it relates to the reporting required by the regulator. So, this is a whole different level, and regular templates won't help here. You need an ontology. Now look at these consulting agreements as well. Accenture and TCS receive money for implementing Claude in client companies.
This is a service business where know-how is accumulated within the consulting firm , rather than in a system owned by the client, which, in my opinion, is closer to the model that Alex has been opposing for 20 years. So, Michael says it took Palantir 20 years to reach $5 billion. When he published the post, Palantir had about 4.5 billion.
It crossed the 5 billion mark only after he posted. These 20 years have not been slow growth . This was the phase of laying the foundation for everything that is happening now. And I believe that this is exactly what the competitor cannot accelerate. You can't raise 30 billion and just buy this knowledge.
Understanding how the Department of Defense moves information takes time and experience. This is the kind of argument that Alex Karp actively promotes in the industry. He believes that most companies are making a huge mistake with artificial intelligence that they have n't noticed yet. He calls it “token maximization,” which simply means putting money into use and calling it a strategy.
So, you subscribe to the services of a leading lab, spend money on usage, and give away your data, your prompts, and your internal know-how. These are all the established methods of operating your business. So all of this becomes material that the lab learns from, and then the lab that you pay uses it to create things that compete with, guess who?
With you. Alex's statements here are actually quite direct. He says that the labs believe they deserve to colonize your facility. He said his own clients refused to become vassals of language labs. During that call, Alex called it a revolution for AI independence and sovereignty. It simply means that you own your own things.
So, putting the complicated words aside, it's about who owns the value. And this is important because the Palantir model works exactly the opposite. The client stores data, prompts, model weights, and importantly, the ontology. And they also keep the money that this whole process brings in. So Palantir gets a share of the value they create.
And this is where the mediator theory collapses. Palantir is model-agnostic , so it doesn't care which model you plug in. Grok works inside its platform, like the others. So, Palantir doesn't need to win the modeling race because it doesn't participate in it . It stands a level above and benefits from someone who is ahead of the competition every quarter, every week, or every month.
And now Palantir is directly bypassing these labs. During the reporting conference, Shyam Sankar spoke about a large technology company from Silicon Valley. She set up a test between the Palantir platform and the model developers' interfaces. Guess who won ? Palantir. And the deal turned into a contract worth $10 million a year.
So if Palantir can beat the modelers in a head-to-head fight, why is the world's most famous shorter still holding onto his short position? Because Michael's argument is based on the fact that companies should go directly to the model developer, bypassing the control layer. And Alex's argument is exactly the opposite.
As models become more powerful, more companies want to have an intermediate level. They need something between these models and their core assets: data, contacts, business knowledge, everything. Now let's look at the practical aspect without philosophy. If you are a bank, a hospital, or a defense department, you have rules: where data is stored, who sees it, and what happens to it.
Giving all this directly to the model supplier is quite uncomfortable. In many cases, it is even legally prohibited. So, in my opinion, the numbers show that customers choose Alex. Last quarter, revenue grew by 93%. This is the highest growth rate in the company's history. The American commercial sector is exactly what, according to Michael, fascinates Anthropic.
It grew, wait for it, by 849% . Net dollar retention shows that customers are spending more than they did a year ago, which is not like a company being absorbed by a competitor. That quarter, they closed 221 deals worth more than $1 million each. These are real companies spending real money. And the remaining value of the agreements, that is, the work that still needs to be done and has already been signed, now stands at $13.1 billion.
This is 83% more. So clients aren't just considering options , they're making commitments for years to come. And the company is profitable in a way that almost no other fast-growing software company manages to be. Last quarter, revenue was just over a billion dollars, and the " rule of 40" score reached 85.
This score combines growth and profit into one number , and a value above 40 is considered good. It has been growing for 12 consecutive quarters. Palantir then raised its forecast for the year to $8.15 billion. This is the largest increase in the company's history, and the market reacted instantly. After that report, the stock rose by about 29%.
On short positions alone—wait for it —traders lost about $3 billion in a single trading day. And Alex called this quarter extraordinary. On CNBC, he told people to forget about consensus. He stated that he would develop the business at or above current commercial rates in the United States. And this is for the next 18 months.
His argument was actually very straightforward. For the first time, people believed us. So, this is the person Michael is pitted against. 20 years passed, and people finally started taking Alex at his word. Wedbush analyst Dan Ives called Michael's post a fabricated story. He described Palantir as the epicenter of the AI revolution's leaders.
And his thesis was simple. Anthropic's $ 60 billion success is not at the expense of Palantir. So what are the "bears'" arguments really based on ? Well, they view it as a zero-sum game. This means that the growth of one company must come at the expense of another. But look at what each of them actually sells.
Anthropic sells intelligence . Palantir sells a map that makes intelligence useful for your company. Therefore, by buying more of one, you have more reasons to buy the other, which means both companies can grow enormously at the same time. And, honestly , that's my own opinion. They solve different parts of the same problem, and Anthropic's success shows that companies are seriously investing in AI.
This is exactly the market in which Palantir wants to sell its services. The simultaneous growth of both is something that, according to the logic of zero-sum theory, should not happen, but that is exactly what we see in the numbers. So, if you've watched the video up to this point, then you probably like the content we create.
I've been creating content for social media for almost ten years, including helping founders, creators, and businesses grow and make money on YouTube. Here , for example, is Jeremy. He has almost a million subscribers. Here's what he said about us. This is a huge help. Well, it also obviously increases revenue.
David has done an incredible job helping me manage this channel overall. And many other authors said the same thing . And then there's Farzad. We helped his channel grow from 40,000 to over 400,000 subscribers. Behind the scenes, the success of going from 40,000 to 400,000 subscribers is a credit to David Carbott and his team, my wonderful editor.
Without him, this channel would be just a shadow of what it is now. Even Elon Musk shared our work, mentioned me, and follows me on X. We help people grow on YouTube: from aspiring creators to companies looking to grow. Follow the link below to learn more or book a call. Additionally, if you want to purchase a course on AI or Palantir, links to those are also below.
Thank you to all the participants who support this channel. They get access to exclusive videos and early access to content. If you haven't already , subscribe, and I'll see you in the next video. See you later.
댓글 0