To The Moon

Palantir and NVIDIA have a key partnership.

$Palantir Technologies Inc.(PLTR)$  

$NVIDIA(NVDA)$  

Palantir + NVIDIA Signal the Rise of Sovereign AI for Governments and Enterprises

On the surface, it’s about customizing AI systems for the U.S. government. But the real point isn’t just another AI model, nor is it simply Palantir winning another government contract.

The real significance is that companies and governments are starting to demand re-control over the means of AI production.

In the past, many companies using frontier large models were essentially handing over their data, processes, prompts, and business logic to those models — and then paying token fees on top of it.

Karp believes this model is acceptable in consumer scenarios. For example, as individual users, if you steal my ideas, then fine — my ideas probably aren’t worth hundreds of billions anyway.

But when it comes to battlefield medicine, critical manufacturing, finance, and key infrastructure, companies that go to frontier model providers face a fundamental problem: they are no longer sure whether they can truly control their core assets.

So this Palantir-NVIDIA partnership is actually packaging three layers together:

NVIDIA provides compute power, AI infrastructure, and open models.

Palantir provides AIP (Artificial Intelligence Platform), ontology, permission systems, security controls, and the business application layer.

What government and critical infrastructure customers get is a more controllable AI system. Questions like:

Where is the data stored?

How does the model run?

Are the prompts secure?

Who controls the model weights?

Will my core business secrets leak?

…can now be re-defined and properly addressed.

Then we look at Karp’s interview video:

“We’re in the U.S., Iran, Islamic countries — all the critical infrastructure sites and everyone using LLMs on the battlefield. They’re all built on our traditional design. Our customers are not just unhappy — in forward operating labs and at Berkeley, they’ve welcomed us. Their discomfort and loss of trust…”

Interviewer: Explain what you mean.

Karp: When you use an LLM, every technician realizes it’s an important resource. But to make it valuable on the battlefield, at the management level, or in manufacturing, you need an application layer. We have what’s called “human-readable language,” but it actually uses the LLM while making it secure, useful, and precise.

It’s secure because it doesn’t touch the data it reads and learns from.

It’s secure because it blocks the LLM from processing and copying your business.

It’s secure because it doesn’t transfer your IP, combat methods, secret data, or the highest-level classified data — or sensitive medical data.

The general approach these people take — like Sam and Dario — I have interesting private discussions with Dario, so I’m not criticizing them personally. But some things are fundamentally wrong.

The basic view of these companies in this country is: “I casually waste my time and resources, I get no real value, and they get my IP.”

This sounds like a tiger trap… no, not a tiger trap. It sounds like…

Interviewer: You’re attacking them.

Karp: No, no, no. This is feedback. This is reporting what I hear. This is what clients are telling me.

I originally thought it was in my own interest to stay quiet — after all, I’m making money, right? But actually, you may not be like us at Palantir, or at Harvard, or Berkeley, but companies in this country trust us — especially when it comes to critical infrastructure, both public and private.

So I hope this country returns to vitality. I hope we and our friends around the world can possess the world’s best technological capabilities.

The secret behind Palantir is the model we’ve been deploying for the past five years. Everyone who has followed us for a long time knows that people used to say Palantir just provides services, that we don’t even know what “vitality” is. Now people only talk about this.

The secret is that we provide the best platform for warfighters. These warfighters have serious trust issues. But it’s not just the Department of Defense — private sector companies have the exact same problem.

“Why should they get my data? If they’re going to build my alpha, why can’t I control it?”

This is exactly the partnership customers want. My relationship with NVIDIA is what tech companies truly desire: control over their data, their models, their databases, and their alpha. They want to own their means of production instead of transferring them to others.

They don’t care about some virtual “Deploy Co” that moves money or transfers alpha to a third party. This problem is serious. We must find a way…

Our product is virtual. We are now selling it to customers, allowing you to switch from model to model. We are completely model-agnostic. But we need to rebuild trust.

This trust comes down to answering basic questions: Who owns the data? Where is it stored? Is the information secure? Has it been transferred to you?

If the information is truly valuable — suppose I could give you $100 million tomorrow — I would say: I can give you $100 million, but I only want 30%. If the information is worth that much, why are they only charging token fees?

Interviewer: So if I understand the key point, it’s a secure, American-made, open-source model. Then how long until your model can compete with frontier models?

Karp: What I’m saying is — although it’s virtual — it’s also self-centered. It’s the model + application layer + compute. Actually three things together.

So in our narrative, the value… look at our margins. Why is everyone leisurely using bad models and losing value? It’s because customers refuse to pay the real cost.

There are two places we can make money: profit and free flow. Our application layer is called ontology and compute.

We can use open models — in both standard and non-standard situations — to achieve frontier-level performance. But first, you control the value.

The real cost is not just what you pay, but what you lose — that is, the value of your business. You can have frontier applications with frontier models without worrying about transferring your business to someone else.

By the way, you can also use closed frontier models, but customers must be able to answer very basic questions: Did you retain the data? Are you joining my business?

In standard situations, when the strategic department tells you they need this application — do they control the value, or do you?

Are you really going to transfer this country’s strategic capabilities to the perspective of Silicon Valley? That is crazy.

By the way, every company in this country — especially private ones — many don’t want to speak publicly because they’ll be labeled as crazy or on drugs. But that’s not me. That’s my role.

But I’m telling you: in this country, every company I deal with is angry. They say, “I’m paying for worthless stock. These people are stealing my alpha inside my own company, and they’re creating a wealth tax. This doesn’t help poor people — it just punishes us.”

The reason is that these models have been irresponsibly over-hyped, and this over-hype is dangerous for everyone.

This is why Palantir and NVIDIA’s partnership is what technology companies want. They want to control their data, their models, their databases, and their alpha. They want to own their means of production instead of handing them over.

This sentence is very important.

In the AI era, the means of production are no longer materials, factories, machines, land, or traditional capital. They are now models, weights, compute, data, and business processes.

If companies hand all of these over to external model providers, then in the short term they are “using AI,” but in the long term they are surrendering their core capabilities.

Karp criticizes that some so-called “deployments” are essentially making companies pay for tokens while transferring their secrets to third parties. He says this game is about to end.

Palantir’s product is model-neutral — customers can switch between different models without being locked into one company.

He also raises a sharp question: If AI can really help you make a billion dollars tomorrow, why is the model company only charging you token fees? A more reasonable business model would be: I help you make a billion, I take 30%.

If they’re only selling tokens, it means the real value of the enterprise isn’t being realized by the model itself — or the risk/reward structure hasn’t been properly clarified.