Why your business needs an AI sovereignty strategy
Palantir's Alex Karp says sovereignty will decide your institution's future, and he's largely right. Every business handing its data and decisions to an AI model it doesn't own or control, not just governments building infrastructure. Here's what owning your AI actually looks like, and why it's the more efficient choice too.
Palantir CEO Alex Karp recently put out a nine-point manifesto on X on the importance of AI sovereignty. In it, he urged institutions to keep their data, own their model weights, and stop "tokenmaxxing."
“Your AI sovereignty dictates your institution’s future,” he wrote, adding: “Relinquishing sovereignty transfers the future choices of your institution to others, who are likely to exploit it for their gain and your loss.”
In layman’s terms, what he was talking about is the difference between owning your tools versus renting them.
If a business feeds all its information into someone else's AI system (like ChatGPT or another big model it doesn't own or control), it's handing over the raw material, the customer data, the internal know-how, the way it makes decisions, to a company that isn't obligated to protect it or keep it exclusive.
Over time, that outside company learns the business's playbook and can use it however it wants, including building a product that competes with or replaces you.
What’s more, businesses are paying for the privilege.
Karp's point is that this isn't a minor technical detail, it's existential. The businesses that own their data and models stay in control of their own future. The ones that don't are, in effect, just tenants in someone else's system.
AI labs are charging for tokens that create no value.
Karp isn't a household name in the way Sam Altman or Elon Musk are. But in the AI economy, he matters. People love him or hate him.
The US data and AI company Palantir is best known for its defence, intelligence and immigration enforcement work, including contracts with ICE, and for software that helps large institutions make decisions in sensitive, complex environments.
He talks about big ideas like ontology, something we're passionate about at Decidr, too.
His last manifesto was a 22-point summary of his book The Technological Republic, posted on X in April. It racked up tens of millions of views, got dubbed "the ramblings of a supervillain" by a UK MP, and put more than half a billion pounds of Palantir's British government contracts under review.
Last week’s statement argued that sovereignty is "the precondition for choice," that "data retention is your treasure," and that tokenmaxxing creates the "addictive feeling of false progress."
Karp then went on CNBC to discuss Palantir's new Nvidia partnership, and took it even further.
He criticised leading AI labs, saying enterprise customers were fed up paying rising token bills without seeing the value back, and called the situation "effing insane."
One outlet described the interview as ‘heated’. That was one of the more polite characterisations.
But the more interesting read wasn't that Karp sounded angry. It was why.
His argument is that frontier models have been oversold, that token usage has been mistaken for business value, and that companies are handing over the data, context and decision logic that make them competitive.
The market seemed to understand the strategic point. Palantir added roughly $22 billion in market value the day the Nvidia deal and the sovereignty argument landed together.
Why sovereignty is the issue of the day
Karp is loud about it, but he's not alone, and this isn't a one-company argument.
Governments are backing that concern with capital: sovereignty requirements could shape $500 to $600 billion of global AI spending by 2030.
Australia is having its own version of this argument right now, just with less profanity.
Government investment is racing to bring the AI giants onshore, and they're racing back: NEXTDC and OpenAI, Microsoft, AWS and Anthropic have all committed billions of dollars to Australian data centre and compute deals over the past year, while Google is stalling over tax concerns.
Strip away the politics and the compute spend, and the question underneath all of it is simple: who controls your data, your decision logic, and the thing that makes you different from your nearest competitor.
That question doesn't stay contained to governments and the Fortune 500. Every business handling AI faces it, whether or not anyone's told them yet.
Karp's manifesto has nine points, we'd add a tenth
This isn't a new argument for Decidr. We've been making a version of it since before Karp posted his first point, and we've used a different name for the same mechanism: intelligence renting.
Businesses are giving away estimated trillions in company knowledge. "Australian companies using multiple AI tools are helping the hyperscalers get really good at learning each part of their business," Decidr Co-CEO David Brudenell said. “Once it's gone, it's gone.”
"If Big US Tech follows their traditional business model, they will eventually learn Australian business so well they will create products out of it and sell it back to us, as they have done with SME products."
"We should learn from the dominance of Meta, Google and others," Brudenell said.
Karp's manifesto calls this the transfer of alpha. Brudenell calls it the great knowledge heist. Different vocabulary, same mechanism, and the same warning: sovereignty lost to a platform in one era gets sold back to you as a product in the next.
Where we've pushed further than Karp is on who else is exposed. It isn't only SMEs handing knowledge to a model they don't control.
Large corporates using external management consultancies face the same problem at one remove: "You currently have no control over the AI model they are plugging your company knowledge into, and you don't know that you are renting your intelligence to a global AI multinational who will learn it to no benefit to you."
How Decidr handles sovereignty
Saying this is one thing. Building for it is another, so here's where it shows up in our own product and our own capital allocation.
DecidrOS allows businesses to bring their own model, but runs it inside a boundary, on infrastructure the business itself controls, which keeps learning their business rather than everyone else's. The company owns the model, the learning, and the data unique to it. That's the difference between using AI and renting it.
There's a practical dividend to owning the model rather than renting it too: it's more efficient, not just more sovereign. A model built to run one business needs fewer tokens to get a useful result than one built to run every business on earth.
Karp's manifesto is an argument. We think the argument is right. We'd add one line to his nine: sovereignty isn't just something governments, armies and big businesses need to get right.
It's something the accountant with 40 staff needs too, and right now, almost nobody is building it for them.
