How ChatGPT and AI Model Monetization Works: Palantir's Manifesto

Palantir has published a manifesto against the dominant business model of the AI industry: monetizing ChatGPT, DeepSeek and other neural networks by the tokens they process. It urges companies to take back control of their own data and infrastructure.

Nine principles against the token economy

The manifesto urges organizations to stop depending on third-party cloud AI models and to regain control over data, algorithms and infrastructure.

Key positions:

• Organizations must accumulate and protect their own data, without handing it to third-party providers
• Model weights should belong to the organization itself, not be rented by subscription
• The main target of the criticism is "tokenmaxxing": the practice of maximizing AI providers' revenue by increasing the volume of requests and the length of responses
• Sovereign AI means deployment on the client's own infrastructure, with no dependence on external APIs

The term "tokenmaxxing" is the central concept of the document. Palantir points to a systemic conflict of interest: a cloud AI provider earns more when the model generates long answers and when the API is called more often. That incentive structurally runs against the interests of a customer who wants accurate, short and cheap answers. The more "verbose" the AI, the more profitable it is for the provider, and the more expensive for the organization.

Who the manifesto is aimed at

The audience is obvious: government bodies, defense agencies and large corporations, Palantir's traditional client base. Since 2003, the company has specialized in supplying analytics platforms to the governments of the US, the UK and NATO partners.

For these clients, data sovereignty is not rhetoric but an operational necessity. Processing intelligence data, trade secrets or citizens' personal data through a third-party cloud provider's infrastructure creates legal and political risks that are incompatible with regulators' requirements.

Palantir's market logic is transparent, though: the company stands to benefit directly from a shift to "sovereign AI." TNW describes the manifesto bluntly, as "a pitch disguised as a principle."

"It's a pitch disguised as a principle,"

TNW said of Palantir's manifesto.

Who is in the crosshairs

The manifesto implicitly attacks the business models of the leading AI providers: OpenAI, Anthropic, Google and Microsoft. Their revenue depends directly on token volume: the more requests the system processes, the higher the company's income. Palantir argues that this very mechanism creates incentives that run counter to the interests of corporate and government customers.

Palantir offers an alternative: its own infrastructure, its own weights, its own data. The initial investment is higher, but the organization gains independence from a specific vendor's terms and avoids vendor lock-in, the situation in which a provider raises prices, restricts functionality or stops operating in a needed jurisdiction.

The irony of Palantir's position is that the company is itself a public corporation focused on growing its contracts. Its arguments for sovereignty are no less commercial; they are simply monetized differently: through licenses, deployment and support rather than tokens.

What this means

Palantir's manifesto marks a widening rift in the AI industry: the cloud model versus the sovereign one. As AI becomes embedded in the critical infrastructure of states and corporations, this dispute will only sharpen. Palantir is betting that large institutional buyers will ultimately choose control over the technology rather than the convenience of cloud access.

How does monetization of ChatGPT and other AI models work?

By tokens: the more characters the neural network processes, the more you pay. Palantir offers an alternative: your own models and infrastructure instead of dependence on cloud AI services.

Why does this matter for business?

AI sovereignty means protecting your own data and algorithms: companies do not hand sensitive information to third parties such as OpenAI or DeepSeek.