In a characteristically fiery, unfiltered press circuit, Palantir CEO Alex Karp has launched a direct offensive against Silicon Valley’s frontier AI labs, issuing a strict nine-point “AI Sovereignty Manifesto” warning corporations and militaries against outsourcing their digital futures.
Appearing on CNBC’s Squawk Box on July 1, 2026, Karp went heavily off-script, tearing into the subscription and pay-per-token models popularized by companies like OpenAI and Anthropic. He labeled their current pricing structures a value-less “wealth tax” on businesses and declared that relying on commercial cloud-hosted models for critical infrastructure or national defense is “effing insane.”
1. The Core Argument: “Controlling Your Weights is Controlling Your Fate”
The backbone of Palantir’s argument—published simultaneously as an official corporate manifesto on X (formerly Twitter)—is that true AI Sovereignty requires an organization to completely own and control its data and model weights, rather than routing proprietary queries to external servers.
- Data Retention is the Treasure: Karp argues that companies are blindly handing over their most valuable institutional memory, code, and operational tradecraft to model providers, who use that data to improve their own commercial products.
- The Weight Protection Mandate: “Weights” are the underlying mathematical values that dictate what a neural network has learned. Palantir insists that if an enterprise does not host and possess its own model weights inside its own secure perimeter, it does not actually own its AI.
- The National Security Risk: Karp directed his sharpest critique at the U.S. Department of Defense’s increasing reliance on public cloud infrastructure:
“Are we really going to outsource the battlefield of this country to the consensus view in Silicon Valley? That is effing insane.”
2. Declaring War on “Tokenmaxxing”
Palantir’s manifesto coins a aggressive new industry term targeting the commercial tech landscape: “Tokenmaxxing.”
[ Tokenmaxxing Loop (The Industry Standard) ] ──► Metered, pay-per-token pricing models (OpenAI, Anthropic)
│
▼ (The Financial Strain)
[ Enterprise Reality Check ] ──► Soaring API bills even as base token costs fall
Fails to generate real enterprise ROI / productivity gains
│
▼ (The Alternative)
[ Sovereign Weights Framework ] ──► Host open models inside your own walls (Nvidia Nemotron)
Build permanent, durable enterprise logic off-cloud
Karp stated that “something has gone completely wrong” with how generative AI is currently bought and sold. He accused frontier labs of “stealing the weights and alpha of my business” by locking corporations into continuous usage loops, noting that heavy token usage simply rewards throwaway scripts and endless, unstructured chatbot prompt cycles rather than solid enterprise software.
3. The Product Pitch Behind the Principle
While the manifesto reads like a deeply principled defense of corporate autonomy, industry analysts quickly pointed out that the argument perfectly aligns with Palantir’s core business model—and a major new product rollout.
Alongside the manifesto, Palantir announced an expanded, deep-tier intelligence engine built in partnership with NVIDIA. The framework allows government agencies and highly regulated enterprise clients to deploy Nvidia’s Nemotron open-weight models directly inside Palantir’s secure, air-gapped stack (AIP, Ontology, and Foundry).
| Evaluation Metric | The Commercial Token Labs | Palantir Sovereign Architecture |
| Hosting Surface | Multi-tenant commercial cloud providers. | On-premise, secure sovereign cloud, or disconnected edge. |
| Pricing Model | Metered, volume-based pay-per-million-tokens. | Enterprise software packaging / value-driven licensing. |
| Data Protection | Data is systematically processed via external vendor API tunnels. | Context, prompts, and weights remain 100% inside corporate perimeters. |
The aggressive framing yielded immediate dividends for the company on Wall Street. Following the CNBC segment and the product unveiling, Palantir stock (PLTR) skyrocketed roughly 9% to close at $127.31, leading a massive rally across the broader AI software sector. By positioning pay-per-token AI as an extractive “wealth tax,” Karp is attempting to carve out an entirely new enterprise category: positioning Palantir as the ultimate cure for companies waking up to the data and financial risks of modern cloud dependencies.
