Going sideways

© 2015 Getty Images / Getty Images News via Getty Images

Going sideways applies when Karp carps about OpenAI and Anthropic and the notion of control
as control over tech as powerful as AI is, becomes a loaded question for the ages.

Karp’s argument was that businesses are exhausted by paying for tokens. As he put it: “The basic view among enterprises in this country is I’m going to chillax and waste my time with tokens.” According to CNBC’s Samantha Subin, Karp took aim at the token model used by OpenAI and Anthropic as AI costs skyrocket. He further argued customers are shifting away from “tokenmaxxing” toward ROI and open-weight models that deliver similar work at a fraction of the cost.

Ahead of the interview, Palantir posted a 9-point “AI sovereignty” manifesto on X. Earlier that week, the company expanded its partnership with Nvidia (Nasdaq: NVDA) to build custom models for U.S. government agencies. Karp’s framing of that alliance was revealing: “What aligns me with Nvidia, and I think is what the technical customers want, which is control over their compute, their models, their data stack and their alpha. They want to know they own the means of production. It’s not being transferred to someone else.”

Hate to say but Karp's correct. 

Token-cost fatigue is showing across businesses: Uber (NYSE:UBER) has reportedly capped employee spending at $1,500 per month for each agentic coding tool, including Claude Code and Cursor, after blowing through its AI budget in four months. For readers tracking the picks-and-shovels layer of this shift, our team’s AI infrastructure research maps the suppliers benefiting most.

The Disconnect and the Bear Case

Palantir shares closed at $126.79 on July 10. The stock is down 28.67% year to date, even as operations accelerate. The stock trades at a forward P/E near 91, and Michael Burry’s Scion Asset Management disclosed a new put position tied to 5,000,000 Palantir shares in its Q3 2025 13F filed November 3, 2025, an underlying notional of about $912 million. 13Fs don’t disclose strikes, expirations, or whether the position is still open.

What to watch: whether the “own the means of production” pitch keeps pulling U.S. commercial customers. Palantir’s U.S. commercial remaining deal value (RDV), a measure of contracted business still left to recognize, reached $4.92 billion in the latest quarter, up 112% from a year earlier.

But it's deeper than that due to the platform on which AI runs, the thermodynamic nightmare of the data center requiring ongoing replacement of every server, in system, every 24-30 months in order
to keep telling grandma how to make a better apple pie.

This doesn't compute.

Number of data centers by country - Statista

What is a data center?

A data center is a physical room, building or facility that houses IT infrastructure for building, running and delivering applications and services. It also stores and manages the data associated with those applications and services.

Data centers started out as privately owned, tightly controlled on-premises facilities housing traditional IT infrastructure for the exclusive use of one company. Recently, they've evolved into remote facilities or networks of facilities owned by cloud service providers (CSPs). These CSP data centers house virtualized IT infrastructure for the shared use of multiple companies and customers.

History of data centers

Data centers date back to the 1940s. The US military's Electrical Numerical Integrator and Computer (ENIAC), completed in 1945 at the University of Pennsylvania, is an early example of a data center that required dedicated space to house its massive machines.

Over the years, computers became more size-efficient, requiring less physical space. In the 1990s, microcomputers came on the scene, drastically reducing the amount of space needed for IT operations. These microcomputers that began filling old mainframe computer rooms became known as “servers,” and the rooms became known as “data centers.”

The advent of cloud computing in the early 2000s significantly disrupted the traditional data center landscape. Cloud services allow organizations to access computing resources on-demand, over the internet, with pay-per-use pricing—enabling the flexibility to scale up or down as needed.

In 2006, Google launched the first hyperscale data center in The Dalles, Oregon. This hyperscale facility currently occupies 1.3 million square feet of space and employs a staff of approximately 200 data center operators.

A study from McKinsey & Company projects the industry to grow at 10% a year through 2030, with global spending on the construction of new facilities reaching USD49 billion.

Eating the world applies along with the word assume.

Our gigafactory of compute.

Built in 122 days — outpacing every estimate — Colossus was the most powerful AI training system yet. Then we doubled it in 92 days to 200k GPUs. This is just the beginning.

Or the end when you think about it, eh?