You are being robbed by your AI vendors. You know it, they know it, but nobody wants to say it out loud because it makes everyone look stupid for buying into the hype.
Then Alex Karp walked onto a live television set and blew up the entire narrative.
During a bizarre, chaotic, and completely unscripted twenty-minute interview on CNBC, the Palantir CEO threw a massive wrench into the Silicon Valley hype machine. He didn't just criticize his rivals; he called the foundational business model of the entire artificial intelligence sector "effing insane."
While the internet argues over whether Karp had a televised nervous breakdown or a moment of pure genius, he managed to expose the ugly truth that corporate America has been hiding. Corporate leaders are spending millions on a technology that might actually be actively stripping away their competitive advantage.
The Great Token Robbery
When you use commercial foundational models from the heavy hitters like OpenAI or Anthropic, you pay by the token. You pay for the words going in, and you pay for the words coming out. Karp calls this "tokenmaxxing," and he argues it is a massive transfer of wealth that yields almost zero actual enterprise value.
Think about how a normal software contract works. You pay for a tool, the tool improves your efficiency, and your profit margins grow. Frontier AI doesn't work that way. Instead, you pay massive fees to process data, and while you pay, the AI vendor watches what you do.
Every single prompt your employees type, every proprietary dataset you funnel into an API to "fine-tune" a model, and every unique business workflow you try to automate is being processed by someone else's infrastructure. Karp argues that these labs are essentially double-dipping. They charge you a premium to use the model, while simultaneously harvesting your business secrets, weights, and alpha to make their next model smarter.
You're paying them to let them steal your data.
If these models truly generated the astronomical value their creators claim, the pricing structure wouldn't look like a utility bill. If a vendor can instantly unlock a billion dollars in value for your business, they should ask for a 30% cut of the upside. Instead, they charge you for the digital equivalent of minutes on a flip phone. Why? Because charging for tokens shifts all the financial risk onto the enterprise while the tech providers pocket the data.
The Illusion of Silicon Valley Trust
Enterprise tech buyers are terrified, but they're trapped by the fear of missing out. No CEO wants to tell their board they ignored the biggest technology shift of the decade. So, they keep signing the checks.
The defense from the closed-model heavyweights is always the same. Building frontier models is expensive. The GPU clusters from Nvidia cost a fortune. The energy grids required to run these clusters are massive. The capital expenditures are terrifying, which is why these firms are racing toward massive public offerings. They need your token fees to survive.
To keep the money flowing, the industry relies on a culture of blind trust. We are told to believe that these companies will protect our data simply because they say they will.
Karp didn't hold back on that front. He mocked the philosophical, safe-for-humanity branding that companies use to shield themselves from corporate scrutiny. The idea that an enterprise should trust a vendor just because they claim they've never lied is completely ridiculous at this level of business. When billions of dollars in proprietary market advantage are on the line, good intentions don't cut it.
The cracks are already showing. Companies are starting to hit a wall. Uber, for example, quietly capped its spending on AI tokens because the return on investment just wasn't adding up. The party is crashing because the bills are arriving, and the productivity gains look remarkably ordinary.
Outsourcing the Battlefield
The corporate data heist is bad enough, but Karp saved his real fury for how this affects national security. Palantir has long been a foundational tech partner for the U.S. military and defense sectors, a position that has drawn plenty of criticism over the years. Karp knows exactly how the defense tech pipeline works, and he is horrified by what he sees happening in Washington.
The U.S. government has increasingly leaned on consumer-facing Silicon Valley firms to build out military and national security applications. Karp called this move completely unhinged.
Are we really going to trust the defense of the country to a consensus view engineered by tech executives in California? It makes no sense. The government is attempting to build national security infrastructure on top of models that are built by organizations openly terrified of their own creations. Tech labs constantly warn the public about how dangerous their models are, yet they simultaneously try to sell those same models to the Department of Defense.
This dynamic is creating massive friction. The Pentagon designated Anthropic as a supply chain risk after the company refused to remove safeguards that prevented its models from being used for fully autonomous systems or domestic surveillance. You cannot build a sovereign defense system using a tech stack controlled by a commercial entity that can pull the plug, change the terms, or modify the model's weights overnight because of a shift in corporate ideology.
Meanwhile, foreign adversaries aren't dealing with these corporate identity crises. The speed of AI progress in China is real, aggressive, and entirely focused on state capability. While American institutions are wasting time playing with commercial tokens, geopolitical rivals are building sovereign, locked-down code stacks.
Taking Back the Means of Production
The timing of Karp's public outburst wasn't an accident. It was a carefully orchestrated corporate pivot. Right before he went on air to torch the industry, Palantir announced a massive expansion of its partnership with Nvidia, focusing on open-weight models like Nemotron. They also published a nine-point AI sovereignty manifesto.
This gives us a clear look at where the enterprise market is moving. The future isn't closed-source APIs. The future is complete control over your own technical infrastructure.
If you want to protect your business or your institution, you have to move away from the commercial token pipeline. Here is the playbook for fixing your AI strategy right now.
Stop sending your core data to external APIs. If you don't own the weights of the model you are using, and if you can't run that model on your own dedicated compute infrastructure, you are bleeding equity. Look toward open-weight models that you can host internally or within a secure cloud perimeter.
Implement an independent governance layer. You need an agnostic software system that sits squarely between your raw databases and whatever AI model you happen to be using. This layer must strictly dictate what data the model is allowed to see, what actions it can take, and ensure that no prompts or outputs are cached or used by the vendor for retraining.
Shift your metrics from consumption to value. Stop measuring your AI success by how many employees are using a tool or how many tokens you consume. If a vendor cannot tie their billing directly to concrete business outcomes—like reduced customer churn, faster software deployment, or actual manual hours saved—cut the budget.
The era of blindly buying into the Silicon Valley hype cycle is over. If you keep renting someone else's intelligence, you will end up paying a permanent tax on your own data. It's time to own the means of production.
