Karp is right. Just not about you.
What last week's Palantir moment actually means if you run a $20M to $200M company.
Last week, Palantir CEO Alex Karp went on CNBC and delivered twenty minutes of television that the internet is still arguing about. The internet mostly focused on the spectacle. We think that's a mistake, because buried in the spectacle was the most important argument in enterprise AI right now.
Here's the short version. Palantir announced a deal with Nvidia to run open models inside secure environments where customers control everything. The next day, Palantir published a nine-point manifesto on "AI sovereignty." The day after that, Karp went on Squawk Box and put it bluntly: enterprises are tired of paying for tokens while handing their data, and potentially their edge, to the model companies. His line: technical customers "want to know they own the means of production."
A defense contractor said that on financial television because the demand is no longer theoretical. Companies want the benefit of better models without surrendering control of the systems around them.
Where Karp is right
For his customers, he's right across the board. If you're a defense agency, a critical infrastructure operator, a life sciences company, or a SaaS business whose product is basically software plus proprietary data, then the frontier labs are a genuine strategic concern. Your data is interesting to them. Your business is adjacent to theirs. Control over your models and weights is a rational demand, and the Palantir-Nvidia offering exists because that demand is real.
He's also right about something bigger: intelligence is becoming a commodity. Karp calls it "commodity cognition." Every quarter, the models get better and cheaper, and the differences between them shrink. When the input is a commodity, the advantage moves to whoever owns the layer where the commodity gets put to work.
Where he's not talking about you
Here's the trap we see mid-market CEOs fall into: overestimating how interesting their data is to a trillion-dollar lab. If you run a $60M distribution business, Anthropic is not coming for your alpha. Neither is OpenAI. Your customer list, your pricing history, your ERP data: valuable to you, invisible to them. Building your AI strategy around fear of theft is solving a problem you don't have, usually at the cost of the problem you do have.
Because you do have one, and it's worse than theft. It's sameness.
If you and your three closest competitors all outsource your thinking to the same model, with the same generic prompts and no proprietary context, you will get the same answers. Same analysis, same copy, same ops playbook. Your margin converges along with your output. The commodification of intelligence doesn't steal your edge. It quietly dissolves it, and everyone else's, into the average.
The test that actually matters
Satya Nadella put his finger on it in June: the real test of control in this era is whether you can swap out the generalist model without losing the company expertise built into your systems.
Run that test on your own business. If your workflows, your SOPs, your institutional knowledge, and your data all live inside rented SaaS and a model vendor's proprietary features, you fail. Switch vendors and your accumulated intelligence stays behind. You're a tenant, and the landlord keeps the improvements.
If those things live in systems you own, you pass. You can change models every quarter, always renting the best intelligence at the best price, and everything that makes your company yours comes along.
Which gets us to the one-sentence version of our whole position:
You don't need to own the model. You need to own everything around the model.
The model is a rented turbine. The plumbing it powers, the data it draws on, the encoded know-how that tells it what good looks like in your business: that's the house. Hold title to the house.
What this means for the middle market
For a $20M to $200M company, sovereignty does not require owning the model. It means owning the systems where your workflows, data, and operating knowledge accumulate, then treating the model as a replaceable input.
That distinction is the starting point for the Runpoint master plan. The plan begins with rented software you can price, moves into the knowledge and agents that make your company different, and ends with the products and services that were not economical before AI. This note is about why the ownership line belongs around those things, not around the model itself.
What to do Monday
One exercise, thirty minutes, no consultants required. Two columns.
Column one: everything your business runs on that you rent. The SaaS, the workflows trapped inside it, the data sitting in someone else's schema.
Column two: everything you know that your competitors don't. The SOPs, the pricing instincts, the way your best ops person handles an exception.
Column one is your exposure. Column two is your edge, currently unencoded and walking out the door every time someone quits. The gap between the two columns is your actual AI strategy.