Key Takeaways

  • Data brokers are paying between $250,000 and $500,000 for complete operational datasets from companies, alongside $50,000 to $100,000 finder fees just for introductions.
  • By 2027, enterprise buyers will rebel against sending proprietary records to Anthropic and OpenAI without receiving tailored model improvements in return.
  • Applied AI companies like legal tech startup Harvey build proprietary internal models partly as pricing leverage against foundation model providers.
  • Reinforcement learning environments function as workflow simulations that replicate how knowledge workers complete tasks step by step.

The Sovereignty Rebellion Coming in 2027

Most software companies treat frontier model APIs like utilities. You send an API call, you pay by the token, you get an answer back, and you move on. Minerva co-founder Daniel Saeedi argues that this honeymoon will end soon. Leaders like Palantir CEO Alex Karp and Microsoft CEO Satya Nadella are already pointing out the flaw in this arrangement: companies give away their best operational context while the underlying model learns nothing specific to their business.

“Number two, I think, is the sovereignty and control issue, and you see Karp and Satya talking a lot about this,” (play from 13:43) Saeedi said. “I imagine this is going to be a huge issue in '27 as companies realize they're sending their data out to Anthropic and OpenAI and they're not getting anything in return for it.” (play from 13:48)

Right now, you pay for inference and supply the prompt context that makes the answer valuable. The frontier model does not retain your edge cases or specialize in your domain. As Saeedi put it: “The difference that I'm highlighting though that Karp talks a lot about is you give all this data, you get your answers back. The model doesn't get any better based on your use case.” (play from 14:38)

Internal Models as Pricing Shields

When you depend on a single frontier provider for your core product, that provider sets your margins. If you want lower unit costs, you need a credible alternative. Saeedi points to legal AI company Harvey, which developed its own internal model, Tenet. Even if a startup does not serve its homegrown model to every end customer, owning that model changes the vendor negotiation completely.

“And then I think third is the competition, right? And that you're no longer beholden to a single vendor,” (play from 14:57) Saeedi said. “If even in the case where you're Harvey, and you might not even be using tenet as a model to your customers, but the fact that you can go to Anthropic and OpenAI and say, 'Hey, I have this homegrown model. You have to now engage in value-based pricing for me because it's no longer no automation or automation.'” (play from 15:03)

This dynamic is creating an aggressive race to collect and structure private operational data. Data brokers now scour industries for workflow logs, paying huge sums to buy complete business records. “They're spending anywhere from $250k to $500k per complete dataset from a company, and they were paying people $50,000 to $100,000 just for referring the companies to them,” (play from 27:27) Saeedi noted.

These brokers use that data to build reinforcement learning environments. As Saeedi explained, “An RL environment is simply you can think of it as like a simulation where you're trying to replicate the environment in which an AI is actually doing a task.” (play from 25:29) If you capture that environment yourself, your company keeps both the leverage and the margin.

What to Do With This

Audit your primary workflow pipeline by tomorrow afternoon. Export two weeks of raw interaction logs, including tool calls, system prompts, and user corrections, then store them in your own cold storage bucket before stripping them into an internal evaluation set. If OpenAI or Anthropic raises rates tomorrow, you must have the ground-truth data required to benchmark an open-weight alternative within 48 hours.