Key Takeaways

  • Open-source AI models raise the general capability floor but don't cannibalize the value of specialized, frontier data for companies like Mercor.
  • The real opportunity lies in identifying and serving "latent demand"—problems enterprises aren't even trying to solve yet with AI—which requires high-value data at the frontier of model performance.
  • Enterprises differentiate data sensitivity: they're more open to sharing data for general workflows (HR, procurement) with proprietary models, but guard core business data that differentiates them.
  • Skepticism around sharing sensitive, core business data with large frontier model providers is a major factor driving enterprise decisions on AI deployment.

The Open-Source Paradox: Raising the Floor, Not Killing the Ceiling

Harry Stebbings, host of 20VC, cut straight to the chase with Oswald Nitski, CPO at Mercor: “I am seeing open open open. Everyone claiming that we will see the mass migration from frontier closed to open... Does open cannibalize Mcore's core business?” It's a question many data-centric AI companies are wrestling with. Nitski's answer isn't what you might expect if you’re only tracking model performance.

Nitski argues that the rise of open-source models isn't a threat to Mercor's core business of providing high-value data. Instead, he sees it as a market expander. “I wouldn't say that open source model improvements cannibalize our core business because data is most valuable on the frontier of model performance,” Nitski explained. “Open models just raise the floor of what people are interested in.” This means more companies are now interested in any AI, creating a bigger market for all AI solutions.

The real differentiator, Nitski points out, isn't just better-performing models on existing problems, but uncovering entirely new use cases. “We think that these calculations might be based off of existing demand... but there's a whole category of latent demand that people aren't even, these are things that people aren't even trying to do with models yet,” Nitski said. This "latent demand" is where specialized, frontier data becomes critical. It's about finding the problems that companies didn't even know AI could solve, and then building the data sets to train models for those solutions. Think of it less like competing on an existing highway and more like building new roads to untapped territories.

Enterprise Data: Fear and the Frontier

Another significant tension Nitski highlighted is the deep-seated skepticism enterprises have about sharing sensitive data. Stebbings referenced Alex Karp of Palantir, noting his observation of “incredible skepticism we see from large enterprises towards data and sharing data with the frontier model providers.” This isn't just abstract fear; it shapes how enterprises adopt AI.

Nitski confirms this caution, drawing a clear line between different types of data. “Things that are just like general things that every company needs to do like HR, procurement, it can be less sensitive and enterprises are more open to putting these workflows on proprietary models,” he explained. For these general, less sensitive tasks, the efficiency gains outweigh the data sharing risk. However, it's a different story for core operations. “It's the core work that the company is doing that's that's vital to its business that differentiate it from competitors where we see more sensitivity,” Nitski added.

This distinction is critical. Companies will self-host or use open models for generic tasks where data sensitivity is low. But for tasks that involve their unique competitive advantage and proprietary information, they demand specialized, secure solutions. This is where companies that can provide frontier data expertise with strong trust and security paradigms win, even against the allure of open-source flexibility.

What to Do With This

Stop worrying that open-source AI devalues your data play. Instead, leverage the increased market interest that open models create. For your next product iteration, don't just optimize for existing demand; explicitly seek out "latent demand" by asking customers what impossible tasks they wish AI could do. Build your data strategy around these high-value, previously unimaginable applications, focusing on the specialized data needed to unlock them, and tailor your enterprise pitches by clearly segmenting for data sensitivity: start with lower-risk, general workflows to build trust, then demonstrate how your secure, specialized approach can tackle their most vital, proprietary problems.