Neda Nia drives Stibo Systems' product vision, shaping strategy, innovation, and growth to create measurable value for customers.
getty
Palantir CEO Alex Karp recently said what enterprise software rarely says about itself: Something has "gone completely wrong" in how companies are handing their most sensitive data to cutting-edge frontier AI labs.
If you ignore the theatrics of his argument, it boils down to one word that is worth paying close attention to: alpha. No, not the finance term. He meant the thing that makes a company a company: the long-term, hard-won understanding of how it actually operates, including trade secrets and customer relationships.
Some have argued that Karp's warning was self-interested. Maybe. But the discomfort he talked so publicly about is real, and I don't think it's going away any time soon. In a podcast interview, venture capitalist David Sacks agreed with him, but he framed the idea more practically: Who controls your compute, your models and your data stack?
I want to add a question: Who controls the context and the accumulated meaning behind your data that makes any of it useful in the first place?
How The Bill Is Arriving Faster Than The Trust Is
Gartner estimated in 2025 that worldwide AI spending would reach nearly $1.5 trillion that year, up almost 50% from the year before.
At that mind-bending pace, every layer of the technology stack shifts at once. Considering this, Nvidia's framing of AI as a five-layer "cake” or system (energy, chips, infrastructure, models and applications) captures something important: Value is not staying put in one layer. It's already sliding up from the model itself toward whatever sits closest to a company's actual data and decisions.
That application layer is precisely the layer enterprises are most protective of, and for good reason. If a frontier lab absorbs a company's proprietary workflows through its prompts and usage patterns, that "alpha" doesn't stay put either. It can, in principle, get folded into the next model update and sold back to a competitor.
How Fast This Trust Can Break
HubSpot recently announced its customers would be automatically enrolled in a data-sharing arrangement that pooled their CRM contact and company data across accounts to improve everyone's AI-powered prospecting tools.
As reported by CX Today, the backlash on LinkedIn was immediate. Within four days, HubSpot said they made a mistake and did away with the policy.
This has nothing to do with the particular company, but consider the implications. A vendor wanted to use customer data as a shared resource to fuel AI features. Customers saw it as a vendor taking their competitive advantage.
The trust broke almost instantly. Quite a lesson for us all.
To Build Or To Borrow?
A question I hear constantly from data and technology leaders is whether their own company should be building a proprietary foundation model rather than relying on a frontier lab.
For nearly all enterprises, the honest answer is no, not yet and possibly not ever. Building a competitive foundation model requires a scale of data and compute that even large enterprises with billions of records rarely have.
On top of that, most companies' own data-sharing agreements with their customers and partners were never written with model-building rights in mind.
To me, the more thought-provoking question is where an enterprise's real advantage sits once models themselves become table stakes. Increasingly, I think that advantage sits one layer down: in how much you can trust the enterprise's data and the context it gives you. That's a much harder thing to build than API access to a model.
It’s also a much harder thing to lose overnight.
Trusted Data As The New Battleground
Karp's comments and the pushback against companies using proprietary data are really the same story from two angles. Trust in how AI touches your data is becoming the actual competitive battleground.
Here's my hypothesis: More enterprises will run AI inference in environments they control, on open models they can inspect, enriched with their own proprietary data. Fewer will default to routing sensitive work through those frontier labs.
If that happens, the model stops being the differentiator. What matters instead is whether a company has spent years building a trusted, governed record of its own data. Trusted data, and the trustworthy intelligence it enables. That kind of record can't be copied overnight, because it was never generic to begin with.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
>> Home