Data Services vs Data Software: Two Different Businesses

There is a persistent confusion in the physical-AI ecosystem about what kind of business a data-infrastructure company actually is. The confusion matters because the two possible answers imply very different growth curves, gross margins, and long-term equity value.

If the primary product is 'we will produce N hours of demonstration data for you, delivered to your training pipeline,' the company is a services business. It scales by adding operators, cells, or capture kits. Gross margins are typical of services, thirty to fifty percent at scale, bounded by operator labor. Revenue is predictable per unit. The business is real, the customers are real, and the enterprise value is a multiple of gross profit, not of revenue.

There is nothing wrong with this. Services businesses in strategic categories can be very large. But they are not software businesses and it is a mistake to price them as such.

If the primary product is 'we will sell you the capture platform and the operator-network software so you can produce your own data,' the company is a software business. Gross margins are software-typical, seventy to eighty-five percent. Growth is driven by seat expansion and platform adoption, not headcount. Enterprise value is a multiple of ARR.

This model is harder to sell into a frontier lab that would rather buy the outcome than run the operation themselves. It is easier to sell into the second tier, enterprises, industrial integrators, defense, that want capability but not a services relationship.

We sell both. The hardware and the platform are a software business. The managed operator network via Talika is a services business layered on top for customers that want the outcome, not the platform. This is deliberate, and it is the same shape most durable data-infrastructure companies land on eventually. The mistake to avoid is pretending one is the other.