Scale AI Hires a Cloud Operator to Become More Than a Data Supplier

The appointment of Francis deSouza signals Scale AI’s move from supplying training data toward selling complete enterprise AI systems.

Scale AI built its reputation behind the scenes. The company supplied the labeled data, human feedback and evaluations that helped leading artificial intelligence laboratories train and test their models.

Now it wants a larger piece of what happens after those models leave the laboratory.

Scale announced that Francis deSouza will become CEO on August 10. DeSouza most recently served as chief operating officer and president of security products at Google Cloud. He replaces Jason Droege, who has served as interim CEO since June 2025. See Scale AI’s announcement.

The appointment is more than an executive change. It reflects an industrywide movement from supplying AI components toward controlling the applications, services and customer relationships built around them.

Scale Is Moving Up the AI Supply Chain

Scale’s original business helped AI developers prepare data for model training. That work remains valuable, particularly as frontier laboratories require increasingly specialized human feedback, testing and evaluation.

The company now describes itself as both an AI applications business and a data-infrastructure company. Its products are intended not only for model developers but also for corporations and governments deploying AI inside real operations. Scale recently added BP and Mayo Clinic as customers and expanded its government work.

DeSouza’s background fits that transition. Google Cloud does not simply sell computing capacity. It sells platforms, security, consulting, integrations and long-term enterprise relationships. Earlier in his career, deSouza led Illumina and held senior positions at cybersecurity company Symantec.

Scale is effectively hiring an operator who understands how to move from technical capability to large, durable customer accounts.

As Reuters reported, the company intends to expand both its data and applications businesses. DeSouza said his focus will include getting Scale’s solutions into more businesses and governments while demonstrating value through measurable outcomes.

Every Infrastructure Company Wants the Application Layer

Selling infrastructure can create a strong business, but application providers usually control the customer relationship.

A data supplier may earn money when a model is trained. An application provider can earn recurring revenue every time employees use the resulting system. It can also sell implementation, security, customization, maintenance and support.

This is why the AI market is beginning to converge.

Cloud companies are building models. Model companies are building agents. Data companies are building applications. Consulting firms are packaging repeatable software. Software companies are adding their own AI platforms.

Everyone wants to own more of the stack.

For customers, the result may be convenient. A vendor that supplies data preparation, models, applications and implementation can reduce the number of separate contracts and technical handoffs.

It can also create dependence.

The Convenient Platform Can Become the Permanent Platform

A company may initially hire an AI vendor for a limited project, such as document classification or customer-support analysis. The vendor then offers a knowledge platform, agent builder, evaluation system, workflow tools and managed infrastructure.

Before long, the company’s prompts, tests, business rules, integrations and operational knowledge may all depend on one provider.

Changing vendors then becomes more complicated than replacing software. Different AI models interpret instructions differently. Evaluation tools may use proprietary formats. Workflow logic may depend on vendor-specific APIs. Employees may have adapted their processes around the platform’s behavior.

The problem is not that bundled platforms are inherently bad. The problem is entering one without an exit plan.

Businesses should understand how AI vendor lock-in now includes model behavior, not merely contracts and data formats. Prompts, evaluation questions, knowledge sources and business rules should remain under the customer’s control wherever practical.

Buyers Should Separate the Layers

Companies evaluating enterprise AI vendors should divide the proposed system into distinct parts:

• The model generates or interprets information.
• The data layer supplies company records and reference material.
• The knowledge layer contains approved policies, procedures and operational guidance.
• The workflow layer initiates tasks or updates systems.
• The governance layer defines permissions, ownership, testing and review.

A single vendor may provide several of these layers, but the business should still understand where each one begins and ends.

It should also determine which information must remain in its existing systems. Customer transactions belong in the CRM. Employee records belong in the HR platform. Financial transactions belong in accounting software. Approved procedures and decision guidance may belong in a governed knowledge hub.

The AskMaisy guide explaining what belongs in an AI knowledge hub and what should remain in operational systems describes this separation as a way to preserve authoritative systems rather than creating uncontrolled copies.

Contracts Must Address Portability Before Deployment

Before adopting an integrated AI platform, buyers should ask practical questions.

• Can company data be exported in a usable format?
• Who owns prompts, evaluations and custom workflows?
• Can another model be substituted without rebuilding the application?
• What happens to indexed company content when the contract ends?
• Which integrations rely on proprietary connectors?
• How are security logs and testing records retained?
• Can the customer operate the system if the vendor changes pricing, products or policies?

These questions are easier to answer before the system becomes operational. Once employees depend on it, every migration becomes more expensive and politically difficult.

Scale’s Appointment Shows Where the Market Is Going

Scale AI is not abandoning its data business. It is using that position as the foundation for a larger enterprise offering.

The strategy makes sense. The company already understands models, evaluations and specialized data. Moving into applications allows it to capture more revenue and participate directly in the business outcomes customers care about.

For buyers, however, the lesson is broader.

The AI vendor that begins as a supplier may eventually want to become the platform underneath the entire company. That can produce a stronger and more integrated system. It can also turn a successful pilot into a single point of failure.

Choose vendors for their capabilities. Design the architecture so none of them owns the keys to the building.