Scale AI interim CEO Jason Droege: two $100M+ businesses, monthly growth since Meta deal, and the trough of AI disillusionment in enterprises

Summary

Jason Droege spent years building growth-stage businesses, most notably founding Uber Eats, which reached roughly $20 billion in GMV by the time he left. He joined Scale AI a year ago as chief strategy officer, drawn by what he describes as the same pattern he'd seen before: demand massively outstripping supply, with a business that needed to be organized to capture it. He became interim CEO after Meta invested over $14 billion for a 49% stake, with founder Alex Wang moving to Meta as chief AI officer.

Scale today has more than 1,100 full-time employees and $1 billion on the balance sheet, alongside a contributor network of hundreds of thousands of people who supply training data. Droege says roughly 15% of contributors hold PhDs, 25% have master's degrees, and around 60% have at least a bachelor's degree, reflecting the shift from general annotation work toward what he calls "skill capture" — getting domain experts to model how senior professionals actually reason.

Two $100M+ businesses

Droege argues most outside observers underestimate Scale's scope. The company runs two distinct businesses, each generating over $100 million in revenue. The data business — supplying training data to frontier model builders — is what Scale is publicly known for. The applications and services business, which helps large organizations actually deploy AI, is less understood but equally large. Droege frames it as independently unicorn-scale: Fortune 500 companies, the US government, and international governments like Qatar are all customers. A recently signed $100 million contract with the US Army is one concrete data point.

Monthly growth since the Meta deal

Droege says Scale has grown every month since the Meta deal closed, which he expects will surprise people given the press coverage. He attributes this to the company's sustained ability to supply difficult, high-volume data at the frontier — something he argues very few providers can actually do at scale, despite widespread claims in the market.

The enterprise trough

Droege is direct that AI is overhyped for near-term enterprise job displacement. The harder and more commercially relevant insight is what he's seeing on the ground with large customers: they went into AI expecting models to "pop out of the box," and instead found that making AI work inside complex organizations requires significant data work, change management, and senior-level human involvement that nobody priced in.

The specific friction he describes is alignment — not in the abstract safety sense, but practically: when you build an agent system for an accounting firm or a hospital, you have to decide whose judgment to encode. A junior accountant's? The CFO's? Compliance's? That question forces the most senior people in an organization to get personally involved in labeling and training work they assumed would be delegated downward. That is where Scale's enterprise business is finding traction.

Droege believes enterprises are approaching a trough of disillusionment. Budget allocated to broad "AI initiatives" will increasingly face ROI scrutiny, and companies that can't demonstrate concrete value will lose contracts. His framing is blunt: the Grim Reaper is coming for AI vendors that aren't delivering ground truth to customers. Scale's pitch is that it sits on the right side of that reckoning — helping organizations extract value rather than just selling the promise of it.

Robotics data and what's next

Scale does robotics data today, though Droege describes it as the newest part of the business and acknowledges the category is early — robots aren't yet walking around the world at volume. The absence of a large pre-existing corpus, unlike the internet-derived data that fueled LLM training, likely increases the long-term opportunity rather than diminishing it. All prior data types — computer vision, autonomous vehicle data, DoD contracts — continue to run in parallel, stacking rather than replacing each other.