Scale AI Raises $1.5B Series G at $25B Valuation
Scale AI has closed a $1.5 billion Series G led by Accel and Google, valuing the data-labeling and AI infrastructure company at $25 billion. The round will fund expansion of its government contracts division and enterprise model evaluation platform.
Original sourceScale AI has raised $1.5 billion in a Series G financing round led by Accel and Google, pushing the company's valuation to $25 billion. The San Francisco-based company, founded by Alexandr Wang in 2016, built its initial business on human-assisted data labeling for machine learning pipelines and has since repositioned itself as a broader AI infrastructure and evaluation platform serving both enterprise and government clients.
The funding comes at a moment when demand for high-quality training data and model evaluation tooling has intensified across the industry. As frontier labs race to improve model capabilities, the bottleneck has increasingly shifted toward data quality and systematic evaluation — two areas where Scale has invested heavily. The company's Donovan platform, aimed at U.S. defense and intelligence agencies, has become a significant revenue driver and is expected to receive a meaningful share of this new capital.
On the enterprise side, Scale is doubling down on its model evaluation infrastructure, which helps organizations benchmark and red-team large language models before deployment. This positions Scale less as a labeling vendor and more as a critical quality-assurance layer in the AI supply chain. Google's participation as a co-lead is notable given that it is simultaneously a Scale customer, a competitor in some AI services, and now a significant investor — a relationship that reflects the complex interdependencies defining the current AI infrastructure market.
The $25 billion valuation represents a substantial premium over previous rounds and reflects investor confidence that the data and evaluation layer of the AI stack will remain a durable business even as model training costs decline. Whether Scale can maintain pricing power as synthetic data generation matures and model providers build more evaluation tooling in-house remains the central question for the company's next chapter.
Panel Takes
The Founder
Business & Market
“The Google co-lead is the most interesting line in this deal — when your customer is also your investor and your partial competitor, you've either built something genuinely defensible or you've created a conflict that unwinds messily in three years. Scale's moat is the combination of proprietary human-feedback pipelines, government clearances that take years to replicate, and workflow lock-in at the enterprise evaluation layer — that's a real stack of defensibility, not just 'we shipped first.' The stress test is synthetic data: if frontier labs solve data quality generation internally, Scale's labeling revenue compresses fast, and a $25B valuation needs the government and evaluation businesses to carry significant weight on their own.”
The Skeptic
Reality Check
“$25 billion is a number that requires the government contracts division to be a real, recurring, expanding business — not a few pilot contracts dressed up for a fundraise deck. The scenario where this breaks is simple: synthetic data generation reaches sufficient quality that the labeling moat evaporates, the DOD consolidates AI vendors the way it always does, and Scale is left with an evaluation platform competing directly against the model providers who are its primary customers. The prediction: Google ships 80% of the enterprise eval tooling natively within 18 months, and Scale's ability to stay ahead of that depends entirely on whether Alexandr Wang can keep government revenue compounding faster than the enterprise side commoditizes.”
The Futurist
Big Picture
“The thesis Scale is betting on is specific and falsifiable: that as AI models proliferate, evaluation and alignment data become more valuable than raw compute, and that no single lab will trust a competitor to be the neutral arbiter of model quality — which means a independent infrastructure layer has to exist. The second-order effect that gets underreported here is geopolitical: Scale's government division is quietly becoming part of how the U.S. military thinks about AI deployment standards, which means Scale gains influence over procurement criteria that shape what 'enterprise-ready AI' means across the federal supply chain. That's a power position that compounds in ways a valuation multiple can't fully capture yet.”
The PM
Product Strategy
“Scale is running two very different product strategies under one roof — a government platform with long sales cycles and compliance requirements, and an enterprise eval product that needs to iterate fast against model-provider encroachment — and the funding announcement doesn't clarify whether those two jobs are converging or diverging. The job-to-be-done for enterprise is clear: 'help me know if this model is safe and capable enough to deploy before it embarrasses me in production.' The question is whether Scale's evaluation platform is opinionated enough to own that job completely, or whether it's a configurable toolkit that leaves the hard product decisions to the customer and therefore loses to whatever the model provider ships natively.”