Google plans split from Scale AI over Meta deal

If your biggest competitor bought half of your chief supplier and hired its CEO to run their own AI division, you'd pull the plug immediately too.

That is precisely what Google is doing. According to a Reuters report citing five sources familiar with the matter, Alphabet's search giant plans to cut ties with AI data-labeling powerhouse Scale AI. The dramatic shift comes right after Meta announced a deal to acquire a 49% stake in Scale AI at a $29 billion valuation, doubling its previous $14 billion mark. Adding fuel to the fire, Scale's CEO Alexandr Wang and select staff are joining Meta to spearhead its internal AI efforts.

The biggest customer heads for the exit

Losing Google is not a minor scrape for Scale AI; it is an existential blow to its commercial engine. Scale brought in $870 million in revenue in 2024, with Google contributing roughly $150 million of that total. In 2025, Google had planned to spend roughly $200 million on Scale's network of specialized human annotators to train frontier models like Gemini.

Post-training advanced models requires PhDs, historians, and scientists to generate complex datasets, with single annotations sometimes running up to $100. To do that work, AI labs share proprietary datasets, unreleased models, and developmental roadmaps with the vendor. With Meta owning 49% of Scale and Wang running Meta AI, competitors fear their intellectual property and strategic blueprints will leak straight to their fiercest rival.

Google is not alone in heading for the doors. Microsoft is backing away, Elon Musk's xAI is actively seeking an exit, and OpenAI began winding down its spend months ago. While OpenAI's CFO noted on Friday that Scale remains one of its multiple data vendors, the exodus is unmistakable.

A feeding frenzy for rival labelers

Scale's competitors are already cashing in. Turing CEO Jonathan Siddharth remarked that neutrality is no longer optional for frontier AI labs; it is essential. Manu Sharma, CEO of Labelbox, told Reuters his company expects hundreds of millions in fresh revenue by year-end from fleeing clients. Meanwhile, Handshake, which specializes in vetted PhD networks, said its demand tripled overnight, and Mercor is helping labs recruit automated, in-house labeling teams to maintain total operational security.

Scale insists its enterprise and government contracts remain solid and customer data will stay protected. But generative AI frontier labs were the company's real cash engine, and Google's contract structure means work can migrate to alternative vendors within weeks.

Why it matters

For developers and enterprise users, the immediate consequence is friction. Migrating mission-critical post-training pipelines to alternative vendors or standing up in-house annotation teams takes time, capital, and fresh security audits. In the short term, this could slightly bottleneck the release cadence of specialized fine-tuning for frontier models outside the Meta ecosystem.

On the industry level, this move marks the end of the neutral arms-dealer era in AI data. Scale AI's meteoric rise was fueled by the fact that nobody saw it as a threat; it sold shovels to everyone digging for gold. By aligning so closely with Meta and handing over its visionary founder, that sacred neutral ground evaporated in an afternoon.

In our view, Meta's aggressive power play is a classic double-edged sword. Zuck secured world-class operational talent and direct leverage over a premier data infrastructure player, which helps combat the narrative that Meta is lagging behind. But by triggering an immediate client revolt, Meta effectively paid top dollar to hollow out Scale's most lucrative revenue stream. The true winners of this fallout appear to be rival platforms and the growing push toward fully in-house, proprietary data operations.