Meta Platforms → Scale AI

Meta bought a 49% non-voting stake in the data-labelling and model-evaluation company at a $29bn valuation and hired its founder, Alexandr Wang, to run a new superintelligence unit — an acquisition in effect, structured to avoid being one in law.

Meta bought a 49% non-voting stake at a $29bn valuation and hired founder Alexandr Wang to run its superintelligence unit.

Value

$14.3B

Announced

June 2025

Completed

June 2025

Sector

AI

Deal type

asset purchase

Acquirer

Meta Platforms (United States)

Target

Scale AI (United States)

Why it mattered

Scale AI supplied the human-labelled training data and the evaluation benchmarks that the frontier labs, and the US Department of Defense, used to train and test their models. Meta, behind OpenAI and Google after a poorly received Llama 4 release, wanted two things Scale had: the data pipeline, and Wang, whom Mark Zuckerberg recruited personally along with a hiring campaign that offered nine-figure packages to researchers at rival labs.

The structure — a minority stake with no voting rights, a founder hired rather than a company acquired — was designed to keep the transaction outside the merger-control thresholds that had delayed Microsoft–Activision and killed Nvidia–Arm, and it followed the template Microsoft had set with Inflection and Google with Character.AI the year before. The price nonetheless valued Scale at more than double its previous round.

What it left behind

Scale's other customers — OpenAI and Google among them — reduced or ended their use of a supplier now half-owned by a competitor, and Scale cut staff within weeks. The deal confirmed that in frontier AI the acquirable asset is people, and that the market had found a structure for buying them: the licence-and-hire transaction that regulators in the US and the UK have since said they will examine as if it were a merger.