When Meta paid roughly $14.3 billion for a 49% stake in Scale AI in June 2025 and simultaneously pulled founder Alexandr Wang into a new role leading Meta's superintelligence research effort, it created an unusual corporate structure: a company whose largest customer relationships depend on labs that now compete directly with its largest shareholder's parent. Scale AI built its business supplying human-annotated and increasingly AI-assisted training data and evaluation services to OpenAI, Google, and other frontier labs, a position that made the Meta deal immediately awkward, since several of those customers began pulling work from Scale within months of the announcement over concerns about data confidentiality flowing to a competitor. Scale's post-Wang leadership, with Jason Droege installed as chief executive, has spent the period since trying to reassure remaining and prospective customers that a firewall exists between Scale's commercial operations and Meta's research organization, while also diversifying the business away from reliance on a handful of frontier-lab contracts. That diversification has pushed Scale further into government and defense work, an area where the company had already built relationships through its Scale for Government unit, and into evaluation and red-teaming services for enterprise customers deploying third-party models rather than training their own. The competitive landscape for data labeling and model evaluation has also shifted underneath Scale. Surge AI, a lower-profile rival that reportedly turned a profit while staying private, absorbed some of the frontier-lab business that fled Scale after the Meta deal, and smaller specialized shops focused on domain-specific data, like legal, medical, and coding annotation, have chipped away at the generalist labeling market Scale once dominated. The broader trend across the industry is toward synthetic data generation and AI-assisted labeling reducing the need for large human annotation workforces, a shift that threatens the core unit economics of Scale's original business model regardless of the Meta relationship. Financially, the Meta capital gave Scale a war chest most competitors lack, and the company has used some of it to acquire smaller evaluation and infrastructure startups to build out an integrated data-plus-evaluation platform rather than a pure labeling shop. Whether that platform strategy works depends on convincing customers, many of whom are direct Meta rivals, that Scale's data pipeline is genuinely walled off from Meta's model training, a trust problem no amount of technical firewalling fully solves given the equity relationship. What to watch: whether major frontier labs beyond Meta restore or expand their Scale contracts through 2026, whether Scale's government and defense revenue grows fast enough to offset lost commercial lab business, and whether Wang's continued informal ties to Scale's board create fresh conflict-of-interest questions as Meta's own model programs mature.
