Since June this year, despite Meta's investment of up to $1.43 billion in the data annotation company Scale AI and hiring executives such as its CEO Alexandr Wang to join Meta Superintelligence Labs (MSL), the relationship between the two sides seems to be facing challenges.
The core issue stems from disputes over data quality. Despite the substantial investment, sources indicate that researchers in Meta's core AI division, TBD Labs, generally consider Scale AI's data quality to be poor and prefer to collaborate with Scale AI's main competitors, Mercor and Surge. TBD Labs had already been working with these two companies before its establishment, but it is rare for a company to continue relying on its competitors after investing such a large sum in a data provider.
This phenomenon reflects changes in the data annotation industry. In the early days, Scale AI's crowdsourcing model relied on low-cost labor to handle simple tasks, but as AI models became more complex, high-skilled professionals such as doctors and lawyers are needed to provide high-quality data. Although Scale AI launched the Outlier platform, competitors like Mercor and Surge, which were built from the beginning on a high-salary talent model, are rapidly rising.
Beyond the tension in commercial cooperation, the integration of personnel between the two sides has also encountered setbacks. Ruben Mayer, a former executive of Scale AI, left Meta only two months after joining, further drawing external attention. Although Mayer stated that he left due to "personal reasons" and was satisfied with his work experience at Meta, there is a discrepancy between his understanding of his responsibilities at Meta and the accounts of internal sources.
Additionally, Meta's AI department is also facing talent attrition. Former and current employees have revealed that since the introduction of Alexandr Wang and a group of talents from companies such as OpenAI, the AI department at Meta has become chaotic. Newcomers have become frustrated with the bureaucratic culture of large companies, while the original GenAI team at Meta feels their capabilities are constrained. Several key employees, including MSL AI researcher Rishabh Agarwal, have recently left, casting a shadow over the future development of Meta's AI.
This investment is generally seen as an emergency move by Meta CEO Zuckerberg after the underwhelming release of Llama4, aimed at catching up with OpenAI and Google. He not only recruited Alexandr Wang but also actively hired top talents from companies such as OpenAI and Google DeepMind. However, given the rapid departure of some new employees and the internal chaos within the teams, whether Meta can stabilize its AI operations and retain talent remains an unresolved question.
