Since June, when Meta made a $14.3 billion investment in the data-labeling company Scale AI, bringing CEO Alexandr Wang and many of the startup’s leading executives on board to operate the Meta Superintelligence Labs (MSL), tensions between the two organizations are already becoming apparent.
At least one executive Wang recruited for the management of MSL – Ruben Mayer, former Senior Vice President of GenAI Product and Operations at Scale AI, has left Meta just two months after joining, according to two sources familiar with the situation who spoke to TechCrunch.
Mayer had a total of roughly five years of experience at Scale AI over two separate periods. In his brief time at Meta, he reportedly oversaw teams focused on AI data operations but wasn’t involved with TBD Labs. The primary unit at Meta focused on developing AI superintelligence, which has attracted top researchers from OpenAI.
However, Mayer challenges some aspects of this account, telling TechCrunch that his initial role was to assist in setting up the lab as needed rather than handling data, and he claimed to have been “part of TBD Labs from day one” instead of being kept out of the main AI group. He also noted that he “did not report directly to [Wang]” and expressed being “very happy” with his experience at Meta, citing a “personal matter” as his reason for leaving.
In addition to staff changes, the partnership between Meta and Scale AI seems to be evolving. According to five sources familiar with the matter, TBD Labs is collaborating with other external data labeling vendors beyond Scale AI to develop its forthcoming AI models. These alternative vendors include Mercor and Surge, two of Scale AI’s biggest competitors.
While it is customary for AI labs to collaborate with multiple data labeling providers. Meta has been working with Mercor and Surge since prior to the establishment of TBD Labs, it is unusual for an AI lab to invest so heavily in a single data vendor. This makes the current situation particularly noteworthy: despite Meta’s substantial investment, several sources indicate that researchers in TBD Labs consider Scale AI’s data to be of low quality and prefer to utilize Surge and Mercor.
Scale AI originally developed its business around a crowdsourcing method that employed a large, cost-effective workforce to handle straightforward data labeling, a process involving the tagging and annotation of raw data to train AI models. However, as AI models have become more advanced, they now depend on highly skilled professionals, such as doctors, lawyers, and scientists, to create and refine the high-quality data necessary for enhancing their performance.
While Scale AI has attempted to attract subject matter experts with its Outlier platform, competitors like Surge and Mercor have experienced rapid growth due to their business models being established with high-paid talent from the very beginning.
A representative from Meta contested the notion that Scale AI’s product has quality issues. Surge and Mercor chose not to provide a comment. In response to inquiries about Meta’s increasing dependence on rival data suppliers, a spokesperson for Scale AI referred TechCrunch to its original announcement regarding Meta’s investment in the startup, which mentions the expansion of their commercial partnership.
Meta’s agreements with third-party data vendors suggest that the company is diversifying its sources instead of relying solely on Scale AI, even after investing significant sums in the company. However, Scale AI does not enjoy the same luxury. Shortly after Meta revealed its substantial investment in Scale AI, both OpenAI and Google announced they would discontinue their collaboration with the data provider.
Following the loss of those clients, Scale AI cut 200 jobs in its data labeling division in July, with new CEO Jason Droege attributing the layoffs partially to “shifts in market demand.” Droege indicated that Scale AI plans to expand in other sectors, including government sales — recently securing a $99 million contract with the U.S. Army.
Initially, some speculated that Meta’s investment in Scale AI aimed to attract Wang, a founder who has been a part of the AI industry since Scale AI’s inception in 2016. He seems to be aiding Meta in recruiting leading AI talent.
Aside from Wang, there remains uncertainty concerning the actual value of Scale AI to Meta.
A current employee at MSL mentioned that several executives from Scale AI who joined Meta are not engaged with the core TBD Labs team.
At the same time, Meta’s AI division has reportedly become increasingly disordered since Wang and a number of top researchers joined, as noted by two former employees and one current MSL employee. New hires from OpenAI and Scale AI have voiced dissatisfaction with maneuvering through the bureaucracy of a large corporation, while the previous GenAI team at Meta has reportedly seen its responsibilities reduced, according to their accounts.
These tensions suggest that Meta’s most significant AI investment thus far may be experiencing a difficult beginning, despite its intended goal of resolving the company’s AI development obstacles. Following the unimpressive rollout of Llama 4 in April, Meta CEO Mark Zuckerberg became annoyed with the AI division, according to insights from one current and one former employee speaking to TechCrunch.
In a bid to rectify the situation and catch up with OpenAI and Google, Zuckerberg hurried to finalize agreements and launched a proactive effort to recruit top-tier AI talent.
Aside from Wang, Zuckerberg has successfully attracted prominent AI researchers from OpenAI, Google DeepMind, and Anthropic. Meta has also taken over AI voice startups like Play AI and WaveForms AI, and announced a collaboration with AI image generation startup, Midjourney.
To support its ambitions in AI, Meta recently unveiled plans for several large data center expansions throughout the United States. One of these significant projects is a $50 billion data center in Louisiana, named Hyperion, after a titan from Greek mythology known to have fathered the God of Sun.
Wang, who does not have an academic background in AI research, was considered a somewhat unexpected choice to head an AI lab. Reports indicate that Zuckerberg had discussions with more conventional candidates to lead the initiative, such as OpenAI’s chief research officer, Mark Chen, and made attempts to acquire the startups of Ilya Sutskever and Mira Murati, all of whom opted not to join.
Several of the new AI researchers recently recruited from OpenAI have already exited Meta, as noted by Wired. In addition, many long-serving members of Meta’s GenAI division have left due to the recent changes.
MSL AI researcher Rishabh Agarwal is among the most recent to announce his departure, sharing on X this week that he would be leaving the company.
“The pitch from Mark and @alexandr_wang to build in the Superintelligence team was incredibly compelling,” Agarwal stated. “But I ultimately choose to follow Mark’s own advice: ‘In a world that’s changing so fast, the biggest risk you can take is not taking any risk’.”
When asked about his experiences at Meta and what influenced his decision to leave, Agarwal opted not to share any comments.
Chaya Nayak, the director of product management for generative AI, and research engineer Rohan Varma have also stated their intention to leave Meta in the past few weeks. The pressing question now is whether Meta can stabilize its AI initiatives and keep the talent necessary for its future growth.
MSL has already begun development on its upcoming AI model. According to Business Insider reports, the goal is to launch it by year-end.
Read also: How AI is Powering Better Communication with Vocal Image
