August 31, 2012 - In June of this year Meta To data labeling service providers Scale AI The partnership between the two companies is already showing cracks after only a few months of a $14.3 billion investment (note: about 101.951 billion yuan at current exchange rates) that brought on board CEO Alexandr Wang and a number of the startup's executives to run Meta SuperSmart Labs (MSL).
Two people familiar with the matter have told TechCrunch that at least one of the executives Alexander Wang brought in to help run MSL has left -- Ruben Mayer, Scale AI's former senior vice president of generative AI products and operations, left after just two months at Meta. Ruben Mayer.
Meyer has worked for Scale AI twice, for a total of about five years. During his time at Meta, Meyer was in charge of the AI data operations team, but did not join the company's TBD lab, which is the core department for developing AI superintelligence within Meta, according to the people familiar with the matter. TBD Labs is Meta's core department for developing AI superintelligence, and a number of OpenAI's top AI researchers already work there.
Meyer told TechCrunch that his initial role was to "help set up the lab and take care of all the things that were needed" rather than focusing on data, and that he was "part of the TBD lab from day one" rather than being excluded from the core AI department. Meyer also clarified that he "did not report directly to Alexander Wang" and that he was "very happy" with his experience at Meta, and that his departure was due to "personal matters".
In addition to staffing changes, Meta's partnership with Scale AI is also changing. Five people familiar with the matter said TBD Labs is working with third-party data labeling service providers outside of Scale AI to support training for its next-generation AI models. The people said the third-party providers include Mercor and Surge-both of which are major competitors of Scale AI.
While it's not uncommon for AI labs to work with multiple data labeling providers (Meta worked with Mercor and Surge before TBD Labs was founded), it's rare for an AI lab to invest so much money in a single data provider. This makes the current situation particularly interesting: multiple sources have pointed out that even though Meta has invested billions of dollars in Scale AI, TBD Labs researchers still consider Scale AI's data to be of low quality and prefer to work with Surge and Mercor.
Scale AI's original business model centered on crowdsourcing, relying on a large and low-cost workforce to perform simple data annotation (i.e., tagging and annotating raw information to support AI model training). But as AI models have become more complex, they now require highly skilled domain experts - such as doctors, lawyers, scientists, etc. - to generate and optimize high-quality data to improve model performance.
While Scale AI has worked hard to attract this type of expertise through its Outlier platform, competitors such as Surge and Mercor are growing at a faster pace because these companies' business models have been built on well-paid talent from the start.
A Meta spokesperson denied that there were quality issues with Scale AI's products, while Surge and Mercor declined to comment. Asked why Meta has become increasingly reliant on rival data service providers, a Scale AI spokesperson only referred TechCrunch to a statement from the initial announcement of the investment -- which said the partnership would broaden the scope of the companies' commercial collaboration.
Meta's work with third-party data providers suggests that even with billions of dollars invested in Scale AI, Meta isn't betting everything on the startup. Scale AI, however, is a very different story. Shortly after Meta announced its massive investment in Scale AI, OpenAI and Google said they would stop working with the company.
Scale AI laid off 200 people in its data labeling business in July after losing two major customers. The company's new CEO, Jason Droege, blamed the changes in part on "changing market demands. He said Scale AI will expand staff in other areas of the business, including government sales -- the company recently signed a $99 million contract with the U.S. Army ($706 million at current exchange rates).
Initially, there was speculation that Meta's investment in Scale AI was really an attempt to attract Alexander Wang -- who has been deeply involved in AI since Scale AI's inception in 2016 and is the company's founder. So far, it looks like he's actually helping Meta attract top AI talent.
But beyond Alexander Wang, questions remain about the actual value of Scale AI to Meta.
A current MSL employee said a number of executives who joined Meta from Scale AI did not join the TBD Labs core team.
Additionally, two former employees and a current MSL employee revealed that Meta's AI department has become increasingly disorganized since Alexander Wang and a group of top researchers joined the company. New talent from OpenAI and Scale AI, they say, have become disenchanted with the bureaucratic complexities of a large company, while Meta's original generative AI team has found its scope of responsibility squeezed.
These tensions suggest that Meta's largest AI investment to date appears to have gotten off to a rocky start, despite the fact that it was supposed to address the company's AI R&D challenges. A current and a former employee told TechCrunch that Meta CEO Mark Zuckerberg was unhappy with the performance of the company's AI team after the mediocre release of the Llama 4 model in April.
In an effort to turn things around and catch up with OpenAI and Google, Zuckerberg has quickly pushed forward with a number of collaborations and launched an aggressive hiring program to attract top AI talent.
In addition to Alexander Wang, Zuckerberg has managed to poach a number of top AI researchers from OpenAI, Google DeepMind, and Anthropic. meta has also acquired AI voice startups such as Play AI, WaveForms AI, and has announced a partnership with AI image generation startup Midjourney.
To support its AI ambitions, Meta recently announced that it will build massive data centers in multiple locations across the US. The largest of these projects is the $50 billion (356,471 million yuan at current exchange rates) Hyperion data center in Louisiana - a name derived from the Titan that gave birth to the god of the sun in Greek mythology. the Titan that gave birth to the Sun God.
Alexander Wang is not an AI researcher by background, and his choice to head the AI Lab is seen as somewhat unconventional. Zuckerberg has reportedly talked to more traditional candidates to lead the lab, such as OpenAI chief research officer Mark Chen, and tried to buy a startup founded by Ilya Sutskever and Mira Murati, but those attempts failed.
Wired magazine previously reported that some of the new AI researchers who recently joined Meta from OpenAI have left. At the same time, many of the older employees of Meta's generative AI department have also left due to a series of changes.
Rishabh Agarwal, an AI researcher at MSL, is one of the latest employees to leave Meta, as he posted on the X platform this week. agarwal said, "Mark and @alexandr_wang's (Alexandr Wang) offer to work on the superintelligence team was extremely attractive. But in the end, I chose to follow Mark's own advice: 'In a world that's changing so fast, the biggest risk is not taking any risks'."
Agarwal declined to comment when asked about his experience at Meta and his reasons for leaving.
In recent weeks, Chaya Nayak, Meta's director of generative AI product management, and Rohan Varma, a research engineer, have also announced their departures. The question now is whether Meta can stabilize its AI business operations and retain the core talent it needs for future growth.
MSL has now started working on the next generation of AI models. According to Business Insider, the team plans to launch this new model by the end of this year.