Meta invests in AI firm Scale and recruits its CEO for ‘superintelligence’ team

Meta said Thursday it is making a large investment in artificial intelligence company Scale and recruiting its CEO Alexandr Wang to join a team developing “superintelligence” at the tech giant.

The move reflects a push by Meta CEO Mark Zuckerberg to revive AI efforts at the parent company of Facebook and Instagram as it faces tough competition from competitors such as Google and OpenAI.

Meta announced what it called a “strategic partnership and investment” with Scale late Thursday but didn’t disclose the financial terms of the deal. Scale said the added investment puts its market value at over $29 billion.

It won’t be the first time a big tech company has gobbled up talent and products at innovative AI startups without formally acquiring them. Microsoft hired key staff from startup Inflection AI, including co-founder and CEO Mustafa Suleyman, who now runs Microsoft’s AI division.

Google pulled in the leaders of AI chatbot company Character.AI, while Amazon made a deal with San Francisco-based Adept that sent its CEO and key employees to the e-commerce giant. Amazon also got a license to Adept’s AI systems and datasets.

Wang was a 19-year-old student at the Massachusetts Institute of Technology when he and co-founder Lucy Guo started Scale in 2016.

They won influential backing that summer from the startup incubator Y Combinator, which was led at the time by Sam Altman, now the CEO of OpenAI. Wang dropped out of MIT, following a trajectory similar to that of Meta CEO Mark Zuckerberg, who quit Harvard University to start Facebook more than a decade earlier.

Scale’s pitch was to supply the human labor needed to improve AI systems, hiring workers to draw boxes around a pedestrian or a dog in a street photo so that self-driving cars could better predict what’s in front of them. General Motors and Toyota have been among Scale’s customers.

What Scale offered to AI developers was a more tailored version of Amazon’s Mechanical Turk, which had long been a go-to service for matching freelance workers with temporary online jobs.

More recently, the growing commercialization of AI large language models — the technology behind OpenAI’s ChatGPT, Google’s Gemini and Meta’s Llama — brought a new market for Scale’s annotation teams. The company claims to service “every leading large language model,” including from Anthropic, OpenAI, Meta and Microsoft, by helping to fine tune their training data and test their performance. It’s not clear what the Meta deal will mean for Scale’s other customers.

Wang has also sought to build close relationships with the U.S. government, winning military contracts to supply AI tools to the Pentagon and attending President Donald Trump’s inauguration. The head of Trump’s science and technology office, Michael Kratsios, was an executive at Scale for the four years between Trump’s first and second terms. Meta has also begun providing AI services to the federal government.

Meta has taken a different approach to AI than many of its rivals, releasing its flagship Llama system for free as an open-source product that enables people to use and modify some of its key components. Meta says more than a billion people use its AI products each month, but it’s also widely seen as lagging behind competitors such as OpenAI and Google in encouraging consumer use of large language models, also known as LLMs.

It hasn’t yet released its purportedly most advanced model, Llama 4 Behemoth, despite previewing it in April as “one of the smartest LLMs in the world and our most powerful yet.”

Meta’s chief AI scientist Yann LeCun, who in 2019 was a winner of computer science’s top prize for his pioneering AI work, has expressed skepticism about the tech industry’s current focus on large language models.

“How do we build AI systems that understand the physical world, that have persistent memory, that can reason and can plan?” LeCun asked at a French tech conference last year.

These are all characteristics of intelligent behavior that large language models “basically cannot do, or they can only do them in a very superficial, approximate way,” LeCun said.

Instead, he emphasized Meta’s interest in “tracing a path towards human-level AI systems, or perhaps even superhuman.”

LeCun co-founded Meta’s AI research division more than a decade ago with Rob Fergus, a fellow professor at New York University. Fergus later left for Google but returned to Meta last month after a 5-year absence to run the research lab, replacing longtime director Joelle Pineau.

Fergus wrote on LinkedIn last month that Meta’s commitment to long-term AI research “remains unwavering” and described the work as “building human-level experiences that transform the way we interact with technology.”