Meta takes $15B stake in Scale AI in bid to catch competitors

Meta has agreed to pay $14.8 billion for a 49 percent stake in the artificial intelligence data firm Scale AI in one of the biggest acquisitions the social media giant has made since its 2014 deal to buy WhatsApp, according to a person familiar with the matter.

The arrangement with Scale AI, which could be announced as soon as Wednesday, will give the start-up’s CEO, Alexandr Wang, a senior position at Meta leading a team focused on developing super intelligence, said the person, who spoke on the condition of anonymity to discuss private company matters.

Super intelligence describes a hypothetical technology that could perform better than humans at all tasks. Meta, which is an existing Scale AI customer, is expected to expand its use of the company’s data-labeling services, the person said. The deal was reported earlier by the Information. The New York Times and Bloomberg News earlier reported Meta’s decision to create a new AI lab. Meta and Scale AI declined to comment.

The deal will give Meta more access to Scale AI’s services, which have shifted over the years. The company first specialized in connecting large tech companies with human contractors to label the massive amounts of data necessary to train AI models, focusing on self-driving vehicles and relying on “digital sweatshops” in areas such as the Philippines.

As the technology progressed to require little labeling, Scale pivoted to the types of contractors and data required for later stages of AI development, including making chatbots sound more human or helping them improve in specific categories, such as health care or defense.

Outlier AI, a platform Scale uses to hire contractors, on Tuesday advertised 28 roles, mostly for specialized human trainers focusing on subject areas such as advanced biology, chemistry and physics, and for a number of voice-training roles in languages including Chinese and Arabic.

Meta’s acquisition could face scrutiny from antitrust regulators in the Trump administration who, while cautious about regulating AI, have expressed skepticism about the power and effect of big technology companies, including Meta, said William Kovacic, a former Federal Trade Commission chair and a law professor at George Washington University.

The FTC earlier this year opted not to drop its lawsuit challenging Meta’s acquisitions of Instagram and WhatsApp, even after Meta CEO Mark Zuckerberg lobbied the White House for a resolution.

Earlier this month, panelists at an FTC event called for stiff regulatory changes to help make the internet safer for kids and teens. But FTC Chair Andrew Ferguson said in September that “AI may pose a much-needed competitive and innovative challenge to incumbent Big Tech firms.”

“A knee-jerk regulatory response will only squelch innovation, further entrench Big Tech incumbents, and ensure that AI innovators move to jurisdictions friendlier to them—but perhaps hostile to the United States,” he added.

Kovacic said that, if antitrust regulators examine Meta’s acquisition, they would have to determine whether the company could be considered a dominant player in the AI field to take action.

“My sense of looking at the commentary from the business community is that Meta is not the leader in the development of artificial general intelligence—that it is trailing the field,” he said.

Meta’s alliance with Wang is the latest in personnel changes as the social media giant attempts to compete with tech firms that have a stronger reputation in generative AI, such as OpenAI, Anthropic, Google, Microsoft and China-based DeepSeek. Zuckerberg has said he expects that Meta’s AI chatbot will become the leading AI assistant, besting rivals such as ChatGPT and Claude.

Meta earned a lot of goodwill from industry insiders when its popular AI model, called Llama, launched in 2023, earning points for its relative openness as OpenAI and Google began to publish fewer papers and reveal less about how their technology was developed.

But, in recent months, Meta has faced public stumbles. In April, it was caught trying to manipulate a platform for evaluating AI models called LMArena. And last month, Meta said it would delay the rollout of its gargantuan AI model called Behemoth, according to a report in the Wall Street Journal.

Meanwhile, Meta’s recently launched AI app, which Zuckerberg touted in podcast appearances, seems to have fallen flat with consumers. It is ranked No. 17 in the productivity category on Apple’s App Store and has fewer than 10 million downloads on the Google Play Store.

In April, Joelle Pineau, who had worked at the company for about eight years, announced that she was leaving her post leading Meta’s AI research lab. During her tenure, Meta made significant moves to reshape the once-independent artificial-research lab to be more aligned with the company’s business and product priorities.

The following month, Meta reorganized its generative AI team, splitting the department into two groups focused on AI research and consumer products, according to an internal memo about the plans seen by The Washington Post.

“I believe this structure will be a major upgrade to overcome the biggest challenges that I’ve heard from many of you, and will help accelerate our overall progress,” wrote Meta Chief Product Officer Chris Cox.