Drew Ander/Getty Images
The biggest clients of AI startups, including Openai and Google, would not want to trust their data in almost half of the companies owned by Mark Zuckerberg's Tech Empire.
sCALE AI's shock sells 49% stakes to Meta in a $15 billion deal, calling for CEO Alexandr Wang to leave the startup that was established to lead a new AI lab with Tech Giant, and the startup's $14 billion data labeling business could explode.
While playing, there is concern that Scale, the dominant player at labeling data that helps large tech companies and AI startups train their models, can share details about the type of data that AI players have used to build the most cutting edge technology in Meta.
As former scale employee said Forbes: “They all want to cut back on scale-offs. When they become part of the meta, they expand as a business and completely collapse.”
One of its most well-known clients, Openai has already scaled its work on scale, according to four sources with knowledge of its business. According to two sources, this has been happening for several months, with Openai examining potential new partners. Openai declined to comment.
The shocking move said it treasured the company for $28 billion and scrambled people on scale to disrupt them. Some staff are concerned about which visibility meta is included in past projects, despite most contracts between the company and the stipulated data being removed after the project is completed, the person added.
Scale declined to comment.
Scale's small rivals have already taken over positions and are welcoming customers who are concerned about conflicts of interest when it comes to data and privacy. Francis Pedraza, co-founder of Invisible Technologies, said Forbes His company is committed to being independent. Startup Turing already provides model training data to Openai, Anthropic and Google, and sees this deal as an opportunity to become a “Swiss” and an impartial distributor of data to the Frontier AI Lab. CEO Jonathan Sidharth said Forbes The “clients want to work with neutrals that can even support all labs.” One investor who supported Scale's competitors said the deal would create new ways for other companies to “get the open space left by Scale AI.”
Access to high-quality human data is important for training powerful AI models, so data is considered human moats for AI juggernauts like Openai and AI races. Having booked revenue of $870 million in 2024, Scale AI quickly captured the data labeling market by providing human-designed data to companies such as Cohere, Openai and Microsoft. However, when Meta owns almost half the scale, the equation changes.
Scale built its business behind the large army of clickworkers, primarily based overseas, by teaching AI models by providing context to large amounts of data. However, this kind of work has become a product. “Anyone who sets up a team can compete with you and that's really quick to get the price,” said Kevin Guo, co-founder of the startup that slid out of the data labeling business in 2023 interview with scales.
One executive at an AI company described the scale as the “bulk food section of the AI training market.” Scale's business-savvy sources said Forbes Startups suffer from quality issues. “They're over-sold, they sell too much, and then they don't hid very often,” the former employee said.
With an estimated net worth of $3.6 billion and one of the world's youngest home-built billionaires at age 28, Wang is heading Meta's new lab focused on building a so-called top spot. information It has been reported. To build the team around him, Zuckerberg is clearly taking part in a new lab north of Scale AI and Openai, Humanity, Google Deepmind researchers north of $10 million a year. Zuckerberg has been closely involved in assembling the team, spending a very long time reaching out to potential recruits personally, such as setting up a WhatsApp group called the “Recruit Party,” and sorting desks for researchers to sit close to him.
The transaction has not yet been closed and could be blocked by regulators. If so, it will be a stairwell for early investors in Wang and Scale, including Accel and Index Ventures. However, it is not clear what will expand at that point. “This was great for Alex and early investors. It was terrible for everyone else, including employees and former employees,” said the former senior AI employee. “It's unclear how the deal will expand.”
The world's most powerful tech giants have spent the past few years trying to jockey each other for AI hegemony. Meta, which first launched the AI Lab in 2013, struggles to keep up with Google, Openai and humanity with its open source Llama models. The company's AI reputation hit in April when it was accused of artificially launching the benchmark score of the Llama 4 model (Meta rejected the claim). Shiny newcomers like Wang shot the arm for the company's AI efforts.
Zuckerberg also began pushing Meta towards a defensive contract. Forbes (As previously reported, some of the businesses had previously struggled to gain traction).
Meta is the latest company to hire CEOs of Splashy AI startups. Last year, Microsoft started Mustafa Suleyman, founder of Lab Deep Mind, and Mustafa Suleyman, a senior team from “Refracture” in 2022, leaving the company's shell behind. A few months later, Amazon signed a deal with Adept, licensed the technology and hired CEO David Luan and his founding team. Google created a similar gambit and rehired researcher Noam Shazeer, who co-invented the transformer architecture that underpins generation AI during his tenure at Google, from a startup character.
The CEO of Invisible Matt Fitzgerald noted that as the model becomes more complex, a higher level of expertise is required. Scale, Turing, and the invisible all emphasise more sophisticated tasks performed by doctoral degrees and highly educated professionals, away from low-paid clickwork. Meta trade reinforces the importance of human workers when it comes to AI training, he said. “This is a 10-year bet that this will really be important for a long time,” Fitzgerald said.
John Paczkowski, David Jeans and Iain Martin provided reports.
More from Forbes
