The surprising agreement of 15 billion dollars from Scale AI to sell a 49% participation to Meta, with its CEO Alexandr Wang abandoning the startup that he founded to direct a new Laboratory of AI in the technological giant, has increased the spectrum that the Startup data labeling business, valued at 14 billion dollars, could implio.
What is at stake is the concern that Scale, which has been the dominant actor in data labeling to help the main technological companies and new companies to train their models, could share details about the types of data that the main actors of AI have used to build their most advanced technology with goal.
As a former Scale employee declared a Forbes : «Everyone wants to dismantle Scale now. Scale, as a business, once integrated, will collapse completely ».
Openai, one of its most prominent customers, has already begun to reduce its collaboration with Scale, according to four sources with knowledge of your business. Two sources indicated that this has been happening for months and that Openai has been evaluating possible new partners. Openai declined to comment.
The surprising decision, which would value the company at 28 billion dollars, left Scale confusing and disoriented, said the former employee. Some employees are concerned about the visibility that target can have on previous projects, despite the fact that most contracts with the company stipulate the data elimination once the projects were completed, the source added.
Scale refused to comment.
Scale’s smallest rivals are already disputed and welcomed any client concerned about conflicts of interest in data and privacy. The co -founder of Invisible Technologies, Francis Pedraza, declared Forbes that your company undertakes to maintain its independence. The startup Turing, which already supplies data for training models to OpenAi, Anthropic and Google, sees the agreement as an opportunity to be “Switzerland” and become an impartial distributor of data for the laboratories of Vanguardia. The executive director, Jonathan Sidharth, declared Forbes that “customers want to work with a neutral company that can support all laboratories equally.” An investor that has backed a Scale competitor said the agreement would create new roads so that other companies “capture the open space left by Scale IA.”
Read more: Geniuses of digital transformation 2025: AI, ally or enemy of cybersecurity?
Access to high quality human data is crucial to train powerful models, so much that the data has been considered a competitive advantage for AI giants such as OpenAi and Anthropic in the AI race. Scale AI, which registered 870 million dollars in revenues in 2024, was seized in advance of the data labeling market by providing a large amount of data labeled by humans such as Cohere, OpenAI and Microsoft. But the acquisition by almost half of Scale would change this situation.
Scale built his business thanks to a huge clicking army, mainly workers abroad that taught AI models providing context to large amounts of data. But this type of work has become a basic product. “Anyone who organizes a team can compete with you, and the price is quickly reduced,” said Kevin Guo, co -founder of a startup that separated from the data labeling business, in a 2023 interview about Scale.
A senior executive of an artificial intelligence company described Scale as “the bulk food sector of the training market in AI”. Several sources familiar with the Scale business reported a Forbes that the startup has suffered quality problems. “They promise too much and sell too much, and then, very often, they do not comply with what is promised,” said the former employee.
Wang, who is one of the world’s younger billionaires made at age 28 with an estimated net worth of 3.6 billion, will direct a new laboratory in the finish line focused on building the so -called superintelligence, an AI system that exceeds human capabilities, reported The Information . To build the team around him, Zuckerberg is apparently offering Scale AI researchers, as well as Openai, Anthropic and Google Deepmind more than $ 10 million a year to join the new laboratory. Zuckerberg is closely involved in team formation and has made great efforts such as creating a WhatsApp group called “Recruiting Party”, personally contacting possible recruits and reorganizing desks so that researchers feel close to it, Bloomberg reported.
The agreement has not yet been closed and could be blocked by regulators. If concretized, it will be an unexpected gain for Wang and the first Scale investors, such as Accel and Index Ventures. But it is not clear what will happen to Scale at that time. “This was excellent for Alex and the first investors, but terrible for everyone else, including employees and former employees,” said a former senior employee from Scale AI. “It is not clear how the agreement benefits Scale.”
The most powerful technological giants in the world have spent recent years competing with each other due to supremacy in AI. Meta, which launched its AI laboratory in 2013, has had difficulty following the rhythm of Google, Openai and Anthropic with its open source call models. The company’s reputation was affected in April when she was accused of artificially increasing the reference scores of her models call 4 (target denied the accusation). A new brilliant incorporation such as Wang could be a boost for the company’s efforts.
Zuckerberg has also begun to boost the goal towards defense contracts, and Scale’s government arm could join those efforts (although as Forbes He previously informed, that part of his business has previously had difficulty gaining traction).
Meta is the last company to say goodbye to the CEO of an outstanding startup of AI. Last year, Microsoft signed Mustafa Suleyman, founder of the Deepmind Laboratory, already its infection management team, the Laboratory of AI that Suleyman founded in 2022, leaving behind a ghost company. Months later, Amazon signed an agreement with ADEPT to license its technology and hire CEO David Luan and his founding team. Google performed a similar maneuver to re-re-re-re-re-re-reached the Naam Shazeer-who co-invent the architecture of transformers that supports the generative AI during its stage in Google-of its startup Character.AI.
Matt Fitzpatrick, Executive Director of Invisible, said that, as the models become more complex, they require higher levels of experience. Scale, Turing and invisible have set aside the work of poorly paid clicks to focus on more sophisticated tasks conducted by doctors and professionals with a high level of training. The agreement with goal reinforces the importance of human workers in AI training, he said. “This is a 10 -year bet that this will be very important for a long time,” said Fitzgerald.
Rashi Shrivastava, John Paczkowski, David Jeans and Iin Martin Contribuyeron with Este Reportje.
This article was originally published by Forbes Us.
You may be interested: Geniuses of digital transformation 2025: AI, allied or enemy of cybersecurity?
