A week after announcing your approach with the nuclear sector, Meta (Ex-Facebook) continues your very expensive purchases. According to Bloomberg’s information, Mark Zuckerberg’s company has entered conversations with Scale AI, one of the main new US companies in the field of artificial intelligence in the United States.
HE deal It could be sealed around $ 10 billion, reports the US agency, which would make it the largest investment made by the field in the field, from the appearance of conversational robots in 2022. The AI of scale had been valued at $ 14 billion in 2024, and even discussed with the investors of a possible offer of public purchases at 25 billion dollars. He gave a turnover of $ 870 million (2024), but estimates that he can double this result in 2025, reaching 2 billion.
Data and a delay to catch up
Founded and administered by Alexandr Wang, Scale AI specializes in the processing and annotation of the data used by the AI models. Therefore, the company allows, according to the meticulous work of several hundred employees in developing countries, to order the texts, images and videos that serve as the basis for training models. A crucial step, since the availability of quality data is one of the main brakes in the development of generative artificial intelligence. Without these data, without conversational robot; With degraded data, risks of errors of much greater “hallucinations”.
Therefore, scale AI offers to calibrate, qualify and sell data to large models operators. It also offers an information exchange platform for researchers and AI engineers, which come from 9,000 different cities and locations according to the AI of the scale. Its clients gather OpenAi or Microsoft, the latter has also financially supported the new company, as well as goal or NVIDIA, the main flea designer.
In order, this possible investment is taking strategic interest: to obtain a rare weird material and led to become more and more, since the models consume existing data. The giant also seeks to compensate for its competitors: while Microsoft admitted Openai ($ 13 billion) and Google or Amazon have invested in anthropic, Meta defends an AA model for several years open sourcecalled flame, whose data, training and results are accessible to all.
Common Strategies in Defense
In early 2025, Meta announced its intention to invest a lot, up to $ 65 billion, in AI, to develop its particular AI infrastructure. After the centers and energy supplies, the purchase of scale would mark a strategic turn point towards a less participatory model.
The AI of the scale is in particular to allow a very safe management of chatbots responses: for this reason, applications in sensitive areas are possible, and the scale has resorted to health or defense to develop. The start -up is part of the “Thunderforge” closed club, a program launched by the Pentagon and its SATH TECHNICAL DEFENSE INNOVATION UNIT, to develop applications of AI in the service of the US army.
This positioning in sensitive segments is of interest to finish, which has approached the world of defense in recent months. At the end of May, the group announced, for example, an association with Andundil, another Pepita by Silicon Valley, to develop connected helmets. It also contributed to the Defensa Llama model, a model dedicated to research in classified documents, put into service, precisely by IA.
Source: BFM TV
