Like Meta with Scale AI, GAFAMs invest large sums in AI young shoots to debum their best elements without triggering the opening of an antimonopole survey.
Meta has just gave a serious boost to its artificial intelligence strategy with a massive investment in Scale AI, a young shoot founded in 2016 which quickly imposed itself as a major player in the AI value chain. The announcement is concomitant to that of the start -up of a new AI laboratory at Meta, with the aim of reaching superintendent.
Meta plans to invest the astronomical sum of $ 14.3 billion in the start-up, enough to acquire 49% of its capital. In exchange, the Californian giant recovers its prestigious founder and leader, Alexandr Wang, who will take the lead, as well as several leading engineers.
Fill the limits of Llama 4
In a little less than ten years, Scale AI, a start-up based in San Francisco, has been able to make a place of choice in the AI ecosystem. The young shoot is indeed specialized in the training phase of algorithms, through the structuring, annotation, validation and optimization of data intended for the training of models. This step is necessary to obtain quality datasets, themselves essential to cause the most efficient algorithms possible. As the Snowflake company repeats, specialist in cloud data management, “there is no AI strategy without data strategy”. Or, as Alexandr Wang writes himself, “data is the vital element of AI systems. This is what motivated us to create Scale.”
Scale AI has the most prestigious names in the AI world among its customers, alphabet (which would spend between 150 and 200 million euros per year for its services) to Openai via XAI. However, all of these companies have announced that they wanted to distance themselves from Meta’s investment.
For the group of Mark Zuckerberg, getting hands on this nugget is therefore a sure way to improve the quality of its large language models, in the fierce race that the Tenors of the AI engage. The latest Meta model, Llama 4, launched in early April, has indeed encountered performance problems, including alignment concerns, and Scale AI technology could prove to be a precious asset to allow the company to catch up on Chatgpt (OPENAI) and Gemini (Alphabet) and Claude (Anthropic).
“Meta is thus reinforcing her position on the AI market. This investment allows her to walk on his two legs, with the LLM branch on one side, on which Meta has always been good, with Llama, and on the other hand the database branch and strengthening, training of models, on which she had a small weakness that Scale Ai will come to fill”, notes Eric Le Quellenec, a lawyer specialized in Flichy Grangé Avocats.
Why the disguised redemptions are on the rise
Formally, Meta did not buy Scale AI. But as illustrated by the reaction of Alphabet, Openai and Xai, it is actually just like. The massive investment of Meta in the young shoot is part of the category of disguised acquisitions, flourishing in the past two years, in particular in AI. This practice consists, for an AI giant, in investing a large sum of money in a young shoot, in order to recruit its founders, its best engineers, and obtain a license on its technology, leaving more often than an empty shell in its wake.
The American giant of AMD semiconductors, who dreams of Nvidia’s rival, has just proceeded the same with Untether Ai, specialist in flea dedicated to inference. After having invested an amount in the amount not revealed in the young shoot, AMD started all of its engineering teams all of its teams. In March 2024, Microsoft had spent $ 650 million in Infection AI to recruit its co -founders, Mustafa Suleyman (one of Deepmind’s co -founders), immediately promoted new CEO of Microsoft AI, and Karén Simonyan, as well as his best IA engineers. Amazon and alphabet did the same with, respectively, Adept Ai and Character ai.
Unlike well -and -for -the -art buyouts, these disguised acquisitions have the advantage of going more easily under the radar of the competition authorities. For Meta, who is currently faced with an anti -trime trial on the part of the American justice department, which could lead to the dismantling of the company, the moment would be clearly poorly chosen for a leading buyback.
But this type of agreement also has a definite interest in the founders of start-ups, at a time when the headed capital towards them tend to dry up and when entering the stock market remains a risky option at which few young shoots are risking in the current economic context. For AI entrepreneurs subjected to astronomical costs, a mirobal investment associated with a promise of hiring with an interesting position and a very high salary at the key can be devilishly attractive, or even a simple question of survival.
Towards an alarm clock from the anti -lord authorities?
It remains to be seen how long this type of chords will be able to go under the radars of the authorities. In the United States, the FTC, which seems to want to keep the line hard against the technological monopolies adopted under the Biden administration, is currently combing the investments of Microsoft and Amazon in Inflection and Adept AI, while the Department of Justice is interested in that of Google in Character AI.
The Clayton Antitrust Act also gives the FTC and the Department of Justice the possibility of investigating an equity investment in a company, even when it is not in the majority, as is the case of Meta’s investment, which ensures 49% of the capital of Scale AI. Admittedly, the Mark Zuckerberg group currently does not enjoy a monopolistic position on the AI, the market being very competitive. This disguised acquisition is therefore less likely to have the competition authorities tick up than if it was for example a new nugget of social networks.
Nevertheless, the Antimonopolian authorities have in the past committed the error of letting Meta buy young promising young shoots, like WhatsApp and Instagram, at a time when the landscape of social networks was more competitive. Affairs which allowed him to reach the ultra-dominant position from which he benefits today, and which the Department of Justice now plans to revoke authorization, which would however represent a real headache, the acquisitions in question dating back more than ten years ago. It is therefore likely that they do not want to repeat the same error. The Senator of the Massachusetts Elizabeth Warren, known for her anti -lone positions, has already called the competition authorities to assess an agreement likely “to break the law by eliminating competition”.
Europeans could also be interested in it. The German Competition Authority is evaluating whether it has jurisdiction to examine Microsoft’s investment in inflection under its antimonopoly legislation, which, in accordance with the principle of European subsidiarity, could then trigger an investigation by the Commission, and would be bad news for other disguised acquisitions. In addition, the DMA requires access controllers (of which META is a part that they inform the commission of any concentration involving their digital services.
In total, according to Eric Le Quellenec, “there is a good chance that this investment has moved to fine, because if there were concentration on the market for these training databases, it will be problematic from the point of view of the competition authorities. Especially since behind Scale AI, there are very large customers …”
