Meta will invest hundreds of billions of additional dollars in IA infrastructure

The boss of Meta Mark Zuckerberg wants to invest “hundreds of billions of dollars” in artificial intelligence infrastructure (AI). Consequently, the last advertisement of major investment in order to achieve his displayed objective, build “super -saked”.

“We are in the process of building computer networks of several gigawatts of power. In addition, ” he said on a message published on his social networks Facebook and Threats on Monday.

These data centers of a completely different dimension than those already existing to house the servers of the Cloud (remote computer) are designed specifically to train. Therefore, accommodate generative AI models.

These require enormous computing power, and therefore advanced computer flea and a lot of energy.

Meta has been late in the generative AI, dominated by Openai, since the launch of Chatgpt, and Google.

All the leaders in the sector have in sight meta will invest hundreds billions an AI called “general”. or “superintendent”, with cognitive capacities superior to those of humans, capable of making scientific discoveries and inventing technologies itself.

“The first (computer network), which we call Prometheus, will be put into service in 2026. We also build Hyperion, which will be able to reach a capacity of 5 gigawatts over the years,” said Mark Zuckerberg.

A power of 5 gigawatts represents the equivalent of annual electricity consumption of 1 to 4 million American households.

In mid-June. the social networks giant paid more than US $ 14 billion to acquire 49% of the capital of Scale AI, specializing in the use of data used to develop AI models.

According to the boss of Open AI. Sam Altman, Meta also offered an individual bonus at the signing of more than $ 100 million to “a lot” of Openai employees and approximately the same annual salary amount.

Mark Zuckerberg meta will invest hundreds billions then confirmed in an internal Memo that no less than seven of them had changed stables. just like the general manager of Scale AI, Alexandr Wang, and several elements of the Anthropic and Google rivals.

The billionaire repeated on Monday that it intended to constitute “the most elitist. dense talent team in all industry”, and that it would have “a level of computing power in the sector”.

“We have capital from our activities to do so,” he said.

The latest big model of Meta AI, Llama 4, launched in early April, disappointed.

He arrives behind all the heavyweights. American, Chinese and French, in the classification established by the independent evaluation platform LMARENA on code writing, and even behind his predecessor LLAMA 3 for the text interface.