NVIDIA Invests $3 Billion in Meta for Muse and Infrastructure

Key Takeaways

1NVIDIA will invest $3 billion in Meta's capital

2The agreement aims to optimize the training and inference of Muse models

3Meta plans to deploy NVIDIA's Vera Rubin, Spectrum-X, and OpenShell processors

4Muse has surpassed ten million downloads in forty days

Why it matters : This alliance marks a strategic turnaround for Meta and creates a challenging duopoly in the agentic AI market.

NVIDIA will invest $3 billion in Meta, just two days after the announcement of an agreement aimed at optimizing the training and inference of the Muse models. The cooperation extends to NVIDIA's chips and network architectures, while Meta marks a turning point after considering its own AI processors.

Meta had aimed for hardware autonomy before sealing massive agreements

In March of last year, Meta announced its intention to reduce its reliance on NVIDIA hardware by designing its own AI-dedicated chips. In the meantime, in February, the two companies signed a multi-year agreement covering Meta's acquisition of millions of chips and the integration of NVIDIA's Grace processors into its data centers. On the investment trajectory, Meta also announced a budget of between $115 billion and $135 billion dedicated to AI infrastructure by 2026.

A $3 billion investment with a target: Muse

NVIDIA will invest $3 billion in Meta's capital. The operation, made public two days ago, explicitly aims at the joint optimization of the training and inference of Meta's Muse models. The agreement concerns Meta's AI models and includes an equity stake.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

The targeted infrastructure: Vera Rubin, Spectrum-X, and OpenShell

Meta plans to deploy its partner's next-generation infrastructure on a large scale. The setup includes Vera Rubin processors as well as Spectrum-X network accelerators, which NVIDIA has described this year as the foundations of the "agentic AI era." The two groups also plan to collaborate on OpenShell, a reference architecture introduced by NVIDIA at the end of September to isolate autonomous agents in secure environments.

Muse, a personal agent already launched and encrypted

Muse, the AI range developed by Meta, was launched in September in the United States as a personal agent capable of acting on behalf of its users. In just forty days, downloads surpassed the ten million mark.

Platforms and public statements from leaders

Meta owns Facebook, Instagram, WhatsApp, and Messenger. Facebook has 3 billion users, and Messenger has 1 billion, totaling "nearly four billion" people overall. In terms of statements, Jensen Huang describes AI agents as a new computing paradigm, capable of understanding goals, reasoning, using tools, and accomplishing tasks on behalf of users, which would shift computer usage from executing applications to collaborating with intelligent agents. Mark Zuckerberg presents Muse as a means to bring personal superintelligence to everyone and expresses his excitement to continue collaborating with NVIDIA to offer personal agents and cutting-edge models to billions of people.