NVIDIA and Meta Expand AI Collaboration with $3 Billion Investment

en.Wedoany.com Reported - On October 27, NVIDIA and U.S. technology company Meta announced an expansion of their artificial intelligence collaboration, under which the two parties will conduct joint optimization around Meta's Muse agents and the Muse series of models, with a focus on improving model training and inference performance. NVIDIA will also invest $3 billion in Meta's equity offering, with the collaboration covering AI models, agent safety technology, and computing infrastructure.

The two parties plan to extend the collaboration to the operational safety and behavior monitoring of AI agents. Technical directions include adopting NVIDIA's OpenShell reference architecture to establish agent runtime isolation and safety control mechanisms, and exploring the use of NVIDIA Sentry to monitor agent behavior. OpenShell supports runtime environment isolation and policy enforcement, while Sentry utilizes the BlueField-4 data processor to perform independent monitoring and hardware-level security policies. NVIDIA released the relevant reference architecture on October 21.

The computing infrastructure collaboration involves the Vera Rubin computing platform, Spectrum-X Ethernet networking, and the BlueField-4 data processor. Meta plans to expand deployment of the relevant technologies to support AI model training, inference, and large-scale agent workloads. The two companies established a multi-generational infrastructure partnership in February 2026, covering GPUs, CPUs, and networking equipment, and jointly conducting model and system optimization.

Regarding the equity investment, NVIDIA has committed $3 billion to participate in Meta's stock offering. Meta announced a new stock financing arrangement on October 26, with the proceeds to be used for AI infrastructure expansion and related business investments.