Nvidia said in a 26 October release that it will invest $3 billion in Meta’s equity offering as the companies expand work on Muse models to improve training and inference performance. The GlobeNewswire distribution of Nvidia’s release is dated 27 October and carries a Santa Clara, California, dateline.
Key points
- Nvidia says it will invest $3 billion in Meta’s equity offering.
- Meta intends to expand deployment of Nvidia Vera Rubin infrastructure and Vera CPUs, and will also deploy Spectrum-X Ethernet networking and BlueField-4.
- The companies intend to optimise Muse training and inference workflows and will consider work on agent deployment and monitoring tools.
Nvidia’s $3 billion equity commitment
The proposed investment adds an ownership position to Nvidia’s existing role as a supplier of infrastructure used by Meta’s Muse models and agents. An earlier AI Affairs report on Meta’s financing plans covered a $35 billion equity offering and plans for another $60 billion. Nvidia describes its investment as proposed in the forward-looking statements accompanying the release.
Nvidia says completion of the investment could depend on applicable closing conditions. Its forward-looking statements also cover expectations about arrangements with partners and the benefits of its products. The $3 billion commitment therefore sits alongside a separate programme of planned equipment deployment and technical work, rather than a stated payment for that work.
Meta’s financing plans provide the immediate capital context. AI Affairs previously reported that Meta’s revenue rose 26% as Mark Zuckerberg set out Muse growth and a cost target. Nvidia’s release identifies the equity offering as the destination for its investment and sets out the companies’ intentions for computing infrastructure.
Meta plans more Vera Rubin infrastructure
Meta intends to expand deployment of Nvidia Vera Rubin infrastructure and Vera CPUs for next-generation AI agents, according to Nvidia. The release says Meta will also deploy Spectrum-X Ethernet networking and BlueField-4 to support data movement across its infrastructure. Those plans extend the companies’ work beyond model development into the equipment used to run workloads.
Nvidia says the companies intend to optimise workflows across training and inference, with better performance and total cost of ownership as their aims. Inference is the stage at which a trained model handles a user’s request. Nvidia also describes performance per watt as an objective for the planned expansion of Meta’s computing deployment.
The release attributes what it calls frontier-level benchmark performance and increased consumer downloads to Muse models and agents using Nvidia infrastructure. Those are Nvidia’s assessments. For the infrastructure work ahead, the company frames gains in performance and operating cost as goals of further optimisation.
Muse agents bring OpenShell into discussion
The companies will also consider collaborating on Nvidia’s OpenShell Reference Architecture. Nvidia identifies OpenShell with the safe deployment of agents and Sentry with monitoring their behaviour. This work is described as a potential addition to the Muse partnership, while the release separately states Meta’s intention to expand its computing deployment.
Jensen Huang, Nvidia’s founder and chief executive, described agents as software that can reason, use tools and carry out tasks for users. Mark Zuckerberg, Meta’s founder and chief executive, said the companies would continue working together on personal agents and models. Their comments place the planned infrastructure work alongside Meta’s ambitions for distributing Muse.
Meta has pursued subscriptions as a distribution route for Muse in Japan, as AI Affairs reported. Nvidia says demand for useful agents is driving the expanded collaboration. Its release connects the proposed optimisation work to the cost of running those agents as well as to the work of training models.
Nvidia’s forward-looking statements identify market acceptance, changes in customer preferences, product defects and unexpected performance losses when its technology is integrated into systems among the risks that could cause outcomes to differ from its expectations.
