NVIDIA 与 Meta 扩大 Muse 合作，聚焦 AI 智能体

NVIDIA and Meta are expanding their partnership around Muse and the broader Muse model family as demand grows for AI agents that can reason, use tools and act on users’ goals. The companies plan to tune the models, agents and underlying infrastructure across training and inference, targeting higher performance and a lower total cost of ownership.

Meta says Muse models and agents already run on NVIDIA infrastructure, have reached frontier-level benchmark results and have seen consumer downloads surge since launch. The next phase will focus on optimizations across complete workflows, linking model development with the computing stack that serves the agents. The announcement does not set out a release timetable or new access details for users.

Security and oversight are also part of the planned work. Meta and NVIDIA will consider collaborating on NVIDIA’s OpenShell Reference Architecture, including OpenShell for safer agent deployment and NVIDIA Sentry for monitoring agent behavior. This would extend the partnership beyond model performance to controls for agents operating on users' behalf.

Meta also intends to expand its use of NVIDIA infrastructure. Planned deployments include Vera Rubin and Vera CPUs for next-generation AI agents, with performance per watt as a central goal. Spectrum-X Ethernet networking and BlueField-4 will support network efficiency and throughput as workloads scale.

The collaboration is backed by a major financial commitment: NVIDIA will invest $3 billion in Meta’s equity offering. The investment ties the deeper technical partnership to Meta’s wider AI infrastructure ambitions and NVIDIA’s role in supplying the computing stack for agent workloads.

NVIDIA founder and CEO Jensen Huang described agents as a new computing paradigm that can understand goals, reason, use tools, and complete work on a person’s behalf, shifting computing away from manually operating applications. Meta founder and CEO Mark Zuckerberg framed Muse as the company’s route to bringing personal superintelligence and frontier models to billions of people. Together, the statements position Muse as both a consumer agent effort and a large-scale infrastructure program spanning models, compute, networking and safety tooling.

Sources and related context

• Meta introduces Muse Wonder and Muse Wonder Full: Related context: Meta's earlier model release explains the Muse family's agent capabilities and its stated plans for Muse access.
• NVIDIA launches OpenShell Reference Architecture: Related context: The earlier announcement explains the OpenShell and Sentry infrastructure named as possible areas for collaboration.

Source