NVIDIA to Invest $3 Billion in Meta as Muse Partnership Expands

NVIDIA plans to put $3 billion into Meta Platforms’ new share sale as the two companies widen their work on Muse, Meta’s artificial intelligence agent, and the family of models behind it, aiming to lift both training and inference performance.

NVIDIA’s proposed $3 billion investment remains subject to closing conditions and has not yet been completed.

Meta priced its offerings, upsizing the Class A common stock sale to $25.0 billion from $22.5 billion at $625 per share. It also sold 250 million depositary shares tied to new 6.50% mandatory convertible preferred stock at $50 each, for net proceeds of approximately $12.5 billion. Both offerings are expected to close on October 29, 2026, subject to closing conditions.

When it launched the sale, Meta named NVIDIA, Advanced Micro Devices and Broadcom as cornerstone investors that had indicated interest in buying up to $5.0 billion of the Class A stock between them. NVIDIA’s statement gives $3 billion as its own figure but does not say which of the two securities it will buy. Beyond this week’s deals, Meta plans a further at-the-market share program worth up to about $60 billion, which depends on its new non-voting Class C stock starting to trade, something it does not expect before the first quarter of 2027.

On the technology side, the companies plan to optimize workflows together to get the most performance from Muse while improving total cost of ownership. NVIDIA said Meta’s Muse models and agents, which run on NVIDIA infrastructure, have achieved frontier-level benchmark results and fast-rising consumer downloads since they launched. Both companies link the expansion to popular demand for AI agents that are useful.

Meta introduced Muse in September as an agent that can open a browser, fill out forms and keep working after a user closes the app, reachable through the Muse app, WhatsApp or the web. It runs on Muse Spark, a model Meta developed. Meta says more than 10 million people in the U.S. and Canada signed up in the agent’s first weeks. Shopping is what they use it for most: more than 25% of weekly active users rely on Muse to buy things.

Meta and NVIDIA will also look at further work on NVIDIA’s OpenShell Reference Architecture, which includes OpenShell for deploying agents safely and NVIDIA Sentry for monitoring how agents behave. A summary at the top of the release presents the architecture as a way to strengthen agent security and also ties Sentry to monitoring and to processing structured, unstructured and graph data.

NVIDIA unveiled both tools in September as part of an agent safety platform. OpenShell sets and enforces the permissions an agent has, while Sentry is designed to run on BlueField-4 data processing units, separately from the agent it watches. Citi and JPMorganChase were named among the financial institutions collaborating with NVIDIA on that work.

Meta is also adding hardware. It intends to expand its deployment of NVIDIA infrastructure, including Vera Rubin systems and Vera CPUs, to run next-generation AI agents with the highest possible performance per watt. It will also install NVIDIA Spectrum-X Ethernet networking and BlueField-4 to support network efficiency and throughput, as part of what the companies call a long-term partnership behind Meta’s AI infrastructure plans.

That partnership was formalized in February, when NVIDIA announced a multiyear, multigenerational agreement covering the large-scale deployment of its CPUs and millions of Blackwell and Rubin GPUs at Meta, along with Spectrum-X Ethernet switches. At that point, the two sides said Vera CPUs could reach Meta’s data centers at large scale in 2027.

Jensen Huang, founder and CEO of NVIDIA, said: “AI agents represent a new computing paradigm — software that understands our goals, reasons, uses tools and gets work done on our behalf,” before adding: “This changes how we use computers, from operating applications to working with intelligent agents. NVIDIA and Meta are expanding our partnership to optimize the computing infrastructure end to end for capable, efficient and secure AI agents, so Muse can empower billions of people to achieve more and pursue greater ambitions.”

Mark Zuckerberg, founder and CEO of Meta, said: “Muse is how we bring personal superintelligence to everyone. We’re excited to keep working with NVIDIA to deliver the best personal agents and frontier models for billions of people."

The stake could tie NVIDIA’s returns more closely to Meta’s AI build-out, since Meta is also deploying millions of NVIDIA GPUs under the February agreement. Meta expects capital expenditures of $135 billion to $145 billion in 2026, and the new equity may give it more room to keep funding that spending.

Meta will not receive the cash from the common stock sale straight away. Under forward agreements with Morgan Stanley and Goldman Sachs, it expects to collect net proceeds of approximately $24.9 billion, before expenses, when those agreements settle on or before November 19, 2027.