Nvidia puts $3 billion into Meta’s stock sale and tunes Muse for its chips

Nvidia said on October 26 that it will invest $3 billion in Meta’s equity offering while the two companies expand their work on Muse, Meta’s family of AI models and agents, as Reuters reported.

What Nvidia and Meta say they will build together

According to Nvidia’s announcement, the companies will optimize Muse jointly across workflows, with the aim of better inference and training performance and a lower total cost of ownership. Meta says its Muse models and agents already run on Nvidia infrastructure and have posted frontier-level benchmark results and strong consumer download growth since launch. Those are Meta’s own claims, and the release gives no figures for them.

On hardware, Meta plans to expand its deployment of Nvidia Vera Rubin and Vera CPUs for next-generation agents, with a focus on performance per watt. It will also deploy Nvidia Spectrum-X Ethernet networking and BlueField-4 to improve network efficiency and throughput.

The two sides will also consider working together on the Nvidia OpenShell Reference Architecture. It includes OpenShell, for deploying agents safely, and Nvidia Sentry, a tool for monitoring what agents do. The release presents this as something under consideration, not a commitment.

Jensen Huang, Nvidia’s founder and CEO, said that “AI agents represent a new computing paradigm”. Mark Zuckerberg, Meta’s founder and CEO, said: “Muse is how we bring personal superintelligence to everyone.”

The $3 billion sits inside a $35 billion raise

Nvidia’s page gives no share price, share count or timing for its investment. It only points to the “anticipated completion and any applicable closing conditions” in its forward-looking statements, so the deal is not yet closed.

The offering itself is described in a Meta filing with the SEC dated October 26. Meta announced about $35 billion in concurrent underwritten offerings: roughly $22.5 billion of Class A common stock sold with forward sale agreements, and roughly $12.5 billion of depositary shares tied to a mandatory convertible preferred stock. The filing also mentions a planned at-the-market program of about $60 billion, which Meta has not yet agreed with anyone and does not expect to start before the first quarter of 2027.

The filing names Nvidia, AMD and Broadcom as cornerstone investors that have indicated interest in buying up to $5.0 billion of Class A stock in aggregate. Those indications are non-binding, and each investor may buy more, fewer or no shares. The filing does not split that figure by company, so it does not by itself confirm Nvidia’s $3 billion.

A supplier funds the customer that buys its chips

The economic logic is circular, and the filing shows why the money matters to Meta. It expects capital expenditures of $135 billion to $145 billion in 2026, against $97 billion of operating cash flow in the first three quarters and about $82 billion of cash and marketable securities at the end of the third quarter. Proceeds from the preferred stock go to general corporate purposes. For the Class A shares, Meta does not expect to receive proceeds until the forward sale agreements settle, on or before November 19, 2027.

Meta’s own growth figures in the filing give the demand side. It reports more than 10 million Muse users in the U.S. and Canada in the early weeks after launch, and says over 25% of weekly active users of the agent engage in shopping. Its Business Agent is used weekly by over 2 million businesses, with more than 30 million conversations a week. Meta also says data centers in 23 U.S. states are in operation or planned, with four superclusters of more than one gigawatt each.

For Nvidia, the arrangement ties a large customer’s spending plans to its newest platform. Neither company says how much Meta will spend on Vera Rubin, and the release names no chip counts or contract values. Whether the optimization work delivers the lower total cost of ownership the companies describe is something the announcement asserts but does not demonstrate.

The pattern of suppliers and lenders helping finance AI buildouts is not new. Our explainer on GPU-backed debt covers another route, in which chips and customer contracts serve as loan collateral.

What the announcements leave open

Several things remain unclear. Nvidia has not said when the $3 billion will be paid or at what price. The filing’s $5.0 billion indication from three companies is non-binding. The OpenShell and Sentry collaboration is only a possibility. No timeline is given for the Vera Rubin deployment, and no performance-per-watt numbers have been published. The Reuters report is the source for the framing of the news as a partnership on the Muse agent, while the financial terms come from the two companies’ own documents.