NVIDIA plans to buy into a major customer. In a press release dated October 26, the chipmaker said it will invest $3 billion in Meta's newly launched equity offering and broaden its engineering work on Muse, the personal AI agent and model family at the center of Meta's consumer AI push. The announcement arrived on the same day Meta unveiled a stock-sale plan that could bring in as much as $95 billion to pay for AI infrastructure.
On the technical side, the two companies describe a deeper joint effort to tune Muse workloads, covering both inference and training, with the stated goal of better performance and a lower total cost of ownership. Meta intends to put more NVIDIA hardware into service for its next generation of agents, including Vera Rubin systems and Vera CPUs, along with Spectrum-X Ethernet networking and BlueField-4 data processing units. The companies also say they are weighing further work around NVIDIA's OpenShell Reference Architecture, which bundles OpenShell for safer agent deployment with NVIDIA Sentry for watching how agents behave. That part is framed as something under consideration, not a signed commitment.
Meta's side of the story is spelled out in its own investor release. The company launched two concurrent underwritten offerings worth roughly $35 billion in total: about $22.5 billion in Class A common shares sold through forward sale agreements with Morgan Stanley and Goldman Sachs, and about $12.5 billion of depositary shares tied to a new Series A mandatory convertible preferred stock that is scheduled to convert around August 15, 2029. A further at-the-market program of about $60 billion in Class C shares is planned once that new non-voting stock class, announced the previous week, starts trading. Meta says this second step is not expected before the first quarter of 2027 and that its size, timing and terms are not yet fixed.
The two releases describe NVIDIA's role in slightly different terms. NVIDIA states plainly that it will invest $3 billion. Meta's filing groups NVIDIA with AMD and Broadcom as cornerstone investors that have, separately, indicated interest in buying up to $5 billion of Class A shares combined, on the same terms as other buyers. Meta stresses that these indications are not binding, so the final allocations could be larger, smaller or zero.
The money is meant to feed an unusually heavy spending program. Meta now expects capital expenditures of $135 billion to $145 billion in 2026 and says it is building four superclusters of more than a gigawatt each while nearly doubling its total infrastructure capacity this year. It reported about $97 billion of operating cash flow over the first three quarters and roughly $82 billion in cash and marketable securities at the end of September. Meta also cites traction for Muse, saying more than 10 million people in the United States and Canada signed up in the first weeks and that shopping accounts for the largest share of weekly activity.
CNBC reported that Meta shares slipped about 1.6% in early trading after the announcement. Coverage noted that investors had focused on Meta's negative free cash flow and higher spending forecast when it reported results the previous week. The same report placed the deal alongside other large AI financings this year, including SpaceX's GPU financing and Anthropic's move toward a public listing.
For the AI hardware market, the arrangement tightens an already close loop: a leading chip supplier is putting cash into a customer that is committing to buy more of its systems. Several pieces remain open. The offerings had not been priced in the materials reviewed, the cornerstone purchases are indications rather than contracts, and the OpenShell and Sentry work is still exploratory. The performance and adoption figures for Muse come from Meta and NVIDIA and have not been independently measured.