Nvidia and AMD plan $4 billion Meta investment alongside expanded AI work

Nvidia said on October 27 that it will invest $3 billion in Meta’s equity offering. AMD announced a $1 billion investment the same day, bringing the combined total to $4 billion. Nvidia also outlined further infrastructure deployment plans for Meta’s Muse models and agents. AMD reported that initial Helios systems have been installed for testing.

The announcements specify individual amounts after Meta’s October 26 financing launch. Meta had said Nvidia, AMD and Broadcom expressed nonbinding interest in purchasing up to $5 billion collectively. Nvidia and AMD have now published their own planned investments; the announcements give no basis for assigning the remaining difference to Broadcom.

Nvidia and Meta plan to improve training and inference performance and the overall cost of operating Muse. Meta intends to deploy more Vera Rubin systems and Vera CPUs, alongside networking equipment. OpenShell for safer agent deployment and Sentry for monitoring behavior remain possible additional collaborations. The release describes performance and efficiency objectives, without new measurements establishing the gains.

AMD said initial Helios racks are installed for tests. EPYC processors support Muse; Instinct GPUs handle training, ranking and recommendations. A customized Instinct GPU is planned.

Adapting software is part of that work. A joint Meta–AMD paper, published October 15 and linked by AMD, examines recommendation-model training on MI350X GPUs. Researchers adjusted model structures and low-level computation code, choosing implementations suited to the hardware. They report up to 1.5 times the end-to-end training throughput of a directly ported baseline. That comparison measures optimization within the AMD implementation, rather than a 1.5-times advantage over an Nvidia product. Throughput measures how much training data the system processes over time. Contributions were upstreamed to PyTorch, Triton and ROCm.

The financing follows rising infrastructure spending. Meta’s October 21 results put third-quarter capital expenditure, including finance-lease principal, at $34.91 billion. Operating cash flow was $32.79 billion and free cash flow was negative $2.12 billion. The company projected $135 billion to $145 billion in capital expenditure for 2026. The lower bound rose from $130 billion. Cash, cash equivalents and marketable securities totaled $82.45 billion at September’s end.

Offering close and cash receipt have different timelines

Meta also increased its common-stock offering from $22.5 billion to $25 billion, pricing 40 million shares at $625 each. Together with $12.5 billion in depositary shares representing preferred stock, the two offerings total $37.5 billion before additional purchase options.

Both offerings are expected to close October 29, subject to conditions. For the common-stock deal, counterparties initially sell borrowed shares; Meta expects no proceeds at that stage. The company expects forward-contract settlement by November 19, 2027. Actual proceeds depend on price adjustments and settlement terms, so closing the offering and receiving the cash occur at different stages.

A separate proposed $60 billion at-the-market program depends on the start of Class C share trading. Meta’s October 26 launch statement says it has entered no agreement for that program and its timing, size and terms remain undetermined. The company describes a financing mix of operating cash flow, investment-grade debt, strategic partnerships and equity capital to support AI infrastructure.