AMD plans $1 billion Meta investment as Helios systems enter testing

AMD announced on 27 October that it would invest $1 billion in Meta’s equity offering, compared with Nvidia’s planned $3 billion investment in the same sale, Stocktwits reported. AMD also said in its 27 October announcement that Meta had installed initial Helios rackscale AI systems and was testing them ahead of a broader deployment.

AMD’s $1 billion investment in Meta’s offering

Meta increased its Class A common-stock offering to $25 billion from $22.5 billion and priced the shares at $625 each, Stocktwits reported. AMD’s planned purchase puts the chipmaker’s capital alongside its work supplying processors, GPUs, networking equipment and software for Meta’s infrastructure. The companies have also aligned engineering decisions around Meta’s workloads, according to AMD.

Meta also priced 250 million depositary shares representing mandatory convertible preferred stock at $50 each. Stocktwits put the two offerings together at approximately $37.5 billion. Meta forecasts capital expenditure of $135 billion to $145 billion in 2026 as it finances AI infrastructure.

Nvidia announced its planned $3 billion investment on 26 October as it expanded collaboration with Meta on Muse AI models. Meta had previously identified Nvidia, AMD and Broadcom as prospective purchasers of up to $5 billion of Class A shares collectively, with those indications described as nonbinding. AMD’s announcement sets out its own $1 billion investment separately from those earlier indications.

AI Affairs previously reported Meta’s $35 billion equity offering and plans for another $60 billion. Meta expects a further share-sale programme of up to $60 billion to begin in the first quarter of 2027.

Meta tests Helios alongside installed EPYC processors

The Helios racks at Meta are in testing rather than the broader deployment AMD anticipates. AMD says its relationship with Meta already spans several generations of EPYC CPUs: Meta has deployed millions of the processors, and EPYC chips support the dedicated computing environments behind its Muse personal AI agent.

Meta is a lead customer for 6th Gen EPYC CPUs, according to AMD. The chipmaker also says the companies are collaborating on later processors, including a generation called Verano. Those projects extend a supply relationship in which AMD’s existing CPUs have a role in Meta’s operating infrastructure, while Helios remains at the installation-and-testing stage.

The companies’ plans build on a February agreement under which Meta intends to deploy up to 6 gigawatts of AMD GPUs across multiple generations, Stocktwits reported. That figure is a planned ceiling across generations, while AMD describes the initial Helios systems as installed and under test. The scale of the February plan places the rack testing within a longer hardware rollout.

Instinct GPUs support Meta’s production workloads

AMD says its Instinct GPUs are already used in Meta’s production infrastructure for training, ranking and recommendation engines. Meta also helped define AMD’s MI450 architecture at an early stage, according to the chipmaker. AMD says Meta will deploy a custom Instinct GPU designed around its workloads, while the collaboration is informing the MI500 series and later rackscale systems.

On the software side, AMD says the companies have expanded their work to large-scale AI training and are optimising recommendation models on MI350X GPUs. They have also worked on ROCm through PyTorch and contributed changes to PyTorch, Triton and ROCm for other developers, according to AMD.

AMD says it delivered support at release for Meta’s Muse Spark Open models on Instinct GPUs. It also provided that support for Muse Glimmer on Ryzen AI PCs and Radeon GPUs.