Meta already has the first AMD Helios rack systems installed and in testing, and AMD will invest $1 billion in Meta’s stock offering, AMD announced on October 27, 2026. The agreement expands a partnership that already includes millions of EPYC processors and Instinct GPUs in production, and adds a custom Instinct GPU for Meta and a future EPYC processor codenamed Verano.
The new phase of the AMD-Meta alliance in 20 seconds
- Meta is already testing the first AMD Helios racks and plans to expand them soon, with no specific date.
- AMD will invest $1 billion in Meta’s stock offering.
- Meta will get a custom Instinct GPU and is a lead customer for sixth-generation EPYC.
- Neither company gives gigawatt figures or infrastructure investment amounts.
AMD’s press release comes eight months after the two companies signed, in February, a deal for up to 6 gigawatts to deploy Instinct GPUs in Meta’s AI infrastructure, with Helios as the first platform. AMD now confirms those systems are already on Meta’s premises in the testing phase and that their broader rollout will arrive, in the company’s words, in the near future.
Helios, a custom GPU, and a new EPYC generation
Helios is AMD’s full AI rack system, combining Instinct GPUs, EPYC processors and Pensando networking, which Revista Cloud compared with its rival’s approach in NVIDIA Vera Rubin vs. AMD Helios. Meta was one of the customers that helped define the MI450 GPU’s architecture and will deploy a custom version of Instinct optimized for its workloads. According to AMD, that joint work is also shaping the future MI500 series and upcoming generations of rack systems.
Instinct GPUs are already running in production at Meta for training and for the ranking and recommendation systems that order content across its apps. On processors, Meta has deployed millions of EPYC chips across several generations, which among other things power the infrastructure behind its personal AI agent Muse, is a lead customer for the sixth generation, and is collaborating with AMD on a future generation called Verano.
The software side is the least visible part and one of the most important for AMD, whose ROCm platform competes with NVIDIA’s CUDA. The companies are improving ROCm through PyTorch, Triton and AMD’s own libraries, optimizing recommendation models on Instinct MI350X GPUs, and AMD gave day-one support to Meta’s open Muse Spark models on Instinct GPUs and to Muse Glimmer on PCs with Ryzen AI and Radeon graphics.
An AMD investment in Meta, and no rollout figures
The financial news is that AMD will invest $1 billion in Meta’s stock offering. The release gives no further detail on that deal, nor any figures on gigawatts, rollout timeline, or the amount of infrastructure tied to this new phase. It talks about building and deploying integrated AI systems at gigawatt scale, without specifics.
AMD chair and CEO Lisa Su said that with Helios now running at Meta, the partnership moves to a new scale in what the company calls the age of superintelligence. Meta founder and CEO Mark Zuckerberg described AMD as a key partner in building the compute capacity that runs its frontier models. As is usual in these kinds of announcements, AMD notes that forward-looking statements about future deployments and products are subject to risks and uncertainties.
For the market, the message is that Meta keeps a multi-vendor accelerator strategy, and that AMD keeps gaining ground with one of the largest buyers of AI infrastructure, even though the specific figures for this phase remain unknown.
Frequently asked questions
What have AMD and Meta announced?
That Meta is already testing the first AMD Helios racks, will get a custom Instinct GPU, and that AMD will invest $1 billion in Meta’s stock offering.
When will Helios roll out at Meta?
The first systems are already installed and being tested. AMD talks about a broader rollout in the near future, with no specific date.
What is Verano?
The codename for a future generation of AMD EPYC processors that AMD and Meta are collaborating on.
Is this related to the 6-gigawatt deal?
Yes. It expands the alliance the two companies signed in February 2026 to deploy up to 6 gigawatts of Instinct GPUs, though the new release gives no capacity figures.
Sources:
