Meta and AMD are scaling up their partnership to build the computing infrastructure behind Meta's superintelligence ambitions, combining hardware, software and engineering development to support AI systems operating at gigawatt scale. The expanded agreement also includes a $1 billion investment by AMD in Meta's equity offering.
According to Frontier Enterprise, the companies are coordinating their technology roadmaps around Meta's existing and anticipated workloads. Their collaboration spans processors, graphics chips, networking and software, with the aim of supporting AI services used by billions of people. Initial AMD Helios systems are already installed at Meta and undergoing evaluation ahead of a wider rollout.
Helios enters Meta infrastructure
The deployment of Helios marks a significant expansion of the relationship. AMD's rack-scale platform combines computing hardware and supporting technologies into integrated systems designed for large AI workloads. Meta is testing its first installations as the partners prepare for broader deployment.
AMD CEO and chair Lisa Su said the companies had developed their relationship through years of technical work. Earlier cooperation included EPYC processors powering Meta infrastructure, alongside Instinct graphics processing units and open software used for AI computing.
Su said the Helios deployment takes the partnership into a new phase, with both companies developing infrastructure and future generations of silicon, software and systems for Meta's most ambitious superintelligence projects.
AMD chips power Muse
The partnership already reaches beyond experimental infrastructure. AMD EPYC central processing units (CPUs), which handle general computing tasks, support the dedicated computing environments behind Meta's Muse personal AI agent.
Meta has deployed millions of EPYC processors across several generations and is a major customer for AMD's sixth-generation EPYC products. The companies are also developing future processor generations, including a platform called Verano.
AMD's Instinct graphics processing units (GPUs), specialized chips designed to accelerate parallel computing, are also running in Meta production environments. They support AI model training, ranking systems and recommendation engines. Meta previously helped shape the MI450 GPU architecture and plans to deploy a customized Instinct processor tailored to its workloads.
Future hardware and software
The cooperation also influences AMD's longer-term product development. Future platforms include the Instinct MI500 series and next-generation rack-scale systems. These products are part of the companies' effort to align hardware development with the computing requirements of Meta's expanding AI operations.
Software remains another important part of the agreement. AMD and Meta have worked to improve ROCm, AMD's open software platform for GPU computing, through integration with PyTorch, a widely used framework for developing and training machine-learning models.
AMD has also provided Day-0 support for Meta's Muse Spark Open models on Instinct GPUs and Muse Glimmer on Ryzen AI-powered personal computers and Radeon graphics products. Day-0 support means software compatibility is available from a model's initial release rather than arriving through later updates.
The companies have extended their engineering work into large-scale AI training, including the optimization of recommendation models on AMD Instinct MI350X GPUs. Their developers have also contributed improvements to PyTorch, Triton and ROCm, making those changes available to the wider developer community.
Meta's growing investment
Meta founder and CEO Mark Zuckerberg described AMD as an important supplier of the computing capacity behind the company's frontier AI models. He said Meta was committing substantial resources to computing infrastructure intended to make personal superintelligence broadly accessible.
The financial component adds another dimension to the technical agreement. AMD's planned $1 billion equity investment gives the chipmaker a direct financial stake in its relationship with one of the world's largest technology companies. The source does not provide further details about the investment's structure or timing.
"AMD plays an important role in that."
— Mark Zuckerberg, Meta CEO
Why it matters
The expanded agreement illustrates how the pursuit of superintelligence is becoming a major infrastructure challenge. Developing increasingly capable AI models requires more than powerful individual chips: companies also need integrated computing systems, networking, software compatibility and reliable deployment at enormous scale.
For Meta, AMD offers an alternative source of computing hardware across several layers of its infrastructure, from general-purpose processors to specialized AI accelerators. For AMD, the relationship provides a significant opportunity to place its technologies inside a major AI platform and influence the hardware requirements of future workloads.
The immediate milestone is the testing of Helios systems already installed at Meta. The next development to watch is the broader deployment, alongside the customized MI450 hardware and future MI500 platforms. Those steps will show how quickly the companies can translate their technical cooperation into production infrastructure for Meta's superintelligence ambitions.