AMD and Meta Partnership Expands with $1 Billion Investment in Superintelligence Infrastructure

AMD Commits $1 Billion to Meta's AI Infrastructure Push

AMD announced a significant expansion of its strategic partnership with Meta, committing $1 billion to Meta's equity offering while deploying advanced hardware and software solutions to power the social media giant's superintelligence infrastructure1

1

3

3

. The announcement sent AMD shares rallying 5.4% in Tuesday afternoon trading, positioning the chipmaker among top gainers in the semiconductor sector3

3

. This collaboration marks a pivotal moment as both companies work to build and deploy AI infrastructure at gigawatt scale, shaping future generations of silicon, software and systems to enable Meta's most ambitious superintelligence models and services1

1

.

Meta Deploys Millions of EPYC Processors as Lead Customer

Meta has deployed millions of AMD EPYC processors across multiple generations and serves as the lead customer for 6th Gen EPYC CPUs based on the Zen 6 core architecture2

2

4

4

. AMD EPYC processors now power the dedicated computing environments behind Meta's Muse personal AI agent, demonstrating the depth of integration between the two companies4

4

. The companies are collaborating on future processor generations including "Verano," specifically designed for Agentic AI workloads2

2

. This multi-year co-design roadmap signals sustained hyperscaler demand as competitors race to secure merchant silicon allocations3

3

.

Helios Rack-Scale Solutions Enter Testing Phase at Meta

Initial AMD Helios rackscale solutions are now installed and undergoing testing at Meta, with broader deployment planned for the near future4

4

5

5

. These Helios rack-scale solutions will anchor Meta's next-generation AI models and gigawatt-scale infrastructure, supporting frontier models such as the Muse AI Agent3

3

. Lisa Su, AMD's Chair and CEO, emphasized that "with Helios now running at Meta, we are taking that partnership to a new scale in the superintelligence era"1

1

. Mark Zuckerberg noted that AMD plays an important role in Meta's big investments in computing power that will enable personal superintelligence for everyone1

1

.

AMD Instinct GPUs Power Training and Recommendation Engines

AMD Instinct GPUs are currently in production at Meta for training, ranking and recommendation engines2

2

4

4

. Meta served as an early definition customer for the Instinct MI450 architecture and will deploy a custom AMD Instinct GPU optimized specifically for its workloads2

2

5

5

. The collaboration extends to future AMD platforms including Instinct MI500 series GPUs and next-gen rack-scale systems2

2

4

4

. Meta's technical paper details hardware-model co-design principles that enable optimization of recommendation models on AMD Instinct MI350X GPUs, delivering up to 1.5x improvement in end-to-end training throughput compared to directly ported baselines2

2

.

Day-0 Support for Muse Models Across AMD Hardware

AMD delivered Day-0 support for Meta's Muse Spark Open models on AMD Instinct GPUs and Muse Glimmer on AMD Ryzen AI PCs and AMD Radeon GPUs2

2

4

4

. The Muse Spark Open model, featuring 816B total parameters with 42B active MoE and 1M context, runs on a single 8x MI350X node with vLLM in multiple precision formats2

2

. Optimizations including decode-step work and expert parallelism boosted throughput from 206 to 350 output tokens per second at 4 concurrent requests and from 719 to 1,180 at 32 concurrent requests2

2

. At 128 concurrent requests, the system achieved 2,197 tokens per second while cutting time to first token by approximately 24%2

2

.

Software Collaboration Advances ROCm and PyTorch Ecosystem

AMD and Meta expanded their software collaboration to large-scale AI training, optimizing recommendation models on AMD Instinct MI350X GPUs and contributing improvements to PyTorch, Triton and ROCm AI software2

2

5

5

. These contributions have been upstreamed to open-source ecosystems, enabling other developers to benefit from the optimizations2

2

. The companies worked together to advance ROCm through PyTorch, ensuring developers can leverage AMD GPUs right out of the box4

4

. Meta's technical paper represents the first hyperscaler report on productionizing AMD GPUs for large-scale training, proposing three reusable principles including hardware-model co-design, analytical performance modeling, and increasingly mature agentic workflows2

2

.

Strategic Integration Spans Silicon to Pensando Networking

The AMD and Meta partnership extends beyond processors and GPUs to encompass AMD Pensando networking solutions, creating a comprehensive technology stack4

4

5

5

. AMD technology now powers Meta's infrastructure across compute, networking, systems and software, demonstrating deep integration across the entire stack4

4

. The companies have aligned product roadmaps and engineering decisions around current and future Meta workloads, evolving years of collaboration into building and deploying integrated AI systems at gigawatt scale4

4

. Meta deploys a heterogeneous fleet comprising NVIDIA GPUs, AMD GPUs, and proprietary MTIA accelerators to reduce costs and mitigate vendor lock-in2

2

. This multi-layered agreement signals that tier-one hyperscale customers continue to diversify their hardware suppliers while demanding performance optimization across different silicon architectures3

3

.