AMD installs first Helios server systems in Meta data centers

AMD has announced the installation of its first Helios server systems in the data centers of its major partner, Meta, initiating preliminary testing phases. The new platform integrates compute components, networking gear, and other infrastructure elements into a single rack unit. This deployment represents the latest milestone in a long-term strategic agreement that aims to roll out extensive infrastructure powered by AMD Instinct accelerators.

• AMD has installed the first Helios server racks for testing in Meta data centers.
• EPYC processors are already powering the infrastructure for the Muse AI agent.
• AMD plans to invest $1 billion in the partner's shares.
• Both companies are jointly preparing future chip generations, including Verano and MI450 accelerators.

Expanding Infrastructure and AI Workloads

As part of the partnership, AMD EPYC processors are heavily utilized in the compute infrastructure supporting the personal AI agent Muse. Meta has deployed millions of these processors, becoming one of the largest customers for sixth-generation EPYC chips. Furthermore, the companies are jointly developing future generations of silicon, including a processor under the codename Verano, whose specifications and launch timeline remain undisclosed.

Instinct accelerators are already handling production workloads, including model training, ranking, and recommendation generation. Meta participated in the early design phase of the MI450 architecture and is set to receive a customized version tailored to its specific workloads. Future plans also mention the MI500 series accelerators, integration of the ROCm software platform with PyTorch, and optimization of recommendation models on MI350X accelerators.

Financial Investments and Future Outlook

In addition to hardware deployments, AMD intends to invest $1 billion in the partner's stock offering. The companies have not disclosed the business valuation, the size of the acquired stake, or the expected closing date of the transaction. These financial investments are separate from hardware supply agreements and highlight the depth of the strategic alliance.

AMD CEO Lisa Su and Mark Цукерберг highlighted the significance of scaling computational power for advanced AI models. While the ultimate goal is to deploy gigawatt-scale infrastructure, the companies have not yet specified the exact deployment timeline or volume for the Helios systems and upcoming processor lines.