Leveling Up: The AI HW/SW Co-Design Group Tackles the “System of Systems” at OCP Global Summit 2025

Date of Post
September 10, 2025
Posted by
OCP AI HW/SW Co-Design Workgroup

The Open Compute Project (OCP) ecosystem is built on innovation and evolution. In a significant milestone for our workstream, the AI HW/SW Co-Design group has been promoted from the Future Technologies Initiative (FTI) to the official OCP Server Project. This move reflects the maturation of our efforts and the critical importance of co-design in building the next generation of AI infrastructure.

Building on last year’s framework for Polymorphic Architecture, our focus has expanded to tackle the challenge of integration at scale. This year, we’ve begun modeling these complex, heterogeneous environments using AstraSim, with a specific focus on networks, topologies, and fabrics. Our research has also broadened to include emerging fields like Neuromorphic Computing, the integration of Chiplets, and the development of AI fabrics that span from individual chips to entire data centers.

We are excited to share our progress at this year’s OCP Global Summit across three distinct sessions, offering a comprehensive view of our work.

Diving Deep: The “System of Systems” in the Server Breakout

A future AI compute environment is more than just a collection of powerful accelerators; it’s  hierarchically composable yet intelligent enough to dynamically transform itself and adapt to the AI application’s behavior to achieve the best performance and resource efficiency. In the OCP Server Breakout Session, we will present our first order incarnation of Polymorphic Architecture, a “System of Systems”. We will deep dive into this crucial concept and cover the full journey from theory to implementation:

  • An Overview: Defining the “System of Systems” concept and why it’s essential for future AI infrastructure. Samantika Sury, Fellow at HPE, will describe this in detail in “Enabling macroheterogeneity through a “system-of-systems” approach”
  • Modeling the Future: Demonstrating how we use AstraSim to model and analyze these complex interactions will be covered with a Lightning Talk by Tushar Krishna, Professor at Georgia Tech.
  • A Practical Example: Walking through a specific model of a compute environment to showcase the methodology in action Jascha Achterberg, a Research Fellow at Oxford, will detail the “Coordination and workload algorithms to unlock the potential of polymorphic ‘System-of-Systems’ architectures for large scale AI”
  • The Path to Standardization: Outlining our roadmap for creating a standardized schema to enable a truly composable and interoperable ecosystem will be covered in a lightning talk by Kingshuk Mandal from Keysight. Kingshuk will highlight the proposed standardization of a schema, via infra-graph schema, for modeling the AI Data center.

This session will provide a foundational understanding of how to architect, model, and standardize the heterogeneous data centers.

Weaving the Future: AI Fabrics in the AI Track

 The performance of AI workloads is increasingly defined by the quality of the connections between components. In the AI Track, Jai Kumar, Distinguished Engineer at Broadcom with 20+ patents in his portfolio, will present on “Scale Up and Scale Out AI Fabrics: A Polymorphic Ethernet Architecture for ‘Systems of System’”

This talk will explore the critical communication pathways that enable AI systems to grow efficiently. Jai will discuss the challenges and innovations in creating fabrics that seamlessly connect resources within a system, rack (scale-up), pod, and across thousands of nodes in a data center (scale-out). With this you will understand the competing and conflicting requirements of scale-up and scale-out AI fabrics, and how to converge them using Ethernet technology to create a robust system of systems.

Join Us at the Summit!

The promotion to the Server Project marks a new chapter for the AI HW/SW Co-Design working group. Our work has evolved from future-facing initiatives to developing the tangible standards and models that will define next-generation AI systems. We invite you to join our sessions to learn, discuss, and collaborate with us on this exciting journey.

Find us at the following sessions:

  • OCP Server Breakout Session: “A ‘System of Systems’ Approach to AI Infrastructure”, October 15th at 8am.
  • AI Track: “Scale Up and Scale Out AI Fabrics: A Polymorphic Ethernet Architecture for ‘Systems of System’”, October 15th at 12:30pm
  • Future Technology Symposium: Continuing the conversation on forward-looking AI research and technologies.

We look forward to seeing you at the OCP Global Summit and collaborating to build the future of open AI hardware and software together!