iTWire – HPE says ‘AI factory grids’ are the next phase of enterprise AI

HPE is advancing a new architectural concept it calls the “AI factory network,” a vision in which multiple AI data centers are interconnected to operate as a single distributed AI system.

The company maintains that while current AI factories focus on scaling up and scaling within a single facility, future demand will exceed the physical limits of power, cooling and space at any site. The answer, HPE He says, it’s climbing.

Using high-performance Ethernet fabrics and advanced routing, AI factories in different locations can be linked together so that workloads, models, and data can move seamlessly between sites.

HPE Positions NVIDIA Spectrum-X Ethernet as the foundation for this scalability layer, while Juniper routing platforms provide the long-distance multi-cloud connectivity needed to build metropolitan to regional AI networks.

The company describes its approach as “3-2-1 integration”: three silicon platforms, two suppliers, one…