Uploaded July 2026 | Updated September 2026, 2 weeks ago
In this fireside chat, Alan Weckel from 650 Group and NVIDIA’s SVP of Networking, Gilad Shainer, discuss how networking serves as the heart of modern AI factories. The conversation explores the shifting landscape of network architectures as the industry enters an 800 gig upgrade cycle and experiences explosive bandwidth growth. Shainer details four essential networking infrastructures for AI: scale-up via NVLink to create single GPU units, scale-out using Spectrum-X Ethernet to eliminate jitter and achieve 95% effective bandwidth, scale-across for giga-scale factories, and a new context memory storage infrastructure for inferencing. The discussion also highlights the production launch of co-packaged optics (CPO), a significant milestone that reduces optical network power consumption by nearly 5x while increasing reliability.
#AIifrastructure, #AIfactory, #acceleratednetworking
Learn more: nvidia.com/en-us/networking
In this fireside chat, Alan Weckel from 650 Group and NVIDIA’s SVP of Networking, Gilad Shainer, discuss how networking serves as the heart of modern AI factories. The conversation explores the shifting landscape of network architectures as the industry enters an 800 gig upgrade cycle and experiences explosive bandwidth growth. Shainer details four essential networking infrastructures for AI: scale-up via NVLink to create single GPU units, scale-out using Spectrum-X Ethernet to eliminate jitter and achieve 95% effective bandwidth, scale-across for giga-scale factories, and a new context memory storage infrastructure for inferencing. The discussion also highlights the production launch of co-packaged optics (CPO), a significant milestone that reduces optical network power consumption by nearly 5x while increasing reliability.
#AIifrastructure, #AIfactory, #acceleratednetworking
Learn more: nvidia.com/en-us/networking
