Everyone is talking about bigger AI clusters. What happens when those clusters need to span multiple data centers? At Cisco Live, I sat down with Rakesh Chopra, SVP & Fellow, Common Hardware Group at Cisco on The Ravit Show, to discuss one of the less talked about challenges in AI infrastructure: scaling AI beyond a single data center.
A few key themes from our conversation:
-- The industry is moving from scale-up and scale-out to scale-across architectures
-- Connecting GPUs across data centers is becoming a critical challenge as organizations build larger AI environments
-- Power efficiency is now as important as raw performance, driving innovation in silicon and optics
-- Network reliability and low-latency communication are essential as AI clusters stretch across geographic boundaries
-- Co-designing networking, silicon, and optics is becoming a requirement rather than an optimization
The AI conversation often focuses on models.
But the real story may be the infrastructure required to make those models work at scale.
Podden och tillhörande omslagsbild på den här sidan tillhör
Ravit Jain. Innehållet i podden är skapat av Ravit Jain och inte av,
eller tillsammans med, Poddtoppen.
