I ripped this off Jose Pozo repost on Linkedin from his colleague Olga Raz's post.

What I can share is a broad view on the main themes. The tone was set by the first two speakers: Nicolaas Viljoen from Meta kicked things off with a no-nonsense breakdown of the massive cost of cluster downtime, and how, at least for the next 2-5 years, latency is not as critical as we might think. Their key point? Reliability and power efficiency are going to be essential for wider AI adoption of optical technology. That message was echoed by several speakers throughout the day.

NVIDIA added a third piece to the puzzle: cost. Roughly quoting M. Ashkan Seyedi, “I’ll give you a chance if your solution is 2-3X the cost of current tech, but not if it’s 10X.”

In the end, it’s clear that any solution meeting all three criteria—reliability, power efficiency, and an acceptable price point—will win, whether for specific applications or across the entire connectivity ecosystem.

My take is Nvidia is driving the need for speed and the hyper scalers want no drama in the optics solutions (failures) and less power is also a must.

https://www.linkedin.com/posts/olga-raz-b78b8227a_peccsummit2024-datacenter-opticalnetworking-activity-7255517243986354177-ZKMH?utm_source=share&utm_medium=member_ios