Nikhil Shetty
I'm a technical lead in AI infrastructure networking, specializing in cloud infrastructure, observability, and control planes. With more than 16 years of experience across startups, network switch vendors, and cloud infrastructure, I bring a practitioner's perspective to the design of cloud-scale platforms.
My current work centers on high-performance networking and accelerator interconnects for GPU clusters, where I lead initiatives ranging from tuning network performance for distributed training workloads to introducing novel network fabrics in the cloud.
I contribute to interconnect standards through OCP and the Ultra Ethernet Consortium, and I have a Ph.D. in EECS from UC Berkeley. Based in Santa Clara, California.
Work
What I've builtIntroducing new GPU systems and interconnect fabrics into a hyperscale cloud - NVL576 scale-up domains and InfiniBand at fleet scale - then tuning them for distributed training and building the telemetry that keeps them honest once the jobs are running.
High-Performance AI Systems
Engineering Hyperscale GPU Clusters
A book for engineers who have to make large GPU clusters work, written with Rik Kisnah and published by Packt. In progress.