Powering Performance for Neoclouds

WEKA, WhiteFiber, and QumulusAI on building for the GPU era

You bought the GPUs. You energized the hall. Now your fleet is waiting on data and idle accelerators don’t stop the invoice.

Val Bercovici of WEKA sits down with Michael Francisco of WhiteFiber and Ryan DiRocco of QumulusAI to talk about the part of a buildout that rarely makes the press release: what broke first, what they’d do differently, and how you light up a cluster when the storage purchase order hasn’t cleared.

Good infrastructure is modular. It scales a piece at a time, and it never makes you tear down what you’ve already built to make it bigger.

What we’ll cover

• Performance — Small files and mixed IO on one cluster, from kdb+ to training, reinforcement learning, inference, and agent swarms
• Consistency — Predictable latency and throughput
• Scale — Hundreds of TBs to hundreds of PBs in a single namespace, without disruption
• Resiliency — A fully redundant data path your GPU fleet never loses

Attend live to receive a $300 LEGO gift card as our thanks so you can go build something that works the same way. One per attendee.

Meet Our Speakers

Val Bercovici

Chief AI Officer, WEKA

Michael Francisco

Vice President, WhiteFiber

Ryan DiRocco

Chief Technology Officer, QumulusAI