Guest: Rob Hirschfeld (LinkedIn)
Company: RackN
Show Name: An Eye on AI
Topic: AI Infrastructure

AI infrastructure comes with staggering cost and complexity. Every idle GPU represents wasted potential and lost revenue. In this clip from his conversation with Swapnil Bhartiya, Rob Hirschfeld, CEO and Co-Founder of RackN, reveals how his team helped a major service provider transform its AI operations through automation — reducing cluster reset times from an entire week to just 90 minutes.

Each 64-machine AI training cluster represented a quarter-billion-dollar investment. These clusters had to be reset, wiped, patched, and reconfigured between training runs to ensure data isolation and system integrity. Before automation, the manual reset process meant a full week of downtime per cluster — equating to roughly $150,000 of idle depreciation for every reset.

Using Digital Rebar, RackN automated the entire sequence: from disk wipe and BIOS updates to network reassembly and OS deployment. In just 90 minutes, the system returned to a clean, certified, ready-to-use state. The impact was immediate — faster turnaround, lower operational overhead, and dramatically higher utilization of high-value infrastructure.

Hirschfeld explains that this transformation isn’t about exotic new hardware but about better processes. AI clusters are inherently complex — blending InfiniBand, fiber channel, and Ethernet networks, each with redundant paths and peer-to-peer traffic. The topology itself is sensitive; every node must be reconnected precisely to its position in the network fabric. RackN’s automation ensures that no data leaks, no misconfigurations occur, and every system comes back online exactly as intended.

The lesson extends far beyond this single deployment. Hirschfeld emphasizes that automation is the real key to AI scalability. When teams rely on manual configuration, they burn time on repetitive, error-prone tasks instead of innovation. By contrast, automated bare-metal operations turn infrastructure into a reusable, elastic resource.

One customer even reduced onboarding of new clusters to just four hours, fully validated and ready for training. In a market where GPU availability and training times define competitiveness, that speed becomes a decisive advantage.

As Hirschfeld puts it, “The biggest cost in AI infrastructure is lost opportunity.” RackN’s approach helps eliminate that loss, empowering organizations to use every GPU cycle productively.

AI is accelerating at an unprecedented pace, and infrastructure operations must evolve with it. RackN’s case study shows how automation isn’t just about convenience—it’s about enabling true operational efficiency at AI scale.