By The most common reason enterprise AI deployments fail is not technology — it is the absence of repeatable process discipline and the organizational silos that prevent teams from building it.
The Guest: Rob Hirschfeld, CEO and Co-Founder at RackN
The Bottom Line:
- Platform teams and AIOps teams that brute-force bare metal provisioning without infrastructure expertise are the leading cause of stalled AI pilots, missed delivery objectives, and involuntary VMware renewals — and the fix is not more tooling, it is process discipline and the organizational will to build it cross-functionally before urgency makes it impossible.
Speaking with TFiR, Rob Hirschfeld of RackN catalogued the self-inflicted wounds he sees repeatedly across enterprise AI infrastructure programs — and issued a direct challenge to every executive overseeing an AI deployment.
THE MOST COMMON SELF-INFLICTED WOUND IN ENTERPRISE AI
Hirschfeld’s opening point is drawn from a pattern he has watched repeat across organizations of every size: platform teams — the people running Kubernetes, managing AIOps, and operating the model layer — assume that because they own the workload, they can also stand up the infrastructure beneath it. They cannot. Not reliably. Not at speed.
“The platform team—the people running the models, managing Kubernetes, and handling AIOps—often assume that, because they’re using the models, they’ll be able to run and set up the infrastructure.”
The consequence is predictable. The AIOps team or virtualization admin team brute-forces their way through bare metal provisioning, learning the infrastructure layer at the same time they are trying to validate a pilot. What should take weeks takes months. Management confidence in the team’s ability to execute erodes. Projects get shelved. And in the VMware migration context specifically, the failure to get bare metal provisioning right has directly caused enterprises to renew VMware contracts they were explicitly trying to exit.
“We’ve seen it result in many missed delivery objectives. We’ve also seen it lead to numerous VMware renewals, as well as AI initiatives and applications that have stalled.”
THE SILO PROBLEM THAT COMPOUNDS EVERYTHING
Hirschfeld identifies organizational siloing as the structural force that turns a technical skills gap into a systemic delivery failure. Operations teams and platform teams do not communicate well. Each team treats the other’s domain as someone else’s problem. Platform teams wall off the infrastructure layer and try to solve it in isolation — even when they already have access to RackN’s Digital Rebar and a proven operations team that could resolve the problem in days.
“I’ve watched this pattern repeat over and over again. Organizations are siloed. Operations teams are separate from platform teams, and they don’t communicate well. That lack of communication and collaboration, and the failure to think long term, slows projects down.”
The irony Hirschfeld points to is sharp: teams that wall off the infrastructure problem from the people best equipped to solve it are not protecting their autonomy, they are guaranteeing their own delay.
INFRASTRUCTURE AS CODE — APPLIED TO BARE METAL
The principle Hirschfeld frames as the foundation of AI deployment velocity is the same one that transformed application development: infrastructure as code. The goal is not to build a system and hope it never needs to change. The goal is to have a fully repeatable, source-driven process that can recreate the entire system from scratch, on demand, with confidence.
“The goal is to have a repeatable way to recreate the entire system and process. Investing the time to do that and iterate through it ultimately translates into greater velocity.”
Applied to bare metal AI infrastructure, this means every provisioning step — inventory qualification, networking topology configuration, OS deployment, cluster join — must be documented, automated, and validated through repeated execution before the system ever goes to production.
THE EXECUTIVE TEST
Hirschfeld closes with what may be the most actionable single piece of guidance in the clip: a direct test any executive can run to determine whether their team has a repeatable AI infrastructure process or simply a well-rehearsed demo.
“Ask your team: you got it working—now reset it to zero and show me the entire process from scratch again. If your team looks scared or nervous and doesn’t want to do that, then they don’t actually know how they achieved that success.”
A team that cannot rebuild from zero does not have a process. They have a fragile artifact they are afraid to touch. At the pace AI infrastructure is evolving — new hardware, new models, new OEM configurations arriving continuously — that fragility is not a temporary risk. It is an existential one.
“If you can’t replicate success, then you don’t actually know why you were successful—you just got lucky. At the pace AI is moving, luck is not enough.”
Watch the full TFiR interview with Rob Hirschfeld here
