By Enterprises comparing self-hosted infrastructure costs against API pricing almost always anchor the calculation to historical execution patterns. Those patterns assume staffing levels, procurement cycles, and operational maturity that no longer reflect reality. When ops teams are overloaded, hardware onboarding takes months, and active migration projects are already consuming every available resource, the cost model is broken from the first cell in the spreadsheet.
In this interview on TFiR, Rob Hirschfeld, CEO at RackN, breaks down what enterprises consistently get wrong when modeling the true cost of self-hosted versus API infrastructure, and how operational process gaps become the primary accelerant for AI-driven technical debt.
Guest: Rob Hirschfeld, CEO at RackN
Show: TFiR
Here is what every platform engineer and infrastructure operator needs to know.
Technical Deep Dive
Q: What do enterprises get wrong when calculating the true cost of self-hosted infrastructure versus API pricing?
Rob Hirschfeld, CEO at RackN, explains that most enterprises anchor their cost models to historical execution patterns, specifically how long things have taken in the past and what resources were available. That baseline is no longer valid. Teams are already overworked, hardware procurement takes months, and the operators running those servers do not have capacity to absorb new workloads without a fundamental change to how infrastructure is delivered.
“A lot of times they’re looking at the way they’ve done it in the past and what it’s taken to get things done.” — Rob Hirschfeld, CEO, RackN
Q: Why does an already-overloaded ops team make AI infrastructure projects fail before they start?
Hirschfeld points to a specific execution trap: teams already behind on critical projects like VMware migration evaluations cannot be redirected to AI infrastructure work without dropping something essential. The workload does not shrink because a new priority appears. If the organization is not actively improving how it executes infrastructure projects, adding AI requirements on top of an already-strained team guarantees failure or delay.
“You can’t just take somebody who’s already struggling to do the VMware Migration Evaluation Project and say, finish that sooner, I need you to do AI.” — Rob Hirschfeld, CEO, RackN
Q: What operational improvements must happen before an enterprise can successfully run AI on self-hosted infrastructure?
Hirschfeld outlines a set of specific operational capabilities that must improve simultaneously: hardware flexibility, hardware currency, onboarding cycle time reduction, OS and platform flexibility, and the ratio of machines each operator can manage. All of these need to move faster than they historically have. The breadth of infrastructure each person in the organization can manage must increase significantly before AI workloads can be reliably delivered.
“The breadth and span, the number of machines that each person in your organization, each operator is running, also has to improve significantly.” — Rob Hirschfeld, CEO, RackN
Q: Can AI tools fix infrastructure process gaps, or do those gaps need to be solved first?
Hirschfeld is direct on this point: AI will not accelerate teams through infrastructure process gaps they have not already addressed. Without controls, proven processes, and automation that works out of the box, AI tooling applied to a broken infrastructure practice creates more technical debt, not less. It produces outputs that appear correct but implement unsustainable patterns that compound over time.
“AI is not going to help them get through that process faster. As a matter of fact, it can create more technical debt and crises by doing things that look okay but are actually wrong or not sustainable patterns.” — Rob Hirschfeld, CEO, RackN
Q: How does RackN address the infrastructure execution gap for enterprises deploying AI?
Hirschfeld describes RackN’s approach as delivering built-in process capabilities and proven operational patterns out of the box. RackN handles bare metal lifecycle, conformance and compliance, OS management, and platform installs as prebuilt capabilities rather than custom work. This means RackN customers redirect their team’s focus entirely toward delivering AI or virtualization value instead of reinventing infrastructure execution patterns for each new project.
“They show up with built-in process capabilities, proven aspects, and as they work with us, we help them stay in front of the bare metal pieces, the conformance compliance, operating systems, the platform installs.” — Rob Hirschfeld, CEO, RackN
Q: What is the core competitive requirement for any enterprise trying to execute AI infrastructure projects today?
Hirschfeld frames execution speed as the non-negotiable competitive requirement. Enterprises must reach results much faster than historical norms allowed, and they must do so without inventing new architectures, new tools, or new processes from scratch on every project. The organizations that arrive with proven, reusable infrastructure execution capability will outpace those still building the foundation while trying to deliver AI outcomes at the same time.
“You have to get to the result much, much faster than we’ve historically been able to.” — Rob Hirschfeld, CEO, RackN
Resources & Documentation
- RackN, bare metal lifecycle automation platform with built-in process capabilities for OS management, conformance, compliance, and platform installs
***
👇 Click to Read Full Raw Transcript
Swapnil Bhartiya: Let’s talk about cost, because that is one of the biggest factors. How do you help enterprises model out the true cost of self hosted vs API pricing? On the surface, things might look cheap. What do people get wrong in that calculation most of the time?
Rob Hirschfeld: A lot of times they’re looking at the way they’ve done it in the past and what it’s taken to get things done. And so if you’re thinking about buying servers and it takes six months and it takes a long time to get them racked or you don’t have the people or the expertise or, or the people that you have running those servers are already behind, they’re overworked, you don’t have enough of them and you’re not going to be hiring more. You really do have to reconsider how you’re running and delivering your infrastructure processes to leapfrog where you are. You can’t just take somebody who’s already struggling to do the VMware Migration Evaluation Project and help say, all right, finish that sooner, move, you know, I need you to do AI or they can’t get off of that because that’s really important to do AI. The challenge is we need to be working, and this is exactly rack and spread and butter. The industry as a whole needs to be working to dramatically improve how we execute infrastructure projects. That means being more flexible on the hardware. That means keeping the hardware up to date. It means improving your cycle time dramatically, reducing your onboarding time, being more flexible about which operating systems and platforms you run. All of that needs to go up faster. And the breadth and span, a number of machines that each person in your organization, each operator is running, also has to improve significantly. Now AI is going to help with that, there’s no doubt about that. But if they don’t have controls and processes and things that work out of the box, that’s, you know, AI is not going to help them get through that process faster. As a matter of fact, it can create more technical debt and crises by doing things that look okay but are actually wrong or not sustainable patterns. This is where we see a really significant advantage for the rack and customers. Because they show up with built in process capabilities, proven aspects, and as they work with us, we help them stay in front of the bare metal pieces, the conformance compliance, operating systems, the platform installs. All of those things are built in the box for us, which means that our customers then focus on delivering AI value or virtualization value. They don’t have to worry as much on the other piece. And from a competitive perspective. I see that as absolutely essential for everybody that we’re talking about here. They have to be able to execute projects without inventing new ways to do things or new architectures or new tools. You have to get to the result much, much faster than we’ve historically been able. To.
