Martin Casado posted something that made me start writing. “Nearly every board meeting: ‘Hiring strong infra folks is incredibly hard right now.’”
I know exactly where those folks went. I used to be one of them.
The Guy Who Spends All the Money
At Cloudera, I built out our datacenter. This was before AWS was a real option for what we needed. Hadoop was so disk-intensive that cloud instances couldn’t touch bare metal. Rack locality mattered. Drive count mattered. The physical infrastructure was the product differentiator, the thing that let us test massive clusters and compete.
When the company hired a new VP of Finance, I got introduced as “Chad, the guy who spends all the money.”
Not Chad, the guy building the infrastructure that makes the product possible. The guy who spends all the money. That’s what being an infra engineer felt like. You were a cost center. The work was essential and the people doing it were expendable.
I moved to WibiData and found myself doing the same job with the same pain points. Same budget conversations, same invisible work. I was burning out. That’s when I left infra engineering for good. A friend at Puppet offered me a sales engineering role, and I took it. My father was a salesman my whole life, literally raised me for it. Turns out I was good at it, and I liked helping people solve problems more than I liked being the person whose problems nobody saw. Docker sealed the transition. I built a team, grew from IC to director to VP. I never went back.
How We Got Here
The industry spent fifteen years systematically eliminating infrastructure specialists. DevOps started as a genuinely good idea. Break down the wall between development and operations. Ship faster by collaborating better. The original vision, the Alspaw-era thinking, was about empathy across disciplines. Developers should understand operations. Operations should understand development. Both get better.
That’s not what happened. What happened was consolidation disguised as culture. “DevOps” became “make the developers do ops.” Then “full-stack engineer” became the hiring goal. Why pay for a database specialist, an infrastructure engineer, and a developer when you can hire one person who does all three?
Except those specialties are whole careers. You don’t compress twenty years of datacenter experience into a bullet point on a job description. You don’t replace a database engineer with someone who took a Postgres tutorial. You get what we have now: jacks of all trades who are pretty good at some things and not great at the ones that matter most when something breaks at 3am.
The companies that understood this, the Googles and AWSes, the ones still racking servers, they kept their specialists. Everyone else let them go or never hired them in the first place. And now every board meeting is the same conversation: we can’t find infra people.
The Phone Calls
Last year I got recruiting calls from AI companies that wanted to build datacenters. They’d heard I’d done it for Cloudera and they wanted me to do it again.
I passed.
The conversations told me everything I needed to know. They valued the expertise in the abstract. They knew they needed someone who’d done it before. The vibe was clear, though: they expected this to be fast and straightforward, like spinning up a Kubernetes cluster. Building a datacenter takes years, specialized people, and those people now work at cloud companies that pay extremely well. Getting them to leave for a startup that doesn’t fully grasp what it’s asking for is a hard sell.
This is the Space Cowboys scenario. In that movie, NASA has to pull retired test pilots out of retirement because they’re the only ones who understand the guidance system on an old Soviet satellite. The agency spent decades moving past those engineers, making them obsolete on paper. Then they hit a problem that only the old expertise could solve.
That’s where infra engineering is right now. Some companies get it. I’m sure OpenAI and Anthropic understand what datacenter buildouts actually require. Plenty of others are learning the hard way that you can’t just hire this expertise back on demand after the industry spent years driving it away.
The Next Space Cowboys
I’ve seen this movie. I know how it ends. And watching AI tools reshape software engineering right now feels like watching a car start to hydroplane in slow motion. You can almost hear it, that long slow “ohhhhh shiiiiit” as the back end starts to swing. I’m not in the car. I’m watching from the shoulder with popcorn. And I can see where it’s heading.
The playbook is the same one that gutted infra. AI tools are getting good. I use them every day. I run an AI company. This isn’t an anti-AI argument. Companies are starting to believe that AI tools eliminate the need for engineers who deeply understand their systems. Juniors aren’t building foundational skills because the tools shortcut everything. Organizations are losing institutional knowledge one “efficiency” at a time.
The logic sounds identical to the DevOps era. Why hire specialists when the tools can cover it? Why invest in deep expertise when a generalist with AI assistance can ship features? Fewer people, same output. We know how this story ends because we just lived through it. We spent fifteen years making infra engineers extinct and now every board in the country is asking where they all went.
DoorDash is a tool. It doesn’t mean you stop knowing how to cook.
I don’t want “people who can actually write code” to be the next recruiting crisis. I don’t want the industry to spend ten years letting tools do the understanding, then act surprised when nobody understands their own systems. Organizations need people who deeply know what they’re running. Not people who can describe what they want to a prompt and hope the output is right.
We did this to ourselves with infrastructure. We don’t have to do it again.