If there's one category of jobs artificial intelligence won't replace anytime soon, it's the trades, so perhaps it's fitting that a lack of plumbers, electricians, and construction workers is hampering the race to build data centers. As Wired reports, the U.S. faces a tradespeople shortage; the construction shortfall alone could eventually total nearly 400,000 workers, per McKinsey. Big Tech can't draw computing power from unfinished data centers, but skilled laborers are retiring with fewer replacements — because many have sent their own children to four-year colleges.
We Can't Build AI Because We Fired All the Plumbers
Silicon Valley has identified a pressing issue: you can't download an electrician. A recent Wired article sheds light on the urgent math behind this situation:
- The U.S. needs 550,000 skilled workers by 2032.
- Current shortage: over 10,000 workers.
- Data centers are projected to consume 8% of U.S. electricity by 2030.
- We are unable to build fast enough to meet these demands.
We have convinced an entire generation that working with their hands is beneath them, leading to situations where we are paying a lot of money for someone to install pipes. The AI revolution is currently on hold because we need professionals to fix the plumbing.
As with every revolution, some professions will see increases or decreases in demand. However, those that are increasing are not solely the high-tech roles.
AI is triggering the revenge of the blue-collar worker.
We keep framing the AI talent war as PhDs vs PhDs. But the real constraint on AI isn’t code: it’s power, pipes, cooling, and concrete.
No electricians, no plumbers, no HVAC techs = No data centers / No AI.
For decades, we told people success meant a white-collar job. AI is quietly flipping that script.
The irony is perfect: the future of artificial intelligence depends on the people who keep the lights on.
Curious how this reshapes careers, compensation, and workforce strategy over the next decade...?
Ten years ago, a 10kW rack was considered high-density. Today we're spec'ing 100kW+. And honestly? Most data centers aren't ready.
Here's the reality nobody talks about in keynotes: you can build the most powerful GPU in the world, but if you can't keep it cool and fed with power, it's just expensive silicon sitting in a warehouse.
I've watched teams celebrate landing that massive GPU order, only to realize six months later their facility can't actually deploy it. The math is brutal:
Traditional air cooling tops out around 40kW per rack
NVIDIA's Blackwell Ultra hits 140kW per rack
Their Rubin roadmap? We're looking at 250-300kW equivalent
This isn't "slightly more cooling." This is rethinking everything, electrical distribution, thermal management, and even the structural weight capacity of your floors. A single AI rack can weigh as much as a small car.
The infrastructure reality check:
🔌 Traditional air cooling: 40kW max (we're already past this)
🌊 Liquid cooling cost: $50K per rack (just for cooling infrastructure)
⚡ New facility cost: $200-300K per rack (to support 100kW+)
⏰ Infrastructure lag: 18-24 months behind silicon roadmaps
🏗️ Grid capacity: Moratoriums in some regions are blocking new data centers
And good luck finding plumbers who understand both fluid dynamics and GPU workload patterns.
What I find fascinating is the tension between speed and infrastructure readiness. Hyperscalers are racing to deploy AI capacity, but the facilities to support it are 18-24 months behind the silicon roadmap. We're essentially trying to run before we've finished building the track.
The winners in this cycle won't just be whoever has the best chips. It'll be whoever figures out the entire stack of silicon, cooling, power distribution, facility design, and can actually deploy it at scale without melting their data center or bankrupting themselves on infrastructure upgrades.
I'm genuinely curious how this plays out over the next two years. Are we going to see a wave of stranded GPU assets sitting in warehouses because nobody can deploy them? Or will innovation in cooling and power delivery catch up faster than I expect?
Either way, if you're only thinking about compute performance and ignoring watts per rack, you're planning for yesterday's infrastructure problem.
Everyone keeps saying AI will take all the jobs.
Funny thing is, look closely at a modern AI data center.
Every part of the picture below maps to very human skills that will be in demand for the foreseeable future:
• Steel, concrete, and shells: civil engineers, construction crews, project managers
• Power lines and substations: electricians, grid specialists, energy engineers
• Cooling pipes and liquid loops: HVAC experts, thermal engineers, plumbers
• Racks and GPUs: hardware technicians, systems engineers, integrators
• Fiber and networking gear: network engineers, cabling specialists
• Operations floors: operators, safety teams, reliability engineers
Yes, AI will take many jobs, but in the next few years, these folks will be busy. We still need people who can build, wire, cool, power, and operate the physical reality of data centers.
This is a real opportunity to support SMBs and local specialists: builders, electricians, HVAC firms, and integrators. They will be in sustained demand as AI infrastructure scales over the next few years.
Time to focus on the real economy while
We continue to surf the AI evolution.