November 10, 2026
Last week, I found myself thinking about server rooms the way most people think about plumbing. You only really notice them when something starts leaking, overheating, or costing you a small fortune.
That is basically where we are with AI infrastructure.
For years, plenty of firms could get away with treating power as background noise. Important, yes, but hardly boardroom material. Then AI turned up like a house guest who plugs in a dozen heaters, three tumble dryers and a crypto mining rig, then asks why the lights keep flickering.
The old assumptions are breaking. Compute density is moving from kilowatts to megawatts. Traditional AC infrastructure is being pushed to its limits. And if you are building, scaling or modernising AI capability, this is no longer just an engineering footnote. It is a business risk, a hiring challenge and, frankly, a bit of a reckoning.
Computer Weekly recently made the case that AI's power demands are accelerating a shift towards high-density DC architectures, largely because they can cut conversion losses and better support the loads modern AI environments demand (Source: Computer Weekly, 'AC/DC: The AI power reckoning driving a shift to high-density DC'). They are right to sound the alarm.
My take is simpler. Companies still hiring and planning as if AI is just another application stack are already behind.
Everyone loves talking about models, use cases and productivity gains. Fewer people want to talk about what sits underneath all of that: power, heat, cooling, resilience, physical capacity.
Not sexy. Hugely important.
AI workloads are not politely asking for a bit more headroom. They are bulldozing through the neat assumptions that underpinned many legacy estates. Infrastructure designed for conventional enterprise applications was not built for dense GPU clusters chewing through serious power and generating serious heat.
This is the bit some leadership teams miss. They think they are making a software investment. In reality, they are making an infrastructure bet.
If that infrastructure still relies on inefficient power conversion chains, fragmented estates and ageing assumptions around load profiles, costs will climb, resilience will wobble and timelines will slip.
That is before we even get to sustainability targets, which suddenly look rather awkward when your AI ambitions start guzzling energy like a 1980s rock band on tour.
Let me be clear. This is not an anti-AC rant. Alternating current has done the heavy lifting for decades and remains fundamental to how facilities operate. But AI is exposing where AC-centric environments become inefficient, especially when every conversion step introduces losses and complexity.
That is why direct current is getting fresh attention.
At high density, those efficiency gains stop being marginal and start becoming strategic. Fewer conversion losses. Cleaner delivery to equipment designed for modern compute loads. Potentially simpler architecture in the right settings. Better alignment with battery storage and renewable integration too.
Again, this is not sci-fi. It is practical engineering catching up with a new reality.
The big idea here is not that every business suddenly needs to rip out its estate and go fully DC tomorrow. That would be a wonderfully expensive way to panic. The point is that leaders need to start asking sharper questions about whether their current infrastructure model is fit for the compute future they keep talking about.
If your AI roadmap is ambitious but your power architecture is an afterthought, you do not have a roadmap. You have a liability wearing a strategy hat.
This is where I put my recruiter hat on, which, conveniently, is never very far away.
Power architecture decisions are not just technical decisions. They change who you need in the building.
The market is already short on people who can connect infrastructure, energy efficiency, resilience, cloud, facilities and AI compute strategy in one coherent conversation. That intersection is where a lot of firms are now heading, often while still hiring in neat little silos.
That is a mistake.
I see businesses trying to solve next-generation infrastructure problems with yesterday's org chart. One team owns cloud. Another owns data centres. Another owns facilities. Another owns sustainability. Another owns procurement. Everyone is busy. No one owns the full picture. Then people wonder why delivery feels like a relay race where half the team forgot the baton.
The firms that move fastest will hire and structure differently.
They will look for people who can:
These people are not always easy to find. Which is exactly why waiting until your AI programme starts straining your estate is a bad plan.
Here is my slightly cheeky prediction.
A chunk of the market is about to over-index on AI talent while under-investing in the infrastructure leadership needed to support it.
In plain English, they will hire shiny model engineers into environments that cannot support the workloads efficiently, securely or sustainably.
That is like recruiting a Formula 1 driver, then handing them a hatchback with one dodgy tyre and asking for podium finishes.
Good luck with that.
If you are serious about AI, your hiring plan should not begin and end with data scientists, MLOps engineers and product leads. You also need people who understand:
In some firms, that means rethinking existing leadership roles. In others, it means creating new ones. Either way, the market is shifting from 'Can we build AI?' to 'Can we support AI at scale without setting fire to the budget or the energy strategy?'
That second question is where the grown-ups enter the room.
You do not need to become a power systems expert overnight. You do need to stop treating this as someone else's problem.
Here are a few questions worth asking internally:
If those questions produce long silences, nervous laughing or a sudden urge to change the subject, that is useful information.
If I were advising a leadership team from scratch, I would keep it brutally simple. Use this lens: Capacity, Capability, Coordination.
Can your estate physically support the power, cooling and density demands AI will create?
Do you have the people, skills and technical leadership to design and run the right environment?
Are infrastructure, cloud, facilities, sustainability and commercial teams aligned, or are they all optimising in different directions?
If any one of those is weak, your AI strategy is wobblier than it looks on the slide deck.
The best operators will not wait for a crisis. They will start treating energy architecture as part of competitive advantage.
That means:
And yes, it also means admitting that old assumptions may no longer hold.
There is no shame in that. Technology changes. Demand changes. What matters is whether leadership changes fast enough.
AI's next bottleneck may not be model performance. It may be whether your infrastructure can actually feed the machine.
That is the real shift here. Power is moving from back-office concern to front-line strategic issue. And as that happens, the businesses that win will not just have smarter tools. They will have sharper operators, better infrastructure judgement and the courage to hire for the future rather than the org chart they inherited.
Because when AI pushes your environment from kilowatts to megawatts, this stops being an IT problem.
It becomes a leadership one.