Most developers building modern products don't own a single server. Their apps run on someone else's hardware in someone else's building — and until recently, that building didn't require much thought. That's starting to change.
- Hosting costs are unlikely to keep falling as electricity becomes harder to secure.
- Region selection and code efficiency now carry real energy implications.
- Energy is becoming an upstream constraint on what gets built, where, and at what cost.
The invisible layer underneath the stack
Infrastructure costs typically show up as line items on a hosting invoice: CPU, bandwidth, storage, CDN. Clean numbers on a clean dashboard.
What those numbers obscure is that they ultimately trace back to a physical box, in a physical building, drawing physical electricity from a physical grid. And right now, that grid is under more pressure than it has been in years — particularly in the UK, where data centre growth is starting to bump into real infrastructure limits.
The picture is straightforward: data centres are among the most energy-intensive facilities running today, they operate around the clock, and cooling alone consumes a significant portion of their electricity. Meanwhile, AI workloads are exploding, grid capacity in some regions is harder to secure, and energy markets remain volatile.
None of that is new exactly. But it changes how the rest of the technology stack should be understood.
What this means for people building on top
Most independent builders won't switch providers based on cooling efficiency — that kind of leverage doesn't really exist at small scale. But the underlying picture still matters in a few practical ways.
Hosting costs are unlikely to keep falling. For most of the past decade, the trend line for cloud compute was "more capability per dollar, basically forever." That trend assumed cheap, abundant electricity and easy grid expansion. Neither assumption is as solid as it used to be. Flat or rising hosting costs are now the base case, not the surprise case.
Region selection matters more than it used to. Regions aren't interchangeable. Some have grid headroom; some don't. Some are racing to add renewable capacity; some are sitting on bottlenecked infrastructure with long connection delays. Latency and price are no longer the only meaningful inputs to that decision.
Code-level efficiency rolls up into electricity. Every wasteful query, every un-cached image, every unnecessarily large bundle eventually becomes electricity somewhere. Performance work used to be a UX concern and a billing concern. It's now also the smallest unit of energy management a developer personally controls.
The cooling problem is bigger than most people realise
One detail worth sitting with: in a typical data centre, cooling can account for a substantial share of total electricity use. Not the servers doing the actual computing — the systems keeping those servers from overheating.
That reframes a lot. An inference call against a hosted LLM isn't just paying for GPU cycles. It's also paying for the cooling required to keep that GPU in a temperature range where it doesn't fail. The bigger and hotter the chip, the more cooling overhead. Modern AI accelerators are very hot.
This is part of why the industry is experimenting with underwater data centres, liquid cooling at the rack level, and deliberate placement in cold climates. It isn't aesthetics. Cooling is one of the few levers operators can pull that actually moves the needle on consumption.
What matters at scale
For anyone running infrastructure at meaningful size, a few things deserve more attention than they used to:
Contract structure. Flexible procurement is increasingly part of how operators talk about energy strategy — the difference between locking in a multi-year price and retaining some ability to ride the market. Customers cannot influence that directly, but they can ask providers what their hedging looks like before committing to long-term deals.
Consumption visibility. Hosting dashboards exist, but most are coarse. If energy keeps being volatile, better signal on which product features actually drive the bill — feature by feature, not just monthly aggregate — becomes a real competitive input.
The sustainability story. Customers, partners and regulators increasingly ask where compute runs and what powers it. "The provider handles it" is a weaker answer every quarter. Knowing the basics is becoming table stakes rather than a differentiator.
The bigger shift
Energy management is no longer just an operational consideration for data centres — it's becoming a core part of long-term business strategy.
That logic extends one rung further. For the people who use data centres — which, in practice, means anyone building anything modern — energy stops being an invisible utility and starts being an upstream constraint on what gets built, where, and at what cost.
There isn't a single clean takeaway. Most builders aren't going to migrate regions or rewrite their code in Rust to save the planet. The shift is more about the mental model: hosting is electricity, electricity is getting harder, and the gap between those two facts will shape a meaningful share of technology decisions over the next few years.
Worth thinking about, even for those who don't own a single server.
"For anyone building anything modern, energy stops being an invisible utility and starts being an upstream constraint on what gets built, where, and at what cost."
Next step
Need visibility on hosting and energy costs?
RAM helps organisations understand where their compute spend meets physical energy constraints — from procurement strategy to consumption visibility.
Talk to us