In our recent post, Alex Marshall, Group Business Development and Marketing Director, argued that AI has entered its reality phase, the point where power, not compute, decides what actually gets built. If that is the constraint, then the next question is the one almost every operator he talks to is now asking: how fast can the power arrive, and what are we prepared to accept to get it sooner?
That question is quietly rewriting how energy infrastructure gets designed. “Speed to power” started life as a delivery problem, a line on a project schedule. It is turning into something closer to a strategy in its own right. Read on as Alex explores these ideas in more depth…
The lead times stopped matching the demand
Energy infrastructure was built around long horizons. Multi-year grid connections, extended permitting, assets designed to run for decades. All of that assumed a fairly predictable demand picture, and for most of the last few decades that assumption held.
AI has broken it. A data center chasing several hundred megawatts is not working to a five-year plan. It often needs that capacity inside a year, sometimes inside months, because the workload it is built for will not wait on a substation. The scale of the shift still surprises people who have been in this industry a while. Five years ago a twenty megawatt distributed generation project was considered large. Today we are working on data center installations several times that size. So the practical question has changed. It is no longer only what we should build, it is what we can get running before the grid is ready, and how that fits what comes after.
Temporary power stopped being temporary
The first place this shows up is in the language. What we used to call temporary power, the interim kit you bring in to bridge a gap, is not really temporary anymore.
On live projects I see interim systems being folded into multi-year operating plans. Not because anyone set out to run them for years, but because the gap they were meant to bridge keeps widening. Grid delays are not always short. Demand is not pausing while the queue clears. So a solution specified to cover eighteen months gets asked to hold a site for four or five years, deliver reliable baseload, and integrate cleanly with whatever gets built around it. That is a different job, and it needs to be specified as a different job from day one, not discovered halfway through.
Fast power and strategic power are not the same thing
The most visible response to all this has been the growth of rental and leased generation, and it is easy to see why. It deploys quickly, it keeps upfront capital down, and it gives operators room to move in an uncertain environment.
But rental solves for speed, and speed is only half the problem. On its own it does not solve for long-term reliability, operating efficiency, or how well the system sits alongside everything else on the site. That is the tension worth naming plainly: fast power and strategic power are not automatically the same thing. The projects that get this right use fast-to-deploy generation to meet the demand in front of them, while designing for the hybrid, longer-lived system the site will actually need. The move I keep pointing to is from rental thinking to lifecycle thinking.
Modularity is doing the reconciling
The thing that lets those two ideas live together is modularity. Rather than waiting on one large centralized build, operators are deploying scalable engine-based generation in containerized, packaged blocks that can be added to in phases.
That buys three things at once. Speed, because units commission in months rather than years, and modern gas engines can be brought to full output in minutes. Flexibility, because capacity gets added as demand actually appears instead of being guessed at up front. And resilience, because a distributed set of units carries fewer single points of failure than one large asset. For high-density, fast-scaling, mission-critical AI load, modularity is less a design preference than a direct answer to time pressure.
Building backwards
There is a quieter consequence in all this that I find the most interesting. Projects are increasingly being planned in reverse. Instead of starting from the finished, ideal end-state and working toward it, teams start from the power they need now, deploy fast and modular, then layer the permanent systems in over time.
That is closer to how projects actually unfold once real constraints hit, and it is the logic underneath the Structured Transition Model, the framework I use to sequence these decisions so that today’s fast deployment does not become tomorrow’s stranded asset. I will give the model its proper treatment in a later edition. For now the principle is enough: the strongest infrastructure strategy is rarely the one that is most complete on day one. It is the one that can evolve fastest without having to be torn up.
“For years, temporary power was where you compromised. In an AI-paced market, it is becoming where you compete. The operators who treat fast, modular generation as the first phase of a designed system, rather than a stopgap they will rip out later, are the ones who reach the market first and keep the option to adapt.”
– Adam Wray-Summerson, Technical Sales Director, Clarke Energy
Where Rehlko’s Clarke Energy business fits
In this environment an energy partner is measured less by the equipment on the delivery note and more by whether the whole thing was designed to move fast, run reliably, and still change shape later. That is where our work sits. Engine-based solutions for rapid, scalable capacity. Hybrid configurations that balance the fast and the strategic. And lifecycle support that keeps a site running well past the initial deployment, in markets where grid access is uncertain and the requirements will not sit still.
Looking ahead
Speed to power is not really a delivery metric anymore. It is becoming a defining feature of how modern energy infrastructure gets planned. The question has moved on from whether fast power is needed to how quickly it can be deployed and how intelligently it can be integrated into what follows.
There is a harder version of this question waiting, and it is the one I want to take on next month: what happens when speed to power and decarbonization pull in different directions. Getting power to a site fast, and getting it there clean, do not always point the same way, and the operators who plan for both from the start will be in a very different position from those who treat carbon as a problem for later. Because in an AI-paced market, the fastest route to power and the most strategic one are increasingly the same route.
Want to learn more?
To learn more about onsite power solutions for data centers, contact Clarke Energy.
Questions and Answers
Speed to power is how quickly usable electrical capacity can be delivered to a site, as distinct from how much capacity is eventually available. For AI data centers it has become as decisive as cost, efficiency, or emissions, because a facility that cannot be energized on time cannot capture the demand it was built for.
AI workloads create step-changes in demand that arrive far faster than the grid can respond. A single facility may need several hundred megawatts within a year, while grid connection, permitting, and transmission upgrades still run on multi-year timelines. When the workload will not wait, how fast power can be delivered often determines whether a project is built on schedule, or at all.
Grid connection delays increasingly outlast the interim solutions meant to bridge them. Systems once brought in for a matter of months are now specified to run for years, deliver reliable baseload, and integrate with permanent assets. In practice they are no longer temporary, so they need to be designed as part of the long-term plan from the start.
Rental generation is attractive because it deploys quickly and keeps upfront capital low, but it solves only for speed. On its own it does not guarantee long-term reliability, operating efficiency, or clean integration with the rest of the site. The strongest strategies pair fast-to-deploy generation with a designed, hybrid system built for the long term, shifting from rental thinking to lifecycle thinking.
Modular generation uses scalable, containerized engine-based units deployed in phases rather than as one large centralized build. It delivers speed, because units commission in months; flexibility, because capacity is added as demand appears; and resilience, because a distributed fleet carries fewer single points of failure. That combination fits the high-density, fast-scaling, mission-critical nature of AI load.
What is modular power generation, and why does it suit data centers?
Pre-engineered, containerized modular systems can be deployed and commissioned in months rather than the years a large grid-connected build can take, and modern gas engines can reach full output in minutes once running. Actual timelines depend on site preparation, fuel supply, and logistics, but the gap versus waiting for the grid is usually measured in years.
Instead of starting from an ideal end-state and working toward it, projects increasingly start from the power needed now, deploy fast and modular, then layer permanent systems in over time. Sequencing those decisions so that early fast deployment does not become a stranded asset is the logic behind the Structured Transition Model. The aim is a strategy that can evolve quickly, rather than one that is complete on day one.
