Data centers are seen as a problem for the grid. Today, that is true.
They are large, continuous, and mostly inflexible loads. They increase peak demand, lock in capacity needs, and do not meaningfully respond to price signals. In a system already dealing with variability, they behave like an added layer of rigidity.
But that is not inherent. It is a design choice.
The power system is shifting from one constrained by production to one constrained by absorption. With growing wind and solar, there are increasing periods where generation exceeds demand. Prices collapse to zero or below, energy is curtailed, and investment signals weaken.
Data centers should play a different role.
Instead of acting as fixed demand, they should become controllable consumption:
- increase load when there is excess generation
- defer non-critical workloads when supply tightens
- align operation with price rather than ignoring it
The technical basis already exists. Workloads are virtualized, geographically movable, and often not time-critical. What is missing is the operating model. Today, uptime is treated as absolute, and load is assumed constant. Flexibility at the workload level is simply not exposed to the power system.
The result is baseload-like consumption in a system that increasingly cannot tolerate it. The craziest voices are even suggesting nuclear capacity just to support their inflexibility.
This is backward. They should work for us, not the other way around.
If data centers were built and contracted as adaptive loads:
- AI training and batch jobs would follow surplus generation
- non-urgent compute would shift in time or locatio
- partial load reduction would be a controlled mode, not a failure
- electricity price would become an input to scheduling
It would also improve nuclear economics.
Nuclear plants are designed for steady, high-output operation. Their cost structure depends on high utilization and stable pricing. Today, they are penalized during periods of high renewable output, when prices fall even though their production remains valuable. Yet, giving up fossil leads to increasing challenges during windless winter days. We urgently need 1-2 new GW-range NPPs but they can't be built for a couple of winter months alone. Adaptive data centers would help here.
We can force generation to follow variability, or we can make part of the demand do it.
Data centers are one of the few loads capable of the latter at scale.
If they remain fixed, they deepen the problem.
If they adapt, they become part of the solution.
Either we make them work for us, or make ourselves work for them.
***
Systems designed for one regime begin to misbehave when the regime changes.
We often treat electrification as incremental. A bit more load here, a bit more capacity there.
It isn’t.
It’s a change in scale.
The amount of power we are trying to move is no longer in the same regime as before. At some point, the question stops being how to extend what we already have and becomes whether the underlying approach still makes sense.
At lower voltages, you can keep adding lines. Spread the load. Build around constraints. It works—up to a point.
But each new line comes with its own footprint. More land taken. More corridors cut. More people are affected, each one just enough to remain acceptable.
Individually, the decisions look reasonable.
Together, they don’t.
What changes at higher voltage is not just capacity. It’s geometry.
A single line begins to replace several. Losses over distance stop dominating. The system becomes something you can deliberately shape, rather than patching together.
At some scale, lower voltages stop being a conservative choice and become a structural inefficiency. At that point, the question is no longer whether 800 kV is excessive.
It’s whether anything below it is still aligned with what we are trying to build.
***
Electricity has no memory.
Once it enters the grid, it loses its origin. A kilowatt-hour from solar, a battery, or a turbine is indistinguishable from any other. The system sees flow, not history.
But taxation still assumes history.
As the system becomes more dynamic, assumptions built for one-way flow begin to fail. You pay when energy enters your storage system. Someone else pays again when it leaves. Not because it is a different energy, but because the system cannot recognize that it isn’t.
Flat taxes make it worse. Every pass through storage looks like new consumption. The system penalizes the very flexibility it increasingly depends on.
There are ways to soften it.
If taxation follows value instead of volume, the distortion shrinks. Charging happens when prices are low. Discharging offsets high-value consumption. The double layer remains, but its effect fades.
Removing it entirely would require something cleaner: taxing only final use, or explicitly netting storage cycles.
Both are straightforward in principle.
Neither is simple in practice.
So the real question is not how to eliminate the imperfection.
It is whether we are willing to keep a system that discourages the behavior we are trying to encourage.
***
Small modular reactors are often presented as a natural fit for combined heat and power.
The argument sounds straightforward: place the reactor close to demand, take heat directly, avoid conversion losses, keep everything local.
But the thermodynamics are less decisive than they first appear.
If you take heat from a turbine cycle in a conventional pressurized water reactor, you are not creating heat from nothing. You are redirecting it. For every unit of heat extracted, you give up a fraction of electrical output—on the order of a few tenths. Not negligible, but not dominant either.
Viewed another way, that same electricity could have been used to drive a heat pump.
And large heat pumps, when designed properly, operate in that same range of effectiveness. One unit of electricity becomes several units of heat. Different path, similar outcome.
So the advantage is not where it first seems to be.
It is not in efficiency.
Which leaves a more uncomfortable question: if the thermodynamics are roughly comparable, why introduce a more expensive way of generating heat?
Because the system is not defined by thermodynamics alone.
There are places where moving electricity is harder than moving heat. Where grid capacity is already constrained, but district heating networks are close and inexpensive to extend. Where the demand is steady, local, and not easily shifted.
There are also cases where the temperature itself matters—where the required heat is above what large heat pumps can easily provide, or where integrating them would place new burdens on the grid at exactly the wrong time.
And then there is siting.
A large reactor assumes space, infrastructure, and acceptance that are not always available. A smaller unit changes that geometry. Not by being better in principle, but by fitting where something larger cannot.
Seen from that angle, the role of SMRs becomes clearer.
They are not a more efficient way to produce heat.
They are a way to place heat where the rest of the system makes other options difficult.
And if electricity is not the constraint—if the need is simply for large amounts of steady, low-cost heat—the logic simplifies further. The turbine ceases to be a necessity and becomes a choice.
Remove the turbine.
The reactor gets smaller.
The system gets quieter.
Nothing is optimized for electricity anymore. Only for heat.
A machine designed to do one thing, in the place where that thing is needed.
In that sense, heat-only reactors are not a compromise.
They are what remains when you stop designing around electricity and start designing around demand.