A datacenter is a machine that turns electricity into heat, and the modern AI datacenter turns a great deal of it. The constraint on growth is no longer chips; it is the ability to get firm, carbon-free power to a site on a schedule that matches construction. Two technologies are competing for that role: enhanced geothermal systems and small modular reactors.
Both promise clean baseload. Both are capital-intensive. But they fail in different ways and on different clocks, and the difference decides which one a given project can actually use.
Part One: Two ways to make firm power
Enhanced geothermal drills deep into hot rock, fractures it, and circulates water through the fractures to bring heat to the surface, where it drives a turbine. It is, in essence, a radiator built into the earth. Its fuel is geology, and its output is steady as long as the rock stays hot.
A small modular reactor is a scaled-down nuclear plant built in a factory and assembled on site. Its fuel is uranium, and its output is steady for years between refuelling. It is, in essence, a battery that discharges heat instead of charge.
Both deliver the firm, dispatchable power that a datacenter needs, and both avoid the intermittency that makes solar and wind require storage. The comparison is not about whether they work; it is about when they arrive and what they cost.
Part Two: The clock is the deciding factor
Geothermal's advantage is time. A well can be drilled, fractured, and connected in a few years, and the modular nature of the plant means capacity can be added incrementally. If the geology is right, power can flow while the datacenter is still being fitted out.
An SMR's disadvantage is also time. Design certification, site licensing, and first-of-a-kind construction routinely stretch to a decade or more. The economics improve with repetition — the tenth unit is far cheaper than the first — but the first unit must be built before the tenth can exist, and that first unit is slow and expensive.
A datacenter that needs power in four years can use geothermal; one that can wait ten can consider an SMR, and that ten-year gap is the single largest difference between the two.
Part Three: Why geothermal is site-limited and SMR is not
Geothermal's weakness is geography. It needs hot rock at accessible depth, and that is not everywhere. The best sites are in volcanic regions, and even enhanced systems, which can work in more places, need favourable geology to be economic. A datacenter built on the wrong rock cannot use it.
An SMR's strength is that it can go almost anywhere, because its fuel is transported rather than mined on site. That portability matters for a datacenter sited near users rather than near geology. But portability comes with the licensing and cost problems above.
The result is a sorting: geothermal wins where the geology is good and the timeline is short; SMR wins where the geology is absent and the operator can wait and can absorb first-of-a-kind risk. Most projects fall into one bucket or the other, and few have the luxury of both.
Geothermal can deliver power in about four years where the rock is hot enough, while an SMR needs roughly a decade of licensing and first-of-a-kind construction before its first megawatt flows.
Part Four: The market and the street
A market estimate frames the stakes. If AI datacenter demand keeps growing and grid interconnection queues stay long, the market for on-site firm power could reach the tens of billions of dollars over five to seven years, split between geothermal, SMRs, and gas with capture. Geothermal's share depends on how many good sites exist near load; SMR's depends on how fast licensing can be streamlined.
The social effect is about where the power comes from and who pays for it. A datacenter that builds its own geothermal plant does not compete with households for grid capacity, which can ease the price pressure that new load otherwise creates. An SMR does the same, but later. In both cases, the household sees the benefit as a grid that stays stable rather than as a cheaper bill.
There is a local consequence too. Geothermal creates drilling and operations jobs that are geographically tied to the site; an SMR creates a smaller, highly skilled workforce and a long-term waste obligation. The two technologies leave very different footprints on the communities that host them.
What is actually transferable
The lesson is not "geothermal beats nuclear" or the reverse. It is that firm power is a schedule problem as much as a cost problem, and the right choice depends on how long a project can wait and what geology it sits on. A cheap option that arrives too late is not cheap.
For planners, the practical rule is to match the power source to the timeline first, then optimise cost. If the datacenter must energise in four years, geothermal or gas is the realistic path; if the operator can wait a decade and wants zero carbon, an SMR becomes viable. Choosing the technology before choosing the schedule is how projects stall.
The winner in firm power will not be the cheapest technology on paper but the one that can be permitted, built, and energised before the servers arrive.
Power is the new real estate. The project that secures it early is the one that ships.
SOL: HCqPmPJM3NVtS6YrmVaTZYnuP8XZ2PFgQZKT94VK1LkS
If this saved you time, a tip is appreciated.
Top comments (0)