Two clocks, badly aligned
A reactor is a sixty-year asset, and increasingly a longer one. A data centre hall is refreshed on a cycle measured in years: the servers turn over roughly every four to six, the cooling and electrical topology is reworked as rack densities climb, and the workload the building was sized for may not exist in a recognisable form a decade out. The building shell lasts; the demand profile inside it does not.
Now put the build schedules next to each other. A hyperscaler can move from land acquisition to live capacity in two to three years, and has every commercial incentive to compress that further. An SMR on a licensed site, with a design through regulatory assessment and a supply chain already producing modules, is still a five to eight year proposition on an optimistic view, and longer for a first-of-a-kind on a new site. The compute demand arrives first, by a wide margin. Whatever powers the early years of that facility is not going to be the reactor.
This is the first thing to get straight in any offtake conversation. The reactor is not the solution to the near-term connection problem. It is the solution to the long-term cost, carbon and firmness problem, and the gap between the two has to be bridged with something else.
Availability is not the same as firmness
Nuclear is often described as always on. The right description is that it has a high capacity factor with planned and unplanned outages. A well-run modern plant might achieve ninety per cent or better over a year, which means a refuelling outage on a predictable cycle and the statistical certainty of unplanned trips. For a grid, that is manageable. For a data centre operator contracting to a tier-level availability commitment, a reactor on its own does not deliver the guarantee, because the failure mode is a loss of the entire supply at once rather than a gradual degradation.
The practical consequence is that behind-the-meter nuclear does not remove the need for grid connection, standby generation or storage. It changes what those assets are for. They move from being the primary supply to being the cover for a unit trip and the planned outage. That is a different sizing calculation and a different commercial structure, and it needs to be worked through at the feasibility stage rather than discovered during connection negotiations.
The grid does not want a sudden gigawatt either
There is a system-operator view of this that rarely features in the press releases. A large reactor dedicated to a single private consumer, with a grid connection held as backup, looks to the network like a demand of several hundred megawatts that can appear instantaneously when the unit trips. Network operators do not welcome that. Connection terms will reflect it, through capacity charges, constraint arrangements, or firm requirements for the consumer to shed load on a defined timescale.
The symmetrical problem is export. When the data centre’s demand falls away, whether because of a workload shift, a refresh outage or a change of owner, the reactor is still producing. Unless there is a grid route for that output and a commercial arrangement behind it, the asset is stranded. Designing for bidirectional flexibility from the start is considerably cheaper than retrofitting it.
What happens when the data centre is obsolete?
This is the question that ought to shape the contract, and it is the one asked least often. The reactor will outlive the computing load it was built for, probably by forty years. Several things follow.
Offtake needs a term structure that does not assume a single counterparty for the asset’s life. A twenty-year power purchase agreement against a sixty-year asset leaves forty years of merchant exposure, and the party carrying that exposure needs to be identified at financial close rather than later. Site planning should assume successor loads: grid export capability, industrial heat offtake, hydrogen production or a different compute tenant. Siting decisions that make sense only for one customer are a concentration risk dressed up as efficiency.
Decommissioning and waste obligations also sit with the nuclear licensee and do not transfer with a data centre sale. Where a technology company is a shareholder in the generating entity rather than simply a customer, the corporate structure needs to be clear about where those liabilities land through ownership changes that are far more likely over sixty years than over six.
How to structure it properly
None of this makes the pairing a bad idea. Firm clean power at predictable long-run cost is genuinely valuable to a compute business, and a creditworthy anchor offtaker is genuinely valuable to a nuclear project. The point is that the value is only realised if the structure respects the asymmetry.
In practice that means a bridging strategy for the years before the reactor connects, a grid connection and backup arrangement sized for a unit trip rather than for normal operation, an offtake term that recognises the asset will outlive the counterparty’s current business model, and a site configuration that keeps successor-load options open. It also means being honest in the investment case about which risks sit with the generator, which with the consumer and which with the network.
NCL advises on exactly these questions, from early feasibility and siting through to connection strategy, offtake structuring and the regulatory pathway. If you are assessing an SMR-backed compute proposition, we can help you test it.