Data Centres Are Being Pushed Offshore and Into Orbit Because the Grid Said No

For most of the cloud era, siting a data centre was an optimisation problem with well-understood variables: proximity to users for latency, cheap and reliable electricity, cool climate to reduce cooling costs, and a jurisdiction whose data-protection rules suited the customer base. The constraint that has come to dominate all of those over the past two years is simpler and harder — whether the local grid can actually deliver the electricity a new facility would draw, on a timeline measured in months rather than the decade a major transmission upgrade typically takes. That constraint is what has moved genuinely unconventional siting proposals from novelty to board-level evaluation. Floating and near-shore facilities use seawater for cooling and can be connected to offshore generation directly, bypassing the congested onshore transmission network entirely. Proposals to host compute on satellites take the logic further: continuous solar exposure above the atmosphere, radiative cooling into vacuum, and no local permitting authority to negotiate with at all. Both approaches trade away the things terrestrial siting optimises for — serviceability, latency, straightforward hardware replacement — in exchange for the one thing that has become scarce. The engineering objections to each are substantial and largely unresolved. Marine environments are corrosive and make routine hardware maintenance an expedition rather than a walk down an aisle, which matters enormously in a business where individual accelerators fail on predictable schedules and get swapped without ceremony. Orbital compute faces radiation-induced errors, launch mass economics that punish every kilogram of heat sink, and a repair story that essentially does not exist. Neither has demonstrated the cost-per-unit-of-compute that would justify the move on its own merits. What makes the trend worth tracking is not the likelihood that either becomes mainstream, but what its seriousness reveals about the underlying bottleneck. When operators with straightforward access to conventional real estate begin funding studies of putting servers in the sea and in orbit, the honest reading is that the ordinary path — buy land, sign a power agreement, build — has become slow enough and uncertain enough that exotic alternatives now clear a comparison they would have lost easily three years ago. The siting question has become an energy question wearing a real-estate costume.