Geothermal and the data-centre build-out.

The demand story is by now well told — we set out the numbers, and the integration argument behind them, in one resource, many energy forms. This article is about the part of the story that decides where and when projects actually happen. The IEA is careful to list the build-out's bottlenecks in the plural — grids, electrical equipment, advanced chips, planning systems, social acceptance, capital — and no single one of them binds everywhere. But the one measured in years in every major market is the time it takes to connect anything to the grid. That queue reshapes what a megawatt of generation is worth, and it is where the case for geothermal is actually decided.

The queue is the constraint

One distinction first, because sloppy versions of this argument blur it: new generation queues for grid interconnection, data centres themselves queue for load connections, and both wait on transmission that must be built. They are related, separately measured processes — and both sides of the meter are slow.

The generator side has the best-measured data. Lawrence Berkeley National Laboratory's annual study of US interconnection queues found that the median time from connection request to commercial operation has doubled: under two years for generation projects built in 2000–2007, over four years for those completed in 2018–2024. The typical project completed in 2024 had spent 55 months in the queue — up from 36 months in 2015 and 22 in 2008 — and the 2026 edition, with data through 2025, puts the median for projects built in 2025 above five years. And a place in the queue is not a power plant: of all the capacity that requested connection between 2000 and 2019, only 13% had reached commercial operation by the end of 2024. Most of what enters a queue never leaves it as electricity.

On the load side, Europe's waits are longer where the data centres actually want to be. In the FLAP-D hubs — Frankfurt, London, Amsterdam, Paris, Dublin — a new data centre waits seven to ten years on average for a grid connection, and some wait as long as thirteen; in Italy, by contrast, data centres can wait as little as three years, on Ember's June 2025 analysis. The concentration explains why: by 2023 data centres accounted for up to 42% of metropolitan electricity consumption in Amsterdam, London and Frankfurt, and nearly 80% in Dublin. The European Commission itself states the problem plainly: data-centre demand growth "is usually met with a lack of available capacity to connect to the grid". The IEA estimates around 20% of planned data-centre projects could be at risk of delay unless connection constraints are addressed.

The policy response has begun, and it changes who gets connected rather than how fast. Great Britain's queue had swollen past 700 GW — about four times what the country requires — before the system operator replaced first-come-first-served with readiness-based prioritisation in December 2025, reordering a 283 GW pipeline and reserving 99 GW for transmission-connected demand, data centres explicitly among it. The system operator's own estimate is that the reform could unlock up to £40bn of investment a year. Queues, in short, are becoming allocation mechanisms: what you bring to the connection now determines whether you get one.

The queues also contain headroom held for later: on IEA and McKinsey figures reported by Ember, Europe's data centres drew on average 44% of their nominal power rating in 2024, and the Irish system operator puts the average at about 34% of contracted capacity. Some of that gap is redundancy, commissioning stages and peak provisioning rather than waste — which is precisely why phased and flexible connection agreements, which contract the ramp instead of the ceiling, are among the reforms on the table.

Keeping the scale honest

It is worth saying clearly what this argument is not. On the IEA's Base Case, data-centre electricity demand roughly doubles over this decade's second half — the agency's April 2026 update puts 2025 consumption at 485 TWh, doubling toward roughly 950 TWh in 2030 — and still accounts for less than 10% of global electricity demand growth between 2024 and 2030. In the EU, the IEA's estimate is 70 TWh in 2024 rising towards 115 TWh by 2030; Europe's growth to 2030 (over 45 TWh, up 70%) is a fraction of the US's (around 240 TWh, up 130%). Anyone presenting data centres as the dominant driver of global electricity growth is overstating the case, and we decline to.

The argument for geothermal here is not volume. It is that this particular load is extreme in three ways at once: it runs continuously, it concentrates in specific places, and — in Europe — it increasingly arrives with a legal obligation attached to its waste heat. Those three properties, not the terawatt-hours, define what the supply side has to be.

What firm power is worth behind a scarce connection

When connections are rationed, the question stops being "what does a megawatt-hour cost?" and becomes "how much energy does each megawatt of scarce grid access deliver?" That is a capacity-factor question. On the EIA's final 2024 figures, US geothermal plants ran at a 64.6% capacity factor — roughly 1.9 times wind's 34.3% and 2.8 times utility-scale solar's 23.2%. (Fleet capacity factors move year to year, which is why we label the vintage; the 2024 figures are final, from the Electric Power Annual.) For a data centre whose load runs day and night, and for a grid operator deciding what deserves a constrained connection, that difference is not decoration: per megawatt of connection capacity, it is roughly two to three times the annual energy, delivered in a shape that matches the load — before a wind or solar portfolio has added the storage and overbuild that hourly matching would require.

Honesty about scale applies on the supply side too: geothermal is today a small industry, a point we have made elsewhere and stand by. The claim is not that geothermal will power the build-out. It is that where the resource, the load and the queue intersect, it is one of very few options that can hold up its end of a 24/7 contract from a compact site.

The deals, read precisely

The hyperscalers have noticed, and the contracts are instructive — provided they are read as what they are.

The most developed structure is Fervo Energy's Corsac Station arrangement with Google in Nevada: a development-stage project contracted to supply 115 MW of enhanced-geothermal power for Google's data centres, to be delivered through the utility under NV Energy's Clean Transition Tariff, approved by the state regulator in May 2025. It is explicitly not a behind-the-meter arrangement — the utility acts as intermediary, and the tariff is structured so that the premium Google pays does not shift costs onto other ratepayers. Note what the structure does and does not do: it settles who buys and who delivers, and it works inside the connection process rather than around it. The larger framework agreement that followed, contemplating up to 3 GW, is covered in our earlier article.

The instructive counter-example is nuclear, not geothermal: the flagship co-located data centre at the Susquehanna plant. Federal regulators rejected an expansion of that behind-the-meter arrangement in November 2024, and by mid-2025 the operator was converting the whole structure to a front-of-meter, grid-connected power purchase agreement — 1,920 MW through 2042 — while the regulatory framework for co-located load remained unsettled. The lesson for anyone planning energy for data centres: the durable structures work with the grid and its regulators, not around them.

And some announcements are exactly that. Meta's 2024 agreement with Sage Geosystems — up to 150 MW — was announced with no site selected, a first phase that "will aim to be online and operating in 2027" in the release's own words, and a developer that had no operating commercial plant at the time. That is momentum worth noting, not capacity delivered; we read our own industry's press releases with the same discipline we apply to everyone else's.

Europe: the heat is part of the deal

Europe adds a constraint the US market mostly lacks — and it is the one that plays to integrated geothermal's strengths. The recast Energy Efficiency Directive (2023/1791) brought data centres above size thresholds under mandatory energy-performance reporting, feeding a European database, and pushes operators to utilise their waste heat where technically and economically feasible — a conditional obligation, not a blanket mandate, but one that changes what a well-prepared project proposal looks like. A data centre in Europe increasingly needs an answer to the question: where does your heat go?

That answer already exists at city scale. Stockholm's Open District Heating scheme, run by Stockholm Exergi, had by 2022 contracted twenty heat suppliers — data centres among them — recovering enough heat to warm around 30,000 apartments a year. Data-centre heat is low-grade, which is precisely why it needs a network designed to absorb it rather than a one-off engineering miracle.

This is where the design brief becomes an integrated one. A geothermal district-energy node can deliver firm power and cooling to a site and give that site's waste heat a network to flow into — one project serving the load's three defining properties at once. Designed carelessly, the same two sources can instead compete for the same network heat load, which is exactly why the system has to be designed as one. On cooling we keep the claim scoped, as we did when we wrote about absorption chillers: heat-driven cooling is established, peer-reviewed technology at building scale, and we do not claim it is proven at data-centre scale. And we do not claim the package is proven because its pieces are — integrating them is real engineering with its own risks, and owning that integration is precisely the work. What European law has changed is the incentive to attempt it as one design.

What we take from this

A queue that runs seven to ten years is not a delay; it is a design constraint. It rewards the developer who arrives with generation, cooling and a heat plan as one proposal — because that proposal asks the grid for less, gives the city something back, and puts more energy behind whatever connection it is granted.

Catoxy Energy is being built to arrive at exactly this intersection: firm power, cooling and a home for waste heat, designed as one project rather than three negotiations. That is a statement of design intent, not of delivered projects — and the numbers above are the reason we think the intent is pointed at the right decade.

---

Sources: Lawrence Berkeley National Laboratory, Queued Up: 2025 Edition (December 2025) and the 2026 edition (July 2026), for US generator-interconnection durations and completion rates; Ember, Grids for data centres in Europe (June 2025, underlying data ICIS), for European connection waits and metro consumption shares; European Commission, "In focus: Data centres — an energy-hungry challenge" (November 2025); IEA, Energy and AI — Energy demand from AI (2025) and Executive summary, for demand projections, the under-10%-of-growth figure and the 20%-at-risk estimate; NESO, grid connection reform implementation (December 2025); US EIA, Electric Power Annual, Table 4.8.B (final 2024 capacity factors); Latitude Media, on NV Energy's Clean Transition Tariff approval (May 2025); Utility Dive, on the Susquehanna front-of-meter conversion (June 2025); Sage Geosystems and Meta, announcement of the geothermal power agreement (Business Wire, 26 August 2024), with development-status context from ThinkGeoEnergy's coverage; EU Energy Efficiency Directive 2023/1791; EU Covenant of Mayors, Stockholm heat recovery from data centres (2023).