For offtakers in the EU and the UK

Europe knows what
it is short of.
Power, and time.

The Cloud and AI Development Act sets out to triple European data centre capacity, and names its own obstacles: permitting, energy, land and financing. That is a five to seven year programme. Our geothermal campus in Kenya commissions in 2027, at a different energy price, and we are taking reservations now.

Talk to the capacity desk → Model your cost →

Read this first

We are not selling you
European sovereignty.

Plenty of providers will. We would rather be useful than flattering, so here is the boundary before the pitch.

A facility in Kenya is not an EU-sovereign option. Under the sovereignty framework in the Cloud and AI Development Act, it does not qualify for sensitive public-sector or regulated workloads, and we are not going to suggest otherwise to win a procurement round you would later have to unwind.

If your requirement is EU data residency for regulated data, buy that inside the EU. If your requirement is sustained compute at a workable energy price, on a timeline that exists, keep reading.

The case

Three things that are true
whatever the regulation says.

Energy cost is the workload cost

At scale, a GPU hour is mostly an electricity bill. European industrial power prices are among the highest in the world, and geothermal baseload from the Olkaria corridor is not exposed to the same market. That gap is the whole argument, and it does not depend on policy.

A date you can plan against

Tripling European capacity is a five to seven year programme against a permitting and grid backlog. Our modules are factory-built and ship on a rail corridor to the campus, with the first 400 kW Tier III module commissioning in 2027. That is a date, not a queue position.

Outside US jurisdiction

For organisations reducing exposure to the US CLOUD Act, we are not a US-headquartered provider and not a regional zone of one. That is a statement about jurisdiction, and we are careful not to let it be read as a claim about EU sovereignty.

Renewable at source

A 95% renewable mix generated where the compute runs. For teams whose own emissions reporting has to withstand a real audit, generated-clean is a materially different claim to offset-clean.

Honest fit

The workloads this is for.

Training runs. Long, scheduled, latency-insensitive, and dominated by the energy bill. The clearest fit we have.
Batch and offline inference. Work measured in throughput per pound rather than milliseconds to a user.
Research compute. University and lab workloads that need sustained hours at a predictable price, and can plan a year ahead.

Not for anything you need running before 2027, not for regulated personal data under an EU residency obligation, not for latency-critical services to European users, and not for anything your compliance function has to certify as sovereign. Those belong in the EU, and we will say so on the call.

Bring us the energy line
from your current bill.

The comparison that decides this is arithmetic, not positioning. Send the workload and what it costs you today, and we will show our working.

Talk to the capacity desk → Model it yourself →

Europe knows whatit is short of.Power, and time.

We are not selling youEuropean sovereignty.

Three things that are truewhatever the regulation says.