CoreWeave brings Denmark AI capacity online as wider network rollout heads toward 2027

CoreWeave says it has activated most of the planned 22.5 megawatts of IT capacity at atNorth’s DEN01 data center in Copenhagen, its first deployment in Denmark. The October 27 announcement pairs that computing expansion with a separate project: additional network points of presence across the Middle East and Asia Pacific, beginning later in the fourth quarter of 2026 and continuing toward a full loop in mid‑2027.

For organizations planning AI infrastructure, those are two different availability signals. Copenhagen already has operating capacity under the announcement. The new network locations have a staged future timetable. A network connection in another country does not, by itself, establish a new location where a customer can rent GPUs.

The Copenhagen deployment uses NVIDIA Blackwell systems connected with NVIDIA Quantum‑X800 InfiniBand, according to CoreWeave’s release. The company says the addition takes its European infrastructure footprint to nine data centers across five geographies, including the United Kingdom, Spain, Sweden and Norway alongside Denmark.

The 22.5 MW figure describes the planned IT power capacity, not a count of available accelerators. CoreWeave says a majority is online but does not state the precise operational amount in this announcement. Reporting the whole 22.5 MW as activated would overstate the disclosed deployment.

For a team comparing places to run a workload, the practical next step is to establish the available system configuration and capacity for its own deployment. The release does not provide a DEN01 GPU inventory, a customer booking schedule or prices. The Blackwell and networking description identifies the announced infrastructure; it does not settle those procurement questions.

CoreWeave plans additional points of presence in the United Arab Emirates, India, Indonesia, Singapore, Australia, Korea and Japan. Its stated schedule begins later in the fourth quarter of 2026 and extends to a full loop in mid‑2027; it does not give a separate activation date for every location.

The company describes the expansion as connecting its existing North American and European network with the Middle East and Asia Pacific. A point of presence is a network access location. It can form part of the path between a customer and compute elsewhere, without being the place that hosts that customer’s GPUs. That distinction matters when deciding whether the announcement meets a requirement for local compute or simply changes connectivity to the platform.

The proposed backbone uses multiple 400‑gigabit‑per‑second connections at each point of presence, CoreWeave says in the announcement. MW measures power, while gigabits per second measures a connection’s data rate. The Copenhagen power figure and the network connection figure cannot be combined into a measure of AI workload speed.

Nor does the rated speed of those links establish the transfer rate a particular application will achieve. A useful infrastructure discussion separates the compute available for the job from the network path used to reach it. For Copenhagen, buyers can ask about the specific systems and capacity they can obtain. For the future points of presence, they can ask which route is active, when the relevant location will be available and how it connects to the compute they plan to use.

The release also describes MACsec encryption and redundant network paths. Those are CoreWeave’s infrastructure statements, rather than an independent assessment of a customer’s security, latency or reliability. The concrete change is a first Danish deployment with most planned power capacity online, plus a dated network expansion that is still being rolled out.

AI‑generated illustration separating compute infrastructure from the network connecting it. It does not depict DEN01 or an actual CoreWeave network topology. Original conceptual artwork by OpenTools Team, based on CoreWeave’s October 27 announcement.