CoreWeave rents GPU capacity to companies training and serving AI models. It is a specialist rather than a general cloud: the hardware, networking and scheduling are built for large distributed training and high-volume inference rather than for running databases and web servers.
Scale, from the company's own disclosures: 51 data centres worldwide and more than 1.5 gigawatts of active power as of 30 June 2026, against approximately 4.2 gigawatts of contracted power reported on the 11 August 2026 earnings call. The gap between those two figures is the build-out pipeline.
In Europe it runs nine data centres across five geographies including the United Kingdom, Spain, Sweden, Norway and, from October 2026, Denmark, where most of a planned 22.5 MW of IT capacity went live at atNorth's DEN01 facility in Copenhagen on NVIDIA Blackwell systems connected with NVIDIA Quantum-X800 InfiniBand. That site supplies surplus heat to the local district heating network.
CoreWeave also announced what it describes as the first global backbone network built for AI workloads, at petabit scale, connecting North America, Europe, the Middle East and Asia Pacific. Each point of presence carries multiple 400 Gbps connections with hardware-based MACsec encryption and redundant paths. New points of presence are planned for the United Arab Emirates, India, Indonesia, Singapore, Australia, Korea and Japan, beginning late in Q4 2026 and completing a full loop in mid-2027.
Pricing is contracted rather than published and was not verified against a vendor page for this listing.