A global compute layer. Trained where power lives. Served where users live.
Frontier model training compute doubles every six months. Inference workloads scale faster still.
Utility interconnect queues stretch past 2030. New generation takes 7—10 years to permit and build.
The bottleneck has shifted from silicon to electrons. Compute now follows where power lives.
AI factory capacity across both continents. Train where power lives. Fine-tune where data lives. Serve where users live.
The AI lifecycle no longer fits one geography. Training is power-bound — it follows cheap, abundant electrons. Fine-tuning is data-bound — it follows sovereign jurisdictions. Inference is latency-bound — it follows users. A multi-region footprint isn't a hedge. It's how the economics of compute now work.
250 MW energized · First live power site
Training operations live
400 MW · Phase 02 begins
Full lifecycle on one layer
We're releasing investor materials and capacity reservations on a rolling basis.