Dedicated, single-tenant NVIDIA B300 and RTX PRO 6000 nodes in a sovereign Dutch AI factory. For teams training foundation models, running large-scale inference and fine-tuning on data that must stay in Europe.
The first AI Mills plant Leeghwater at Aalsmeer comes online with 16 racks and 860 kW: HGX B300 nodes for frontier training and RTX PRO 6000 nodes for inference and fine-tuning, with direct-to-chip liquid cooling, an NVIDIA Ethernet front-end and an InfiniBand XDR back-end, aligned to the NVIDIA reference architecture.
Two node types, one sovereign factory. Pick by workload.
| Node | GPUs per node | GPU memory | Interconnect | Best for | At Leeghwater |
|---|---|---|---|---|---|
| NVIDIA HGX B300 | 8 × B300 | 288 GB HBM3e per GPU | NVLink in-node · InfiniBand XDR between nodes | Frontier training, large-scale inference | 256 GPUs (32 nodes) |
| NVIDIA RTX PRO 6000 | 8 × RTX PRO 6000 | 96 GB GDDR7 per GPU | PCIe in-node · NVIDIA Ethernet | Inference, fine-tuning | 256 GPUs (32 nodes) |
Every element designed to maximise GPU performance and minimise your operational overhead.
Direct physical GPU access, no hypervisor layer. 100% of the silicon.
Provisioned to the NVIDIA reference architecture, with drivers, frameworks and networking configured. Deploy in minutes through the portal.
Reserved capacity with contractual SLAs. Your GPUs are physically allocated and always available.
Dedicated servers with hardware-level isolation. Your data never shares memory, storage or network paths with other tenants.
Ownership, operations, stack and people. The GPUs are owned by a Dutch cooperative, operated from the Netherlands, and every component in the stack is checked against European and Dutch principles.
Sovereignty & compliance →One transparent price per GPU-hour in euros, per node type, with the platform and NL support included.
Request capacity →Straight answers. For anything deeper, you get a direct line to our engineers.
Frequently asked questions →