Infrastructure
Where your GPUs run, and how they are kept running.
CoreGPU servers are hosted in professional colocation facilities with redundant power, cooling sized for 1,000 W+ GPUs and staffed security. Here is what every facility we use must provide.
Power
Redundant power feeds and UPS for every rack. Blackwell and Instinct nodes draw up to 1,000–1,400 W per GPU, so racks are provisioned for high-density loads.
Cooling
Direct liquid cooling for the hottest parts (such as MI355X at 1,400 W) and high-capacity air cooling for H100, H200, L40S and RTX PRO 6000 servers.
Network
Redundant upstream connectivity with DDoS mitigation. Inside each 8-GPU server, NVLink or Infinity Fabric connects the GPUs at full bandwidth.
Physical security
Staffed facilities with access control, video surveillance and visitor logging. Only authorised technicians touch hardware.
Hardware
NVIDIA HGX and AMD UBB platforms for SXM and OAM GPUs, enterprise servers for PCIe cards, local NVMe on every node.
Lifecycle
Every allocation is burned in before sale, and disks are securely erased before a server is assigned to the next customer.
Locations and data residency
The location of each allocation is confirmed with your reservation. If you need a specific country, latency target or data-residency guarantee, tell us before you reserve and we will match you with the right allocation.
Our Data Processing Addendum and security practices describe how hardware and data are handled. Enterprise customers can request additional documentation.