GMI Cloud Supports NVIDIA Confidential Computing in NVIDIA NVL72 Racks

GMI Cloud is excited to support the new Confidential Bare Metal deployment option with NVIDIA Confidential Computing announced at NVIDIA GTC Berlin.

The problem: your model, someone else's hardware

Teams training and serving frontier models rarely own all the GPU capacity they need. More and more, they run on infrastructure that a third party owns and operates. That raises a hard question: how do you keep model weights and sensitive data safe when the servers aren't yours?

What this means for GMI Cloud

GMI Cloud is among the providers supporting NVIDIA Confidential Computing deployments. With Confidential Computing now extending to every NVL72 rack-scale system, our NVIDIA GB200 NVL72 and NVIDIA GB300 NVL72 clusters can support confidential workloads. GMI Cloud’s NVIDIA recognition is strong: we are an NVIDIA Exemplar Cloud on NVIDIA GB300 NVL72 systems for training workloads. A new bare-metal option

NVIDIA's newest addition is Confidential Bare Metal. It lets frontier model builders run sensitive jobs on dedicated infrastructure operated by another company, while they stay in control of their own models and data.

One security design, three ways to deploy

NVIDIA Confidential Computing now brings three deployment modes under one unified security design:

●      Confidential Bare Metal — for bare-metal servers

●      Confidential Containers — for Kubernetes clusters

●      Confidential Virtual Machines — for virtualized setups

Teams can pick the mode that fits where they draw their security lines and how they run their operations.

How workloads stay protected

The technology isolates workloads at the hardware level and uses cryptographic attestation to verify the environment. Together, these help keep models and data protected while they are actively running.

Coverage across NVIDIA's newest systems

Support now reaches NVIDIA GB200 NVL72, NVIDIA GB300 NVL72, and NVIDIA Vera Rubin NVL72, which means every NVL72 rack-scale system can run Confidential Computing. It also covers the x86-based NVIDIA HGX Rubin NVL8 and NVIDIA DGX Rubin NVL8 systems.

NVIDIA also plans to build Confidential Computing into NVIDIA DSX OS, its reference architecture for cloud providers building AI factories. DSX OS will support all three confidential modes, giving providers a software base for running AI factories that host confidential workloads. Global systems partners will offer these capabilities through NVIDIA-Certified Systems.