Root and privileged hardware access across the full rack, your own operating system, serial console, power operations, and direct hardware visibility on a dedicated NVL72 rack. Without owning the hardware, liquid cooling, or fabric.
Root is yours. The rack is ours.
Bare metal should mean more than SSH access to a pre-provisioned OS. Mantle gives infrastructure teams an owner-operator model for GB200 NVL72: bring your own OS, boot it your way, reach the serial console when networking is down, and control power and hardware visibility through an authenticated portal/API path.
You get a dedicated GB200 NVL72 rack with the operator surface exposed: a custom iPXE boot flow, root and privileged hardware access, browser serial console, node power operations, scoped Redfish visibility, private access, public reachability, and a network delivered ready for production use. Mantle keeps the parts that should stay platform-managed: firmware, fabric, BMC credentials, hardware management, liquid cooling, power, and break/fix. Available now, in the Pacific Northwest.
You control everything from the operating system up, with the operator surface beneath it.
Chainload a custom iPXE menu you control for a custom boot flow, with control over kernel, init, and OS services, then netboot your own provisioning stack and base image.
Control disk layout, kernel, drivers, agents, packages, updates, and rebuild cadence.
Open a browser serial console even when the OS is unreachable or not yet configured.
Power on, power off, restart, and reset nodes through portal controls and policy-governed Redfish paths.
Inspect hardware telemetry, inventory, platform health, and Redfish resources through the authenticated, policy-governed portal/API path.
SSH-key-only access through a jump host, with customer-managed keys.
Public IPs, ports, and internet egress are configured around the services you choose to expose.
Private subnets, DHCP/DNS options, and NIC guidance are delivered up front with configuration tuned for speed and redundancy, so your stack can come online quickly.
Add-ons — High-performance VAST storage and non-GPU compute are available on request.
It's the shared-responsibility model every serious infrastructure provider runs on, drawn where it belongs for bare metal. You run everything on top of the hardware: OS, drivers, workloads, orchestration, data, agents, runtime, and rebuild cadence. Mantle runs the infrastructure that keeps the rack online: firmware, NVLink and InfiniBand fabric, BMC credentials, hardware management plane, control plane, power, cooling, monitoring, and break/fix.
That clear split is what makes this much control safe to hand over. You can rebuild the system as often as you need without taking responsibility for the parts that can brick the platform or destabilize the rack.
Root is yours. The rack is ours.
This is not a fractional GPU slice and not a node carved out of someone else's platform. The model is built for dedicated GB200 NVL72 capacity: a full rack, 72 Blackwell GPUs, one NVLink domain, and the rack-scale fabric delivered as part of the environment. When multiple racks are allocated, the InfiniBand fabric between them is delivered preconfigured.
// The platform we run is the platform we ship into every site we build.
You already have an OS image, provisioning pipeline, Kubernetes distribution, scheduler, or cluster lifecycle tooling. Bring it and boot it.
Build your own product on operator-grade rack capacity without buying, racking, cooling, and maintaining NVL72 hardware yourself.
Use custom kernels, drivers, agents, low-level debugging, and repeatable rebuilds on the real machine.
Run a full stack end to end on dedicated hardware, with ownership of the OS and everything above it.