Bare Metal

If the only thing that matters is the lowest price per server, a commodity host will win and we will tell you so. What you get here instead is an engineer who tells you which tier fits before you order, hardware replacement on every component with no parts billing, the same Private Link and direct peering that front our cloud, and one European jurisdiction over all of it with no US parent above us. The minimum term is six months, which is short on purpose: the argument only works if leaving is easy.

We publish the ceilings, not the per-tier configuration. Intel tops out at 4 TB of RAM and takes one or two sockets; AMD tops out at 3 TB with EPYC 9000 and is always single-socket. Storage reaches about 369 TB on AMD, either SSD or NVMe, and about 77 TB SSD or 307 TB NVMe on Intel. What we do not publish is what a given tier ships with today, because the fleet rotates and a parts list is what a configurator is for. Tell an engineer what the workload does and you get a tier, a configuration and a price back, in writing, before you order.

Software RAID by default. We do not fit RAID cards on customer bare metal, so if you need a hardware controller it is a custom build and we source it with the rest of the machine.

Every server is dual stack from the start, IPv4 and IPv6. IPv6 ships as a single /128, and a /56 is rentable if you need to subnet. Additional IPv4 is rentable the same way, in blocks from a /29 up to a /24.

Yes, and sometimes in the same racks. A bare metal server, a cloud instance and your own colocated hardware sit behind the same edge in Madrid, Amsterdam, Munich or Helsinki, and reach each other without going out to the internet.

Over Private Link, in whichever of its three delivery modes suits what you already have. It is the same product that fronts our cloud, our telecom services and colocation, so one private path covers the whole estate rather than one per product.

Yes, managed or unmanaged. We source the parts and build it. Chassis are Dell or Supermicro. Two GPUs per box is the usual density, and a larger chassis can be ordered if you need more cards.

We rack the server, power it on and hand over access, including remote IPMI and KVM. The OS, the RAID and everything above them are yours, which is the same unmanaged model you will know from other operators. If you would rather not run that layer, any tier can be ordered managed, and then OS patching, monitoring and backup are ours.

You open a ticket and we replace it, free, whatever the component is. In business hours we aim to have the server back in service within 4 working hours. Weekends and out of hours are covered by the advanced support plan, which is paid. There is no parts billing at any point, on any plan.

Not on an unmanaged server. Bare metal has no backup product of its own: backup exists inside the managed option, and its scope is agreed with you at purchase. Running unmanaged, the backup strategy is yours, the same as the OS.

Support is free on every product we sell, with a 4-hour target response time on support requests. The advanced plan is paid and buys shorter targets plus weekend and out-of-hours cover. Hardware replacement is separate from both: a defective component is the product failing to work as sold rather than a support entitlement, so it is free on every plan, and the only thing the advanced plan changes is that the repair window keeps running outside business hours.

99.9% product availability, by contract, with automatic credits if we miss it, and the same figure covers every product we tie to an SLA. What it measures is the product being available, not the hardware. There is no failover underneath a single-tenant machine, so a failed component is not an SLA event and we will not pretend to measure it: what we commit to instead is a repair target: back in service within 4 working hours, free. If a workload genuinely needs cover around the clock, that is the advanced support plan rather than the SLA.

In stock, 4 hours. Built but not yet racked, or waiting on a part, 48 hours. Sourced and assembled to order, up to 14 days. Those are the three cases, and we tell you which one you are in before you order rather than after.

Entry and Standard include 10 TB a month, Performance+ 20 TB, and High end and GPU are unlimited. On the three metered tiers you can buy unlimited traffic as an extra, and that upgrade comes on a 10G port. If you stay on an allowance, traffic above it is billed per additional block, or blocked if you ask for that when you order. Either way it is in your contract, not a surprise line on an invoice.

Six months. At the end we delete your data and securely erase the disks to NIST SP 800-88 before the machine goes back into the fleet.

The US CLOUD Act lets American authorities compel any provider under US jurisdiction to hand over data in its possession, custody or control, wherever it is stored, without notifying you. Dedicated hardware does not change that: if the company holding the rack answers to a US parent, the order reaches the rack through the parent. DIVERSITY has no US parent: no US company owns or controls us. There is no entity above us to serve.

Encryption protects data in transit and at rest from third parties. It does not protect against an order served on the company that holds the hardware, under its own jurisdiction. If that company holds the keys, or can be compelled to help reach the disks, encryption is not a legal shield. Jurisdiction is, and on an unmanaged server the disks are the only copy we hold.

That the company operating the machine is incorporated, operated and legally subject to European law. Not a subsidiary answering to a parent somewhere else. When a regulator or a government asks for your data, what decides the answer is EU law rather than a foreign one. That is not a feature we added to the product. It is the company the product runs inside.