A real VPC is a property of the hardware, not a row in someone else's database.

"Virtual private cloud" is one of the better pieces of naming the hyperscalers ever shipped. It sounds like a place. It is actually a database row: a software-defined overlay carved out of infrastructure you share with other tenants, enforced by the same control plane that runs everyone else's. But private is doing a lot of work in that sentence, and it is worth asking what the word means when it is not backed by hardware.

Isolation that's real

On our fabric, tenant separation is not a firewall rule somebody typed. Each tenant gets its own routing separation and its own virtual network identity on our leaf-spine fabric, not a shared VLAN with an access list bolted on. Your network exists as its own thing in the silicon's forwarding tables: your routes are not in a table with anyone else's, and there is no rule ordering to get subtly wrong between you and your neighbor.

This is what blast radius means at the network layer. A misconfiguration, a compromise, or a flood in one tenant's segment has no path into yours, because there is no shared segment for it to cross. For anyone answering to an auditor, single-tenant networking is also the cleaner sentence to defend: the isolation is architectural, which is a posture that maps naturally to ISO 27001 conversations instead of a policy that maps to a promise.

Resilience is the default, not the upsell

At the edge, the network is multi-homed on multiple upstream carriers plus peering at GR-IX and NetIX, so losing a carrier is a routing event for us, not something you notice. Inside, the fabric is leaf-spine. There is no resilience tier to buy. The boring failures are engineered out before you arrive.

The private interconnect

Traffic between our datacenters rides an always-on IPsec-encrypted mesh, datacenter to datacenter, not the open internet. If you replicate storage across sites or run DR to a second location, that traffic is private by default: no VPN for you to build, no public transit for your data to cross, no per-gigabyte meter running on your own replication. The link is engineered for replication, not best-effort internet.

The storage story and the network story meet exactly here: the cross-site replication described in Your DR traffic commutes on the public internet rides this mesh. When we say your recovery point is a design decision and not a hope, the private interconnect is one of the parts that makes the sentence true.

Why this needs owned hardware

None of the above is available to a host that rents its network. You cannot give tenants real silicon-level separation on a rented backbone, and you cannot promise a private interconnect when your inter-site traffic rides the same transit you bought by the gigabyte. This is the networking half of the own-the-hardware thesis, and the reason the first question to ask any host is who owns the network.

Designed to the workload

Network posture is designed per engagement, the same way we size compute and storage: isolation depth, DR paths, and peering needs are questions we ask, not checkboxes you tick. And the honest-sizing rule applies here too. If your workload does not need a second site, we will say so before you pay for one. The wizard asks the disaster recovery question twice for exactly that reason.