Why start with a map
When someone asks you to "run this outside the region," the first problem is the names, not the architecture. Google has renamed its hybrid products several times. Blog posts, Terraform modules and job adverts still use names like Anthos, GKE on-prem and GKE on Bare Metal, while the console now uses others. Before you can compare anything, you need one page that answers three questions about every service:
- Which group is it in? That tells you who owns the hardware and where it sits.
- What was it called before? You'll need this when you search old runbooks.
- Is it still alive? That keeps you from building on something that is being shut down.
Think of the six groups as the shelves in a hardware shop. On one shelf you buy only the remote control. On another the shop installs a machine in your building. On another you rent a rugged box to carry into the field. Every service sits on exactly one shelf.
The six groups
# hybrid-edge-map.yaml: one page, six shelves (status as of Oct 2026)
management_planes: # Google runs the control; you run the hardware
- name: Fleets # logical groups of clusters for policy, config, identity
- name: GKE attached clusters # register EKS, AKS or any CNCF-conformant cluster
- name: Google Distributed Cloud (software only) for VMware
formerly: [Anthos clusters on VMware, GKE on-prem, GKE on VMware]
- name: Google Distributed Cloud (software only) for bare metal
formerly: [Anthos clusters on bare metal, GKE on Bare Metal]
- name: GKE on AWS / GKE on Azure
status: maintenance mode since 2025-03-17; unsupported, APIs disabled 2027-03-17
provider_hardware_on_site: # Google supplies and manages the racks in YOUR building
- name: Google Distributed Cloud connected
formerly: [Google Distributed Cloud Edge]
note: a rack is 6-24 machines plus switches; needs a link to Google Cloud
- name: Google Distributed Cloud air-gapped (racks)
formerly: [Google Distributed Cloud Hosted]
note: no connection to Google Cloud needed; base footprint is 3 racks, 22 servers
metro_and_telecom_edge: # provider-run hardware close to a region or a network
- name: Bare Metal Solution # dedicated servers in regional extensions near Google regions
- name: Distributed Cloud connected at operator sites # same product, run at telco edge
rugged_devices:
- name: Google Distributed Cloud air-gapped appliance # portable ruggedized hardware
iot:
- name: Cloud IoT Core
status: RETIRED 2023-08-16 # move to partners, e.g. ClearBlade (drop-in MQTT/JWT)
data_transfer_and_gateways:
- name: Transfer Appliance # ship-a-box: TA40 (40 TB) or TA300 (300 TB) usable
- name: Storage Transfer Service # online copies; agents run on your own servers
- name: Cloud Interconnect # private link from your site or colo into Google
Gotcha: "software only" isn't "on site hardware". Google Distributed Cloud (software only) uses the GDC brand, but you buy, rack and repair the servers yourself. The brand tells you the family, and the group tells you who holds the screwdriver.
The rename trail
The single most useful thing you can learn here is this timeline:
2019 Anthos launches (GKE on-prem becomes "Anthos clusters on VMware")
...plus "Anthos clusters on bare metal", "Anthos clusters on AWS/Azure"
2023 (Aug+) Anthos brand folded into GKE Enterprise (GA 15 Nov 2023)
Anthos clusters on VMware -> GKE on VMware
Anthos clusters on bare metal -> GKE on Bare Metal
2024 Both re-badged as Google Distributed Cloud (briefly "GDC Virtual")
-> today: Google Distributed Cloud (software only) for VMware / bare metal
2 Sep 2025 GKE Enterprise stops being a separate edition: its features join GKE,
and Config Sync, Policy Controller, Cloud Service Mesh etc. are add-ons
So "GKE Enterprise" is now a former name as well. GKE is a single offering with no tiers, and fleets are simply part of it. If an old design document says "requires Anthos licensing," read that as "check current GKE and add-on pricing."
Walkthrough: inventory what you already have
Before you choose anything new, find out which shelves you're already using. These read-only commands are safe to run in any project:
# 1. Every cluster registered to the project's fleet: GKE, attached, GDC software only
gcloud container fleet memberships list --project=my-platform-prj
# NAME UNIQUE_ID LOCATION
# gke-eu-prod 0f3c... europe-west2
# eks-payments 7a91... us-east1 <- attached EKS
# vmw-plant-01 c22d... europe-west2 <- GDC software only
# 2. Which Kubernetes versions attached clusters accept in a region
gcloud container attached get-server-config --location=us-east1
# 3. Any legacy GKE on AWS clusters? These need a plan before 17 March 2027
gcloud container aws clusters list --location=us-east4
If step 3 returns anything, put it at the top of your backlog. Google's suggested route is to run the workload on GKE or another distribution and attach it. You keep the fleet view and lose the retiring product.
Best practice: tag each membership with its group, for example
--update-labels=group=management-plane. Later scorecards and cost models can then filter by shelf.
Spotting the dead and dying
Entries change status over time, so date every entry on your map:
| Service | Status (Oct 2026) | What replaces it |
|---|---|---|
| Cloud IoT Core | Retired 16 Aug 2023; device APIs gone | Partners such as ClearBlade, Losant, Kaa |
| GKE on AWS | Maintenance mode; unsupported 17 Mar 2027 | EKS + GKE attached clusters |
| GKE on Azure | Maintenance mode; unsupported 17 Mar 2027 | AKS + GKE attached clusters |
| GKE Enterprise edition | Merged into GKE 2 Sep 2025 | GKE plus separate add-ons |
The IoT shelf on Google's own catalogue is now empty. Any IoT answer from here on is a partner platform feeding Pub/Sub or a data service, not a first-party product.
Trade-off to remember: a migration partner can offer a near drop-in replacement (ClearBlade copied IoT Core's MQTT topics and JWT auth). The price is that a third party is now in your support chain and your contract.
Key takeaways
- Six shelves: management planes, provider hardware on site, metro and telecom edge, rugged devices, IoT, data transfer and gateways. A service's shelf tells you who owns the hardware.
- Fleets and GKE attached clusters are management planes. GDC software only runs on your hardware. GDC connected and air-gapped racks are Google hardware in your building, and the air-gapped appliance is the rugged one.
- Bare Metal Solution lives in provider-run regional extensions near Google regions. Transfer Appliance, Storage Transfer Service and Cloud Interconnect move data and traffic in.
- Rename trail: Anthos → GKE Enterprise (2023) → GKE (Sep 2025). GKE on VMware and GKE on Bare Metal → Google Distributed Cloud (software only).
- Retired or retiring: IoT Core (16 Aug 2023), GKE on AWS and GKE on Azure (unsupported 17 Mar 2027).