Tenancy Design

Tenancy is the one decision that gets more expensive every month you leave it. Which kind of organisation, where quota actually lives, and the constraints that arrived with 9.1.1.

Tenancy Design
Updated 5 November 2026
The first version of this post treated tenancy as a general design problem and skipped the decision that comes before all the others, which is which kind of organisation you are creating. That section is new, and so are the two diagrams, the namespace ceiling that arrives with 9.1.1, the VPC and public cloud changes, and Broadcom's own six consumption patterns. Information captured on 5 November 2026 against VCF 9.1.1. Sources are at the foot.

Of all the decisions in a VCF deployment, the tenancy model is the one I see made worst. Not because it is hard, but because of when it gets made.

The pattern is familiar. Infrastructure goes in, the platform team is busy, and somebody needs organisations and projects created so the pilot can start. So they get created, sensibly enough, based on whatever the org chart looked like that week. Eighteen months later there are forty projects, a quota model nobody can explain, and a request to reorganise it all that quietly never gets scheduled because too much now depends on it.

Tenancy is the one decision in the whole build that gets more expensive every month you leave it. Worth spending a fortnight on properly.

Decision zero, which kind of organisation

I used to open this conversation with hierarchy. That was wrong, and it took watching a customer get a long way down the wrong road to see why. VCF Automation 9.1 is bimodal. There are two kinds of organisation, they are not two flavours of the same thing, and almost everything below depends on which one you are in.

Two organisation types in VCF Automation 9.1, VM Apps and All Apps, sharing one vCenter and NSX instance under Mixed Tenancy Mode
The bimodal consumption model in VCF Automation 9.1. Captured 5 November 2026 from the bimodal consumption design documentation.

An Organisation for VM Apps carries traditional VM-based workloads on VLAN-backed or NSX-backed networking. It is the road out of Aria Automation, and if you are upgrading a legacy Aria estate, the converted environment becomes the first organisation in the new instance whether you planned it that way or not.

An Organisation for All Apps carries the modern platform services: VM Service, vSphere Kubernetes Service, Database Services Manager, VPC-based networking. It needs a vSphere Supervisor cluster, either enabled on a cluster with spare capacity or on a new one. Broadcom's own guidance for greenfield with no Aria dependency is to start here.

The useful part of 9.1 is that both can share one vCenter Server and one NSX instance. That is Mixed Tenancy Mode, and it is a feature flag rather than an architecture. Before this, each organisation generally wanted its own workload domain with its own vCenter and NSX, which is why so many tenancy designs from the 9.0 era look more expensive than the requirement justified.

The ordering constraints are the part that catches people
The VM Apps organisation has to be created first, and it has to be configured, with its vCenter and NSX cloud accounts mapped, before a region is configured against that same vCenter and NSX Manager. The multi-fleet design, where VM App workloads span multiple fleets or workload domains, does not work with the All Apps model. So if you stand up All Apps first and then decide you want VM Apps alongside it on the same vCenter, you are not making a small change. Decide in week one.

On deployment shape, there are three. Both organisations on a shared cluster, which is cheapest and carries real contention risk, and which Broadcom positions for small or pilot environments. Separate clusters under the same vCenter, better performance isolation, more hardware, and the documentation calls this mandatory for large deployments with diverse mission-critical workloads. And multi-fleet, which as above is VM Apps only.

What a tenant actually is

With the organisation type settled, you can ask the question that usually gets asked first. In most enterprises a tenant is one of three things, and plenty of places have all three.

A department or business unit
The most common. Finance, retail banking, claims. Boundaries follow budget, which makes chargeback straightforward and makes the isolation requirement moderate. Usually these people trust each other, they just need their spend kept separate and their capacity protected.
A project or application team
Finer grained, and the one that multiplies. Easy to create, hard to retire, and the reason estates end up with forty projects and no owner for half of them. If this is your unit of tenancy, decide the retirement rule at the same time as the creation rule, because you will not get a second chance to make it routine.
An external customer
Different requirement entirely. Isolation has to be a security boundary rather than an expectation, the service description becomes a contract, and the organisation boundary is doing real work. This is where the dedicated models in the deployment model documentation earn their cost.

Write down which one you mean before you create anything. If the answer is more than one, you need a hierarchy that accommodates both, and that is much easier to design up front than to retrofit.

Where quota actually lives

Quota is not a property of a tenant, which is the part that catches people out. It sits at two specific places in a four-level containment, and if your design does not say which, it is not finished.

Containment from organisation through region quota, project and vSphere namespace to a VKS cluster, showing which levels carry a number
Organisation, Region Quota, Project, vSphere Namespace, VKS cluster. Captured 5 November 2026 from the All Apps organisation multi-tenancy model.

A Region Quota is the allocation an organisation receives from a region. It is defined per vSphere Zone and it covers CPU and memory, as limits or as reservations, plus the VM classes and the storage classes that organisation can see. The sentence worth underlining is that a single Supervisor cluster can be mapped to multiple organisations through a Region Quota. That is the mechanism that makes shared infrastructure multi-tenant rather than merely crowded.

Below that, projects hold members, shared catalogue items, approval and reclamation policies. vSphere Namespaces sit inside projects, bound to the scope of the project they belong to, and they carry the CPU, memory and storage limits along with the network the workloads land on. Isolation between All Apps organisations is described as total, and the thing doing the isolating is that namespace boundary. Tenants do not need their own service accounts into the infrastructure, which is worth saying out loud to a security reviewer who is expecting to see them.

The five decisions

1. Hierarchy depth
Three levels, not two, and the third is the one people forget. Organisation, project, vSphere namespace. Do you use organisation as a business unit and project as a team, or organisation as a legal entity and project as a business unit. Both work. Picking one and being consistent is what matters, because inconsistency here is what makes reporting impossible later. The namespace is the level your developers actually feel, since it is where their limits and their network come from, so it is worth deciding whether a project gets one namespace or many before the first one is created. The tenancy deployment models documentation sets out five supported shapes, from consolidated management and workloads in one instance through to a dedicated fleet with its own VCF Automation per tenant.
2. Quota strategy
Three broad approaches. Hard quotas per tenant, which is predictable and wastes capacity. Soft quotas with overcommit, which uses capacity well and produces arguments. Or tiered quotas where a small guaranteed allocation sits under a larger burst pool. I favour the third for most enterprises, because it gives finance a number and gives the platform team room to breathe. Whichever you pick, anchor it to the Region Quota, decide who can raise one and how fast, and know that a self-service platform where quota changes take three weeks is not self-service. One operational note that belongs in the design rather than the runbook: deleting a region quota created before an upgrade from 9.0.x currently fails, with a KB workaround, so a tenant created in the 9.0 era does not decommission as cleanly as one created since.
3. Isolation boundaries
Network isolation, storage policy separation, identity scope. The question to answer honestly is whether tenants need to be unable to reach each other or merely expected not to. Those are very different designs and very different costs. For business units inside one company, expected-not-to is often adequate. For external customers it never is. In All Apps the network side is VPC-based, and 9.1.1 adds VLAN-backed VPCs consumable through VCF Automation, which run directly on physical VLANs with no overlay encapsulation. Simpler to reason about and easier to sell to a network team that distrusts overlays, but private networks are not supported in that model, because addressing is limited to the CIDRs behind the physical VLANs. Know which one you are picking before the first VPC is cut.
4. Who is the provider
Someone curates the catalogue, owns quotas, and answers when a self-service workload misbehaves. If that role is not named, it defaults to whoever is most responsive, which is not a sustainable operating model. This is the organisational decision inside the technical one, and it is the reason a lot of platforms stall at pilot.
5. What happens at the end
How a tenant gets decommissioned. Data, quota return, identity cleanup, and for anything GPU-backed, what happens in accelerator memory between occupants. Nobody designs this at the start and everybody needs it eventually. Worth ten minutes now, and worth testing once on something disposable, because the region quota deletion issue above is exactly the kind of thing you want to find on a lab tenant.

Where the newer constraints bite

There is a ceiling on namespaces, and it is lower than most people assume
In 9.1.1 a single deployment setup is limited to 30 vSphere Namespaces across all Argo CD instances. If your tenancy model gives every project its own namespace and you expect to run GitOps across the estate, that is the number that decides how far the model scales, not the organisation count. Worth putting on the diagram rather than discovering at tenant fifteen.
Region is a hard edge for GitOps
The Argo CD based GitOps service in 9.1.1 runs one instance per hosting namespace, and it can sync across multiple namespaces and VKS clusters only within the same region. A multi-region estate needs multiple instances by construction. There are two forms of it: a Supervisor Service installed from vCenter and instantiated per namespace with the CLI, and a VCF service that provider admins install from Service Management and organisation admins then deploy into their own namespaces. The second is Tech Preview in 9.1.1, so treat it as a design input rather than a build item. One more thing to put in the service description: Argo CD support for new VKS Kubernetes versions lags by two to four months, so your tenants cannot have the newest Kubernetes and GitOps at the same time. Detail here.
The public cloud door has closed in VM Apps organisations
Native management of AWS, Azure and GCP resources is deprecated and disabled by default in 9.1.1. Cloud accounts, cloud zones, projects, profiles, networks and security groups are all affected. Existing deployments with public cloud resources are marked Stale, create, update and delete are blocked, and Day 2 actions are off apart from lease management. New blueprints containing public cloud resources are flagged Invalid and cannot be released to the catalogue. It can be re-enabled by a Service Broker or Assembler administrator through a KB, but the direction is clear. If anyone has sold a tenancy model on the strength of one catalogue spanning private and public cloud, that promise needs revisiting now rather than at renewal.
GPU tenancy is its own conversation
Broadcom's September performance testing found a single tenant using 36 to 38% of GPU streaming multiprocessors on a light inference workload, rising to 71% with four tenants sharing. So consolidation works. The constraint worth knowing at design time is that co-locating multiple tenants inside one VM is supported only with manual tenancy management, because VCF Automation requires each tenant in its own VM. If self-service is the goal, that decides your isolation model for you. Source.
Edition boundaries shape what tenants can be offered
Application and database management packs are VCF or VCF Edge only. If your catalogue promises database provisioning with monitoring included, check the entitlement covers it before it appears in a service description. Feature comparison.
Two smaller things that will bite an integration team
Requests to the cloudapi endpoints now require a token carrying the new vcfa_api scope, and VCF Automation and the VCF Operations orchestrator have to match through the subminor version. Neither changes a tenancy model, but both will change somebody's automation on upgrade day, and the person who wrote that automation is usually the provider team.

Broadcom publishes six patterns worth reading first

There is a named set of multi-tenancy consumption patterns in the design library, and it sits a couple of clicks away from the deployment models page that most people land on, so it is easy to miss. Read it before you draw anything. The shape I describe below sits alongside these rather than replacing any of them, and four of the six cover ground that otherwise gets improvised in a workshop.

  1. Resource isolation with a single VPC for multiple businesses in one Supervisor zone.
  2. Resource isolation with multiple VPCs for multiple lines of business in one Supervisor zone.
  3. Resource isolation with a single VPC for multiple lines of business across three Supervisor zones.
  4. Resource isolation for multiple organisations with dedicated VRFs or transit gateways.
  5. Multi-region, multi-tenant resource isolation with dedicated and shared gateways.
  6. Multi-organisation resource isolation with shared NSX across workload domains.

If you are a business-unit tenancy with everything in one region, you are pattern one or two and the choice is whether each line of business gets its own VPC. If isolation has to be a security boundary, pattern four is where that argument gets settled. If you are multi-region, pattern five is the starting point and the GitOps constraint above is already waiting for you there.

A shape that usually works

Not a recommendation for your estate, just the pattern I find myself drawing most often for a mid-sized enterprise. In the vocabulary above it is an All Apps organisation per business unit, sitting closest to patterns one and two.

Organisation per business unit, with the business unit owning its budget. Project per application team or product, created from a template so they are consistent. Tiered quota expressed as a Region Quota with a guaranteed floor per project and a shared burst pool at organisation level. Network isolation at the organisation boundary, not the project, because project-level isolation multiplies quickly and rarely earns its complexity. And a named provider team with a published service description, so people know what they are getting and what they are not.

The bit people skip is the service description. If a tenant cannot find out what the platform promises without asking someone, they will keep raising tickets, and you will have built self-service that nobody self-serves.

Before you create anything

  • Decide VM Apps or All Apps, and if both, create the VM Apps organisation first.
  • Write down what a tenant means in your organisation, in one sentence.
  • Decide hierarchy depth across all three levels, and what sits at each.
  • Pick a quota model, anchor it to the Region Quota, and name who can change one and how quickly.
  • Decide whether isolation is a security boundary or an expectation.
  • Decide whether your VPCs are overlay or VLAN-backed, knowing the private network trade-off.
  • Name the provider team and write a one-page service description.
  • Design decommissioning, including accelerator memory if GPUs are in scope, and test it once.
  • Check the region constraint and the 30-namespace ceiling against your GitOps plans.
  • Check nothing in the catalogue depends on native public cloud management.
  • Read the six consumption patterns and say which one you are closest to.
  • Model what this looks like at three times the current tenant count.

That last one catches most design problems. A model that works for six tenants and falls apart at twenty is common, and you only find out when you are already at fifteen. The namespace ceiling is the first wall you will hit, and it arrives sooner than the organisation count suggests.

Want a score rather than a reading list?
I built a short self-assessment covering the same ground. Fifteen questions, about four minutes, and it scores your estate across five areas with what I would do next in each. Nothing is sent anywhere, it all runs in your browser.

Tools that go with this

Tenancy is a design problem, but it sits on top of two counting problems. Estate Readiness Triage tells you which clusters can carry any of this, and Upgrade Runway tells you whether there is time to get them there before support runs out.

Free to use, including with customers. Nothing you enter leaves your browser. All tools.

Sources

All of the below checked on 5 November 2026 against VCF 9.1.1. These pages move, and at least one of them has moved since this post first went up: the tenancy deployment models page now carries a last-updated date of 26 October 2026.

Read next

Write down what you could not find

A design document asserts things its format never earned. The fix is structural: a numbered section for what you could not establish, and attribution by source type in the body rather than only in the references.