Disaggregating from HCI: Moving from VxRail to External Storage
If you run VxRail, the Broadcom licensing shake up has reopened an older architectural question that has nothing to do with the hypervisor. Is hyperconverged still the right shape for your estate, or would separating compute from storage serve you better. Here is an honest, vendor neutral view of when disaggregating to an external array is the smarter long term position, and when staying converged still earns its place.
Most of the noise around VMware right now is commercial. The renewal quote, the per core maths, the bundle you are being pushed toward. Underneath that sits a quieter question that will outlast this renewal cycle, and it deserves to be raised on its own terms. The architecture you chose years ago, hyperconverged infrastructure, fused your compute and your storage into a single appliance. That was a deliberate trade. It bought you simplicity and a single point of support. It also welded three separate decisions together, and the current licensing model has made that coupling more expensive than it used to be.
What disaggregation actually means
Hyperconverged infrastructure, HCI, runs storage as software across the same nodes that run your workloads. VxRail is the best known example: Dell hardware, VMware vSphere, and vSAN providing the storage layer, all sold and supported as one engineered system. A disaggregated, three tier architecture pulls those layers back apart. You run your workloads on standard compute hosts, and your storage lives on a dedicated external array reached over a storage network. It is the model most enterprises ran before HCI arrived, and for a large class of estates it is quietly coming back into favour.
The distinction is not academic. In HCI, adding capacity means adding nodes, which means adding compute and storage and software licensing whether or not you needed all three. In a disaggregated design those are independent purchases. That single difference is the root of most of what follows.
Disaggregation is not about chasing a benchmark. It is about decoupling decisions that HCI deliberately joined: when you refresh storage, when you refresh compute, and how much VMware licensing you are obliged to buy. For estates where those three no longer move on the same clock, separating them is a cleaner long term position.
Why the licensing maths made this relevant again
Under the Broadcom model, VMware is licensed per physical core with a minimum of 16 cores for every populated processor, and the storage layer in an HCI design is part of what you license. vSAN capacity is bundled inside VMware vSphere Foundation and VMware Cloud Foundation, so the storage you run on your nodes sits within the same subscription whose price has just jumped. When you move storage onto an external array, that capacity steps outside the vSAN tied portion of the maths. You still license the hosts that run your virtual machines, but you are no longer paying VMware subscription on the storage tier as well, and you can often run those workloads on fewer, denser hosts because the nodes are no longer sized to carry the storage too.
This is not a loophole and it is not free. You are buying an array instead, with its own cost and support. The point is that the array is a known, negotiable, one off purchase plus maintenance, while the vSAN portion of a per core subscription scales with every core in every node and renews on Broadcom's terms, not yours. Before assuming a saving, model both sides. Our VMware renewal calculator is a quick way to see what your licensed core count looks like today and how sizing host count to compute alone might change it.
The three things disaggregation decouples
It helps to be concrete about what separating the tiers actually buys you, because the value is structural rather than a single headline number.
1. Your storage refresh
Best when storage and compute age on different cycles
In HCI, storage and compute refresh together because they live in the same node. Flash that still has years of useful life gets retired early simply because the node around it is due. An external array lets you refresh the storage on its own schedule, usually a longer one, and keep it when you cycle the hosts.
2. Your compute refresh
Best when you want to track CPU generations without touching data
New processor generations arrive faster than you want to migrate petabytes. With the data sitting on an external array, swapping hosts becomes a comparatively light operation. You are moving compute, not storage, so you can adopt denser or more efficient CPUs without a data migration each time.
3. Your licensing position
Best when the vSAN portion of your subscription is doing the damage
Taking storage off the nodes removes it from the vSAN tied portion of the VMware subscription and lets you right size host count to compute alone. For estates where vSAN capacity is a large part of the licensed footprint, this is where the structural saving sits, not in the array itself.
The external array landscape
Naming the destinations matters, because this is where the long tail of the decision lives and because we get asked specifically. The credible external array options for an enterprise leaving vSAN include Pure FlashArray, NetApp, and Dell's own PowerStore and PowerFlex, among others. Each has a different character. Pure tends to lead on operational simplicity and consistent latency. NetApp brings deep data management and strong hybrid cloud integration. Dell PowerStore and PowerFlex keep you inside a single vendor relationship where that has value, with PowerFlex in particular built for a scale out, software defined block model.
We have no preferred array, and that is deliberate. The right choice depends on your workload profile, the data services you need, your existing operational skills and your data, not on the badge. What matters more than the logo is matching the array's strengths, latency, data reduction, replication and snapshot model, to the workloads that actually run on it. The honest answer is that several of these will do the job well, and the differentiation is in the fit, not a spec sheet shoot out.
This is not a fringe view. Dell itself is signalling toward disaggregated infrastructure within its private cloud portfolio, and even VMware's own roadmap has softened the purely converged story. VCF 9.1 now allows mixing VxRail and Dell vSAN Ready Node clusters within a single instance, a step away from the fully fused appliance, though note it is still a vSAN based design rather than external array storage. The direction of travel is toward more architectural choice, not less.
When HCI still earns its place
Disaggregation is not the right answer for everyone, and it would be dishonest to pretend otherwise. Hyperconverged still earns its place when:
- Your estate is small or distributed, where a self contained appliance per site is genuinely simpler than running a storage network and an array at every location.
- You have edge and remote sites, where you want a compact, standardised unit you can ship and support without a storage specialist on the ground.
- Your teams have no dedicated storage skills, and the operational simplicity of one engineered system outweighs the flexibility of separate tiers.
- Your workloads genuinely scale compute and storage in lockstep, so the coupling HCI imposes happens to match how the estate actually grows.
If most of that describes you, the converged model is doing its job and you should be slow to unpick it. The real question is whether it still describes you, or whether it described you five years ago and quietly stopped.
When disaggregation is the better call
Equally, there are clear signals that the fused model has stopped paying its way. Disaggregation tends to be the better call when:
- You run a large estate where vSAN capacity is a meaningful share of your licensed core count.
- Storage and compute are clearly refreshing on different cycles, so binding them together wastes life on one or the other.
- Your workloads need data services an external array does better, such as rich snapshots, array based replication or fine grained quality of service.
- You are facing a sharp renewal uplift where right sizing host count materially changes the number.
- You already run a storage network and have the skills to operate an array, so the operational tax of separate tiers is one you already pay.
What the move actually involves
The architecture is the easy part to describe and the hard part to execute, and we will not pretend otherwise. The effort is driven far more by data gravity, downtime tolerance, replication and disaster recovery interlock, and operational tooling than by the array swap itself. You are introducing or expanding a storage network, rehoming data off vSAN, and rebuilding the runbooks your team relies on. None of that is exotic, but it is real work, and it is the same class of work as any serious platform move. We cover the mechanics in detail in our guide on migrating off VMware, the broader stay versus go question in should you leave VMware, and the array choice itself in enterprise storage migration.
You do not have to decide this at the same moment as your renewal, and you should resist being forced to. The useful first step is to ask whether hyperconverged still fits, model what taking storage off the nodes would do to your licensed core count, and price a representative array against it. That gives you a real number to weigh, and it restores leverage in the renewal conversation either way.
How C4C helps
This is our credibility ground. We spent years on the vendor side across Dell, EMC and the wider storage market, so we know these arrays, the HCI model and the VMware licensing from the inside, and we know how the quotes are built. We will model what disaggregation does to your licensed core count, evaluate the array options against your actual workloads on a consistent basis, and help you decide and execute. Our independence is the point: we have no array to sell and no preferred destination, so the recommendation is the one that genuinely fits your estate and your numbers. We did exactly this kind of modelling and negotiation on a recent VMware renewal, holding the uplift well under 80 percent against a projection of up to 410 percent. You can read that VMware licensing case study for the detail.
Weighing HCI against a disaggregated future?
Send us your estate, rough node count, vSAN footprint and what is prompting the review. We will give you an evidence based view: what taking storage off the nodes would do to your licensing, which class of array fits your workloads, and whether disaggregation is genuinely worth it for you. Independent, with no array to sell. We have sat on the other side of the table.
Prefer email? Reach us directly at hello@c4cgroup.co.uk.
Frequently asked questions
What is the difference between hyperconverged and disaggregated storage?
Hyperconverged infrastructure, such as VxRail, runs storage as software with vSAN across the same nodes that run your workloads, sold as one engineered system. A disaggregated, three tier architecture separates them: standard compute hosts plus a dedicated external storage array reached over a storage network. The practical difference is that HCI scales compute and storage together, while disaggregation lets you scale and refresh each independently.
Does moving off vSAN actually reduce my VMware licensing?
It can, and that is often the point. Under the Broadcom model VMware is licensed per physical core with a 16 core per processor minimum, and vSAN capacity sits inside the VMware vSphere Foundation and VMware Cloud Foundation subscriptions. Moving storage onto an external array takes that capacity outside the vSAN tied portion of the maths and lets you size host count to compute alone, which can lower the licensed core count. You are buying an array instead, so model both sides before assuming a saving.
Which external array should I choose?
There is no single right answer, and we have no preferred array. Credible external options include Pure FlashArray, NetApp, and Dell PowerStore and PowerFlex, among others. The right choice depends on your workload profile, the data services you need, your operational skills and your data, not on the badge. Match the array's strengths in latency, data reduction, replication and snapshots to the workloads that actually run on it.
Is hyperconverged infrastructure obsolete?
No. HCI still earns its place for small or distributed estates, edge and remote sites, teams without dedicated storage skills, and workloads that genuinely scale compute and storage in lockstep. The question is not whether HCI is obsolete, it plainly is not, but whether it still fits your estate today rather than the estate you had when you bought it.
Do I have to disaggregate at my next VMware renewal?
No, and you should resist being rushed. The decision is architectural and outlasts any single renewal. A sensible first step is to model what taking storage off the nodes would do to your licensed core count and price a representative array against it, which gives you a real number and restores leverage in the renewal either way.
Is this the same as a full VMware migration?
Not quite, though they overlap. Disaggregating from HCI is a storage architecture change, and you can do it while staying on VMware. The effort is driven by data gravity, replication and disaster recovery interlock and operational tooling rather than the array swap itself. If you are also weighing leaving VMware entirely, see our guides on migrating off VMware and the stay versus go decision.