1. Why the choice is about total cost, not hardware price
Storage is often compared by the price of an array or of a set of servers. Over several years, the costs that matter most tend to appear later: software licenses, support contracts, power and cooling, staff time, the cost of expansion steps, and the cost of an outage or a migration. A total cost of ownership (TCO) view, built over a fixed planning horizon, makes these costs visible and allows options to be compared on the same basis.
The planning horizon is the number of years you model, and it should follow your hardware refresh cycle, your contract lengths and your expected growth. A model that ignores the refresh will favor whichever option is cheaper to buy today. Include a scenario in which the refresh is delayed, because delays are common and their cost, in support risk and in performance, is rarely written into the original proposal.
2. How HCI and SAN architectures differ
A storage area network (SAN) separates compute from storage. Servers connect over a dedicated network to an external storage array, which has its own controllers, disks and management software. Hyperconverged infrastructure (HCI) combines compute and storage in the same nodes, with software that pools the local disks across a cluster. Both can run virtual machines. They differ in how capacity and performance grow, and in how they fail.
In HCI, adding capacity usually means adding nodes, which brings compute and storage together. In a SAN, you can often add disks or shelves without adding servers, but the array becomes a shared dependency with its own lifecycle and support contract. The better fit depends on whether your workloads grow in compute and storage at a similar pace. If storage grows much faster than compute, a SAN may be more economical, and the opposite is often true for HCI.
3. Comparing the main cost areas
The table compares the main cost areas in qualitative terms. These are tendencies, not verdicts. The actual result depends on the quotes you receive, the licensing model of each product and the configuration you need to meet your performance and resilience targets. Use the last column to make sure each figure in your model comes from a written source. Quote the source next to each entry.
The areas with the largest uncertainty are usually licensing and operations. Licenses may be charged per node, per core or per capacity, and each model grows differently. Operations include routine tasks, upgrades and incident response, and their hours are often underestimated. Put these items into the model explicitly rather than leaving them to a general overhead percentage. Give each one a named owner and a source.
| Cost area | HCI tendency | SAN tendency | What to verify |
|---|---|---|---|
| Initial hardware | Often bundled per node, simpler purchase | Separate array, switches and servers | Full bill of materials from each supplier |
| Software and licensing | Per node or per core, may recur | Array features may be licensed separately | Recurring fees, renewal terms and price protection |
| Power, space and cooling | Scales with the number of nodes | Array and fabric add fixed overhead | Measured or rated consumption for the real configuration |
| Staff and operations | Fewer silos, but cluster skills are required | Specialized storage skills often needed | Estimated hours for routine tasks and upgrades |
4. Scaling and failure economics
Scaling matters because the cost curve is rarely linear. HCI grows in steps of nodes, so you may pay for compute you do not need in order to gain storage, or the reverse. A SAN can grow storage independently, but it requires capacity planning for the fabric and the controllers, which can become a step cost of its own. Model the next expansion steps explicitly, not only the average cost per unit of capacity.
Failure domains also carry costs. In HCI, a node failure affects the storage pooled on that node, and the rebuild traffic competes with production workloads while the data is restored. In a SAN, a fault in the array can affect many hosts at once. Include the cost of downtime, the performance impact during rebuilds and the skills needed to recover, because these costs often decide the real comparison.
5. Checklist for a TCO model
A TCO model is only as good as its inputs. The checklist below lists the inputs that most often go missing, from licenses and power to staff time and migration. Each input should have a source, such as a quote, a measurement or a documented assumption, and the source should be visible to the people who will approve the decision.
Run the model at least twice: once with the base assumptions and once with a stress scenario in which licenses rise, growth is slower or faster, or the refresh slips by a year. If the ranking of the options changes under those scenarios, the decision needs more evidence before it is made. Record which assumptions drive the difference, so that the discussion stays on facts.
- Define the planning horizon and the hardware refresh cycle for each option.
- Collect full quotes that include hardware, licenses, support and maintenance.
- Include power, cooling, rack space and network switches.
- Model expansion steps for compute and for storage separately.
- Add staff time for routine operations, upgrades and incident response.
- Include migration costs from the current systems and the risk of downtime during the move.
- Run a sensitivity check: what changes if license prices rise or growth is slower?
6. Common mistakes and what to ask suppliers
The most common mistake is comparing a firm quote for one option with a list-price estimate for the other. Use the same assumptions for both: the same redundancy level, the same performance target and the same support level. A model that gives the SAN a fully redundant design but gives HCI no extra capacity for resilience is not a fair comparison, and the result will mislead the decision.
Another mistake is ignoring the licensing structure. Per-core, per-node and per-capacity licenses grow differently, and renewal terms can change the picture after the first contract period. Ask what happens at renewal, whether support is required to receive updates, and how you would leave the platform if you needed to. Ask also for the migration path and for the exit terms in writing, before the contract is signed.
7. How to start: build a comparison from your own numbers
Start with your current inventory, your workload list and the date when your current hardware reaches the end of its support. Use your own figures for power, rack space and staff hours rather than generic benchmarks. Ask two or three suppliers for quotes built from the same specification, so that the differences you see come from the products and not from the assumptions.
Then build a simple spreadsheet with the cost areas described above, run the planning horizon and a stress scenario, and review the result with the people who will operate the system. The first concrete step is to write the specification: the capacity you need, the performance target, the redundancy level and the workloads that must share the same platform. Agree on that list with the operations team.
Frequently Asked Questions
Is HCI always cheaper than a SAN?
No. HCI often simplifies purchasing and operations, but licensing and step-based scaling can raise the cost for some workloads. A SAN can be more economical when storage grows faster than compute, or when an existing array still has useful supported life. Compare both options on the same specification and the same planning horizon.
Should we keep our existing SAN?
Often you can keep it for a period if it is still supported and meets your performance goals. Include its remaining support cost, the risk of reaching end of support and the migration cost when you model the alternative. Keeping an unsupported array is rarely a saving once the risk is priced in.
What is the most overlooked cost?
Staff time and migration effort. Routine tasks, incident response and upgrades take hours that are rarely written into the model. A migration from an older platform also requires testing, a rollback plan and a downtime window that the business has approved. Include all of these before comparing the totals.
Can a managed service provider reduce the TCO?
It can move part of the operational cost into a recurring fee, which can make budgeting simpler. Whether it reduces total cost depends on the scope of the service, the exit terms and how your internal costs compare. Ask for the scope in writing, including what is excluded and how changes are priced.
How should we account for power and cooling?
Use measured values or the manufacturer's ratings for your actual configuration, and apply your site's energy and cooling costs. Avoid a generic estimate, because the real load and the room conditions vary considerably between sites and change as the platform grows.