Storage decisions have a long tail. Choose badly and you live with the consequences for years, because migrating high volumes of data is expensive, disruptive and rarely urgent enough to get funded until it becomes a crisis.
That makes it worth resisting the framing the market usually offers, a choice between “modernise to cloud” and “keep control on-premises”. The useful question is narrower: for this specific set of data, with this access pattern and these obligations, where should it live?
Choosing between data storage solutions: start with the data
Before evaluating platforms, characterise what you are storing. Five attributes determine most of the answer.
Access pattern
How often is it read, and how quickly does it need to come back? Data read constantly at low latency behaves very differently, commercially, from data written once and read twice a year.
Growth rate and predictability
Storage that grows 5 per cent a year predictably is a very different procurement from storage that might triple if a new product succeeds.
Regulatory and contractual constraints
Data residency requirements, client contracts specifying where data may be held, and sector rules on retention or recoverability. These are constraints, not preferences, and should be established before anything else.
Performance sensitivity
Whether the workload is latency-bound (transactional databases), throughput-bound (analytics, media) or neither.
Lifespan
How long you are obliged or intend to keep it, which for archival data is often measured in decades.
Only once you have that picture does a platform comparison mean anything.
When is cloud storage the right answer?
Cloud storage is the stronger option in several specific situations.
- When growth is unpredictable, because you buy capacity as you use it rather than forecasting it years ahead.
- For archival data, where cold tiers of object storage are low cost per terabyte.
- Where geographic distribution or resilience across regions would be expensive to build yourself.
- When the workload using the data already runs in cloud, because keeping storage and compute together avoids both latency and egress.
- When you want capability such as versioning, object lock immutability, lifecycle policies and cross-region replication without engineering it.
The operational argument is also real: no hardware refresh cycle, no capacity planning 18 months ahead, no data centre contract, and no five-figure emergency when a controller fails out of warranty.
When is on-premises storage the right answer?
On-premises remains the better answer for large, static, predictable datasets where cost per terabyte over a set period is the dominant factor and the capacity is known. It remains better for on-premises latency-critical workloads, particularly where a database and its storage need to sit close together. It is better where you have already invested in high-performance flash and the depreciation has not run out. And it is often better where data residency, client contract terms or sovereignty requirements make cloud placement complicated enough that the complexity outweighs the benefit.
There is also the egress question. Reading data out of a cloud provider costs money, and workloads that read heavily and repeatedly, or that move data between providers, can produce bills that bear no relation to the storage line in the budget. The UK Competition and Markets Authority’s cloud services market investigation found in 2025 that egress fees are a barrier to switching and to using more than one cloud provider.
“The number we ask people to work out is cost per terabyte over five years, including egress, for the workload as it actually behaves, not as the business case assumed,” says Tom Richards, Lead, Systems and Storage at Northdoor. “That single calculation settles most of these arguments, and it quite often surprises the person who was certain which way it would go.”
Why do many organisations end up with hybrid storage?
Hybrid is less a compromise than the natural result of having varied data. The transactional core stays close to the applications that use it; archives, backups and burst analytics go to cloud; and the two are connected.
What separates hybrid that works from hybrid that sprawls is placement discipline. Five things to establish:
- A written placement policy. For each class of data, where it lives by default and what would justify an exception. Without one, placement is decided by whoever provisioned it, and drift follows.
- A single view of capacity and cost across both halves. Two separate consoles produce two separate blind spots.
- Consistent data protection on both sides. Encryption, access control, retention and immutability should not be weaker in one half of the estate than the other.
- A defined path for data to move. Data that becomes cold should migrate to cheaper storage automatically, rather than through an annual manual exercise nobody has time for.
- Backup that respects the split. A cloud-resident dataset is not automatically protected against deletion or ransomware because it is in cloud. Immutable, tested backups matter wherever the data sits.
A short decision framework
Working through these in order usually produces a defensible answer.
- What must not move? Establish regulatory, contractual and sovereignty constraints first. They eliminate options rather than trading against them.
- What does the workload need? Latency and throughput requirements, measured rather than assumed.
- What does it cost over the life of the asset? Including egress, licensing, refresh, power, space and the staff time to run it, on both sides.
- How certain is the growth forecast? High uncertainty argues for cloud elasticity; high certainty removes much of cloud’s advantage.
- What can your team realistically operate? A technically optimal design nobody has the skills to run is not optimal.
- What does exit look like? For any option, including staying put, how hard would it be to change your mind in three years?
Modernisation and migration are not the same thing
Storage estates do need refreshing, and ageing arrays past their support window are a real risk. But “modernisation” and “move to cloud” are not synonyms. Replacing an old on-premises array with current flash, properly sized, is a modernisation. So is tiering cold data to cloud object storage while leaving the hot tier where it is. The objective is a storage estate matched to the data it holds, at a cost you can predict, with protection you can evidence. The same placement logic applies to compute, which we cover in our article on modern infrastructure and virtualisation.
What this means for your organisation
Storage decisions last five to seven years, so the time spent characterising your workloads before comparing platforms is the highest-return time in the whole exercise. Taking the time to understand your requirements properly can also reveal costs and considerations that are easily overlooked in an initial comparison.
Our data storage solutions approach helps organisations model the longer-term position, including the costs that rarely make it into the first draft, across cloud and on-premises platforms. If you are considering a storage refresh or reviewing your current strategy, talk to us about how we can help you assess the options and build a clearer five-year view.