The year-on-year (YoY) growth of cloud storage and computing suggests an unstoppable trend of customers that have moved away from owning and running their own datacentres, compute, storage and networking hardware.
But although the cloud continues to grow – accounting for as much as 50% of the world’s IT infrastructure – the trend no longer only favours the cloud. Perhaps they’re a minority, but a significant number of organisations have gone in the opposite direction and brought storage and compute back in-house.
This “repatriation” of data and workloads from the cloud does not, however, mean cloud computing has fallen from favour. Rather, it is evidence the market has matured and that cloud is one of several valid ways to run IT.
Here, we look at some of the reasons customers repatriate IT from the cloud, and how they decide which data to keep there.
What is cloud repatriation?
Cloud repatriation is the process of moving workloads back from public cloud infrastructure to on-premises hardware. This could be to a business-owned datacentre, co-location or other shared facilities.
According to a survey by Citrix, 25% of UK organisations have moved at least half their cloud workloads back on-premise.
In some rare cases, customers might opt to bring only data and storage back in-house. This is usually because of specific concerns about data sovereignty, security and regulations.
More often, firms will repatriate the storage and compute architecture. This allows CIOs to keep a tight coupling between compute and storage, and potentially tap into performance benefits from optimising their technology.
Organisations might also opt to repatriate virtual machines, reversing earlier “lift and shift” cloud migrations.
Why do organisations do cloud repatriation?
Firms typically move workloads away from the cloud because cloud costs have failed to achieve the savings they expected.
“Cost concerns are a prime driver,” says Rahul Gupta, a cloud transformation expert at PA Consulting. “Many organisations are finding public cloud costs unexpectedly high at scale.”
In other cases, CIOs want a more stable environment – the constant development and innovation that characterises cloud platforms is not always an advantage – or because they have specific data protection, security or compliance requirements that are easier to meet with on-premises technology.
Then there are trends within the cloud computing marketplace that include price pressures from higher energy costs and rising demand, not least from AI. A shortage of higher-end processing power is pushing up costs here.
“There is upward pressure from inflation and from AI on cloud pricing,” says Adrian Bradley, head of cloud at KPMG UK. “That is changing the relative economic balance between cloud and on-premises, and therefore it’s rational for an enterprise to ask if they are getting the right value from their cloud estates.”
For other customers, however, the switch comes down to more fundamental reasons. CIOs believe they can better optimise their technology architecture in-house and make more of their investments. And, as the market matures, organisation see that some workloads are more practical and economical to run locally.
“The general tightening of datacentre capacity in the public cloud means the arbitrage has changed,” says Bradley.
How do you decide what data to repatriate?
After cost, the two main criteria for data repatriation are compliance and regulatory issues where organisations want specific assurances around data location and performance.
In highly regulated industries, such as healthcare and finance, (re-)locating data on premises might be forced on a customer. Performance is less clear-cut. Some organisations do run high performance workloads successfully in the cloud, but for others there will be bottlenecks.
“Certain workloads that require high performance and low latency, like market data, achieve better performance on premises,” says PA Consulting’s Gupta.
Sometimes, these bottlenecks can be reduced by redesigning the cloud infrastructure, or by taking advantage of the provider’s higher performance storage options.
In others, especially systems that depend on sensors or instrumentation – such as R&D or production line – bringing storage closer to the data source will bring clear performance gains.
What are the drawbacks of cloud repatriation?
The disadvantages of cloud repatriation are largely a mirror image of cloud’s advantages.
Organisations might need to scale their infrastructure rapidly, which is harder in-house. They need to buy equipment, often on long lead times, and find datacentre space and deploy it. Customers might also lack funds for capital investment or prefer to spend money on other priorities. And, as with any large-scale IT project there are costs as well as risks.
The chief “hidden” cost of cloud repatriation is actually quite well known. These are data egress charges where cloud providers levy hefty fees to move data away from their systems. But, these charges have become more transparent of late, and firms might still decide the cost is outweighed by the benefits.
As KPMG’s Bradley points out, egress charges only worry a minority of customers. But, for those it matters to it matters a lot. This is borne out by the UK CMA’s investigation into the cloud market.
“For most enterprises, egress costs are not a big issue,” says Bradley. “But for those where it is an issue, it’s a big issue.”
Enterprises might also find they pay more for licenses than they expected for on-premises technology, including for management tools and security. In the cloud these costs are typically bundled with the main product.
Staffing is another cost that organisations overlook. Does the enterprise have the skilled staff to optimise sophisticated, on-premises hardware, including the latest storage arrays? If firms have invested heavily in cloud skills, not all that expertise will carry across.
And that is why it is the more sophisticated and mature organisations that repatriate cloud workload, suggests KPMG’s Bradley. They have the skills to optimise their technology across both cloud and on-premise platforms, and to analyse the relative costs.
