When cloud first emerged as a practical solution in the 2000’s, it offered something revolutionary – an operational model that organizations could no longer afford to ignore.

Peter Sjoberg, Worldwide Vice President for Solution Architects at Cloudian, recollects, “prior to the cloud, if you wanted to solve any business problem with IT, it was a big project – a significant capital investment with substantial upfront spend just to see if it might work. The consumption economics of the cloud changed this, allowing you to test, pay by the month, prove it out, and only then move toward production.”

Today, the environment is shifting. Two forces in particular are driving a move towards repatriating data from the public cloud back to on-premises data centers: data sovereignty and public cloud cost. In addition, the growing desire to assert AI dominance is exposing vulnerabilities in a fully cloud-centric model.

There’s no place like home

IDC reported in 2024 that 80 percent of its survey respondents “expected to see some level of repatriation of compute and storage resources in the next 12 months.” Today, an overwhelming majority (93 percent, according to Cloudian) of enterprises are exploring repatriation. Many have already moved AI workloads from the public cloud to on-premises or private infrastructure, or are currently in the process of doing so. Similar findings have been reported by Citrix among US IT leaders, and by Barclays among enterprise CIOs.

Movement away from the public cloud is now driven in part by cost. The economics of the cloud are less favorable – and certainly less predictable – than many initially expected. With on-premises infrastructure, organizations are able to buy or lease equipment and maximize its utilization at less cost than cloud, while gaining the added benefit of greater cost predictability. Complexity plays a role as well. The sheer number of cloud deployments and accounts that organizations end up managing becomes a burden unto itself.

At the same time, increasing global regulation around data governance, protection, and sovereignty – GDPR being one of the most widely recognized – is exposing limitations in cloud data management. Where is it stored? How easily can it be accessed or retrieved? The concept of data sovereignty varies around the world, but in Europe in particular, maintaining full control over data is critically important, as Sjoberg explains:

“Many European customers have been telling us they no longer feel comfortable running their IT assets outside of the EU given geopolitical influences.”

“Furthermore, organizations that take governance seriously increasingly choose to exert greater control over their environments. The public cloud can make that more challenging, which is why repatriation has become a prominent trend today,” he says.

Shifting cloud economics

The cloud’s value as a low-commitment testing ground is real. But as workloads mature and scale, the economics tell a different story. Having worked extensively both in and outside the public cloud, Sjoberg knows this problem well. Cloud costs, he warns, grow increasingly difficult to control – and the picture shifts significantly depending on where an application sits in its maturity lifecycle:

“Early in an application lifecycle, the cloud gives you a chance to prove things before making the big financial commitment. And so, upfront, cloud can be very cost-effective.”

As workloads grow, however, the costs mount, making the cloud economics less attractive than first anticipated. Simultaneously, the organization’s knowledge of workload requirements grows as well: they know exactly the scale their systems need to reach. They are now able to size the requirements and to finance on-prem assets over an extended period at a more predictable, fixed cost.

“I saw this time and again – customers would build an environment, run it, and say ‘wow, this is running great, it costs $100,000 a month,’ which doesn’t seem like so much compared to a $3 million capital investment. But that $100,000 continues every single month for years, and over time, the total cost becomes very significant,” says Sjoberg.

Other cost factors, including ancillary fees such as egress charges, retrieval fees, or API call fees, are frequently underestimated, leading to billing surprises. Ultimately, it becomes a question of total cost of ownership that must be considered over time.

Sjoberg has also observed that operators frequently underestimate what it takes to manage cloud environments effectively.

“Even if you’ve been an IT professional for 20 years, engineering on-premises systems and becoming highly efficient at running them, it doesn’t necessarily mean you’ll be proficient with cloud tools.”

Charges for inadvertent capacity increases or compute instances left running can easily spiral out of control. For steady-state workloads, on-premises infrastructure is the more cost-efficient and predictable option.

Harnessing a cloud-smart approach

The cloud and on-premises infrastructure each have natural use cases. Cloud excels for development and testing, and for workloads where demand is bursty, seasonal, or unpredictable. For continuous, steady-state workloads – where capacity requirements are known, and volumes are stable – on-premises infrastructure consistently delivers lower total cost of ownership.

“If you’re looking to take advantage of the latest tools and capabilities – such as data analytics platforms or emerging AI models – cloud systems integrate with these tools very quickly and easily, so you can test and experiment. That’s why the cloud is often seen as the leading edge of technology,” Sjoberg explains.

The practical implication is clear: use the cloud where it makes sense – for development, testing, and variable workloads – and invest in on-premises infrastructure for the workloads that run continuously at scale. A well-managed hybrid model captures the advantages of both.

A cloud-smart approach means continuing to leverage the cloud’s distinct economic model but aligning it closely with business needs. It means evaluating new projects in the cloud as a testing ground – then letting them graduate to on-premises environments once proven. In this way, organizations can maintain the sovereignty and cost-effectiveness critical to their operations.

“We’ve moved beyond cloud-first to cloud-smart as the driving force in IT operations. I don’t see a future that is entirely one way or the other – fully on-premises or fully in the cloud. The future is hybrid,” says Sjoberg.

Data management in a hybrid environment

A successful hybrid approach hinges on disciplined data management. Organizations must maintain clear oversight of where data resides and how it is accessed. A hybrid approach may leverage traditional tiering, which keeps data accessible locally while moving inactive data to the cloud, where it can be stored cost-effectively at scale.

A second technique, reverse tiering, allows datasets that already exist in the cloud or in another on-premises location to be linked back to an on-premises system for centralized access and control.

For organizations looking to repatriate data from the cloud, Cloudian’s HyperStore platform is designed to make the process straightforward – eliminating the technical complexity and risk that can otherwise make data migration a costly and disruptive undertaking.

Whatever method is chosen, every pathway between systems must be secure by default, with encryption applied at rest and in flight. For example, through mechanisms such as Object Lock, Cloudian’s HyperStore object storage platform prevents unintentional deletion or loss, while audit capabilities provide full visibility into what has changed, when, and by whom – or even who has accessed the data.

“HyperStore supports this natively. It integrates on-premises volumes with cloud-based volumes, allowing data to be replicated for protection or tiered so that less frequently used data is stored in the cloud. It may be accessed on-prem, but it lives in the cloud,” emphasizes Sjoberg, adding:

“Using metadata with reverse tiering, HyperStore can recreate the data structure on-prem, so organizations can manage cloud-resident data as if it were local. From there, you can pull data back selectively or in bulk, effectively repatriating data on your own terms.”

For IT decision-makers, the defining principle is maintaining control from a single point. Whether data is replicated, tiered, or reverse-tiered, each investment can be aligned to the most cost-effective outcome. Ultimately, it comes down to one central question: what are you trying to accomplish?

Explore the survey “The great cloud rebalancing: Why enterprises are repatriating data from public cloud and what comes next” here.