The public cloud is growing almost exponentially. According to Forrester Research, the public cloud services market is expected to grow to $1 trillion worldwide by 2026. Its primary value propositions are simplicity, speed and easy access. All you have to do is wave a magic credit card and voila! Cloud services of all flavors and with incredible scalability are available courtesy of the hyperscaler cloud providers.

“The public cloud market will consolidate its dominance of IT by 2026,” said Lee Sustar, an analyst at Forrester. “But getting there will be quite a ride.”

Part of that ride now includes workloads and applications flowing the other way. Instead of in-house data centers bleeding applications and functions to the cloud, there has been a noticeable uptick in the reverse: workloads being repatriated from the cloud to on-premises systems.

“The issue is that cloud services do not scale cost effectively,” said Rich Gadomski, head of tape evangelism at Fujifilm Recording Media USA and co-chairperson of the Active Archive Alliance. “That is why many corporations are repatriating some or even all of their long-term data to on-premises, private cloud or hybrid-cloud solutions.”

According to Enterprise Strategy Group (ESG), 60% of companies have repatriated some workloads from the cloud back on-premises. That said, there is no stopping the unrelenting growth of the cloud. Yes, workloads are being repatriated to the data center and on-premises, but many more are going to the public cloud. It is a case of sharply defining what goes where.

Here are four factors for organizations to consider in deciding which workload goes where, followed by some best practices for the repatriation process.

Factor 1: Cost 

The alluring promise of the cloud was vast cost reduction. This certainly panned out at first. But as the amount of data and applications in the cloud rises, costs tend to mushroom — and sometimes spiral out of control.

“Following multiple years of 30% to 50% growth in the major public clouds, we expect to see that slow as more companies look to optimize their workloads and costs,” said Anand Babu Periasamy, cofounder and CEO at MinIO.

Tools are emerging to provide a tighter rein and more accurate estimates of overall cloud costs, as well as decent forecasts about future costs and assessments of cloud TCO for existing and new cloud applications. Such data should be front and center when it comes to determining which workloads should be sent to the cloud and which would be cheaper on-premises.

Some workloads are perfect for the cloud. Others, not so much. If the data is written once, sent to the cloud, and rarely if ever accessed again, cloud costs will be low. But if information is continually accessed, passed from one application to another, and data is regularly dragged back from the cloud for analysis, costs may rise considerably.

One tool that has gained traction as a way to achieve better governance of cloud usage and costs is the FinOps framework. It brings a level of financial discipline, cost management and accountability to the cloud.

According to J.R. Storment, executive director of the FinOps Foundation, FinOps improves the unit economics of cloud spend by delving into billing data and metadata through the use of machine learning to analyze spend and optimize it.

“People do not necessarily want to just save money, they also want to fine-tune operations, move faster and improve quality,” said Storment. “It is challenging to map cloud spend to business value as huge charges can be incurred without POs.”

Factor 2: Data sovereignty and privacy regulations

Data sovereignty is a serious business. Privacy laws such as the EU GDPR and others, which state that data must remain within a country, state or region as a way to maintain privacy, can result in major fines for the illegal movement of data out of certain geographical regions. GDPR fines are now up beyond 4 billion euros. Meta has been hit with more than 2 billion euros in fines in 2023 so far.

When the cloud is involved, it is all too easy for data to be shunted anywhere across the planet for the sake of convenience. Hence, highly sensitive data may be better housed in a data center in certain regions than being forwarded to a geographically nebulous cloud. “Cloud storage may impact data sovereignty regulations for which a corporation is legally responsible,” said Gadomski of Fujifilm and the Active Archive Alliance.

Factor 3: Security 

According to 451 Research, nearly half of organizations have experienced a cloud-based data breach or failed audit in the past 12 months. Part of the problem is “out of sight, out of mind.” When data is in the cloud it is easy to decide that its security is someone else’s problem. Certainly, the cloud providers share a responsibility. But they define in their terms of service what their security responsibility is and what is the user’s. At times it can get a little muddy. It’s up to the owner of the data to ensure that all possible protections are given for data in the cloud. Because of this, the needs of security and compliance may dictate that certain data and applications belong on-premises only.

Factor 4: Latency

Certain workloads or data transfers may have time-sensitive requirements. For example, financial services firms that deal with millions of transactions every hour can’t have latency above about 2 milliseconds, and sometimes they need sub-millisecond responses. Use of the public cloud makes those kinds of speeds virtually impossible.

When all factors are taken into consideration, the round-trip time between sites 500 miles apart amounts to more than 12 milliseconds using good equipment and optical fiber. Double that distance and it rises to more than 21 milliseconds. If latency is a concern, it’s best to keep data in-house, served by top-of-the-line equipment, and place your systems as close to users as possible.

Best practice: Plan carefully

When repatriating, careful planning is essential. There are examples of companies like Dropbox finding success and savings in bringing a ton of data back from their long-term Amazon Web Services (AWS) location in an effort to gain control over costs. Similarly, poor initial planning or irrational executive exuberance about going all-cloud has caused plenty of workloads to be sent to the cloud that really didn’t belong there. Therefore, it is wise to avoid drastic all-or-nothing policies. Being didactic about all-cloud or all on-premises is likely to be misguided. Consider all factors, send what belongs there to the cloud and keep the rest in the data center.

Similarly, take the time to carefully plan all aspects of any cloud repatriation. Take an inventory of data, apps and workloads that are likely to be touched by the migration. Pay attention to data formats and platforms, as well as other technical aspects that can derail a migration. Repatriating data from the AWS public cloud to a private cloud backed by Azure may lead to some compatibility challenges. Know before you go. If format conversion is needed, find the right tools to carry it out smoothly. A VMware Converter, for example, may be needed to ensure all VMs are re-created correctly after the migration.

And when things get complicated, don’t be afraid to bring in outside help from vendors or third parties who specialize in migrations. In these days of containerization, virtualization and multiple clouds, cloud repatriation may become a little tricky. The likes of Dell Technologies, HPE and local specialists can be found to streamline repatriation efforts.

Best practice: Hybrid is probably best 

For some, 100% on-premises will make sense or will be a regulatory necessity. For others, all-cloud may be perfectly fine – especially startups with no previous investment in equipment or legacy applications. But for most, a mix of workloads is best. Let the cloud have what makes sense (and is affordable) and keep the rest in-house.

“It's not all coming back, and it is not all going to the cloud,” said Derek Dicker, CEO of Nyriad. “Hybrid is here to stay.”