Think carefully before considering cloud repatriation

Most IT leaders have assets moved to the cloud to achieve some combination of better, faster, or cheaper compute and storage services. They also expect to benefit from the expertise of cloud providers—expertise that isn’t easy for companies to develop and maintain in house, unless your company happens to be a technology provider.

“While computing power and hardware costs are lower on the cloud, your approach may not allow you to enjoy these savings,” explains Neal Sample, consultant and former CIO of Northwestern Mutual. “For example, if you move the front end of an application to the cloud, but leave the back end in your data center, then all of a sudden you’re paying for two sets of infrastructure.”

Another common reason companies are disappointed is they put information assets on the cloud in a “lift and shift” operation so applications never benefit from the advantages of cloud, such as elasticity. “A good elastic app doesn’t happen magically,” says Sample. “It needs to be written native for AWS or for another platform.”

The dilemma is you never really benefit from going to the cloud until you start using native functions. And even then, you can get trapped—not just to the cloud, but to a single cloud vendor. “There are a lot of differences between an AWS, for example, and an Azure,” Sample adds. “Using the native functions of one versus the other can lock you in. However, you won’t benefit from what the cloud has to offer until you re-architect your application for the cloud—and that means using native functions.”

A third reason companies are disappointed is because of a lack of control over their information systems. This is particularly pronounced in heavily regulated industries, such as financial services and healthcare, where companies can be held liable for non-compliance—nobody wants to trust a third party to keep them from legal troubles. Similarly, large data aggregators feel the need for control because they don’t want to leave their core business in the hands of a cloud provider.

Overall, disappointment comes from poor planning most of the time. Gartner has been offering advice in this area—most recently in The Cloud Strategy Cookbook, 2023—which can be summed up as: develop a cloud strategy, ideally before moving to the cloud; regularly update the strategy, keeping a record in a living document; and align your cloud strategy with desired business outcomes. Many companies that ignored this advice failed to reap the benefits of the cloud. As a result, some have decided to repatriate information assets, and too many of them do so with equally poor planning.

Repatriating is not for the faint hearted

Migrating back from the cloud is not an easy process—no matter what region you’re in. “Cloud repatriation is generally a last-ditch effort to optimize the cost structure of a business,” observes Sumit Malhotra, CIO of Time Internet in India. “But pulling off such a transition requires a deep technical understanding of the applications, skills in multiple technologies, and executive sponsorship of possible negative impact on user experience at the time of this transition. The journey is not for the faint hearted.”

It’s particularly difficult for smaller companies to repatriate, simply because, at their scale, the savings aren’t worth the effort. Why buy real estate and hardware and pay extra salaries only to save a small amount? By contrast, very large companies have the scale to repatriate, But do they want to?

“Do Visa, American Express or Goldman Sachs want to be in the IT hardware business?” asks Sample, rhetorically. “Do they want to try to take a modest gain by moving far outside their competency?”

Switching can also be complicated when the cost of change isn’t considered part of the calculation. A marginal run rate savings gained from pulling an application back on-prem may be offset by the cost of change, which includes disrupting the business and missing out on opportunities to do other things, such as upgrading the systems that help generate revenue.

A major transition may also cause down time—sometimes planned and other times unplanned. “A seamless cutover is rarely possible when you’re moving back to private infrastructure,” says Sample. “And that’s a really big concern in an era where 24/7 access is expected.”

Irrespective of the details, when a big name repatriates, word gets around. Dropbox made a splash when they migrated away from AWS storage service to their own custom-designed infrastructure starting in 2015. The company reported a cost of revenue savings of nearly $75 million starting in the first two years after the transition ($39.5 million from 2015 to 2016 and an additional $35.1 million in 2017).

More recently, in October 2022, web software company 37signals made news when its CTO and co-founder David Heinemeier Hansson wrote in a blog post that they’ll move their two main platforms—Basecamp (a project management platform) and HEY (a subscription-based email service)—off the cloud. However, they don’t intend to run their own data center, but rather work with a company that has carved out a niche providing a hybrid environment as a service.

“There are companies that specialize in this work,” Hansson says. “If your budget is of a size that this is appealing, that is, most likely, millions of dollars, you can afford to do this several times over with the savings you reap.”
Both Dropbox and 37signals have the motivation and capacity to make a switch since tech companies often rely more on compute and storage, and have a higher need for control and performance. They also have the expertise to pull off a reverse migration. Even though 37signals is working with to repatriate, the move back from the cloud will require significant changes to the apps and data structures to get similar functionality in the new environment—the kinds of changes not every company has the skills to make.

For the Dropboxes and 37signals of the world, the move might make sense. But for non-tech companies, the equation is different. The cloud is getting more efficient and cheaper in ways their private data centers could never match. As cloud providers become better, faster, cheaper, and more ubiquitous, doubling down on a temporary cost advantage might cause these companies to miss out on future proofing their applications.

Both tech and non-tech companies should be careful to avoid winding up with the worst of two worlds. This happens when they try to recreate cloud functions on-prem. “If you decide to repatriate, avoid the situation where engineering teams seek to imitate the public cloud environment when building on-prem counterparts,” says Malhotra.

The same kind of mistake in the opposite direction is often one of the reasons companies are disappointed in cloud services. This happens, for example, when a system that depends on an on-prem architecture, such as client server, is moved to the cloud without being redesigned. Applications written with an older, client-server architecture will wind up on the cloud with the processor in a different location than the database. The resulting latency could be unbearable.

A hybrid enterprise is often worse than either one that’s strictly cloud or strictly on-premises. “In a hybrid environment, web pages take longer to load, applications aren’t as snappy for clients, and batch processes take longer to run as they move data in and out of the enterprise,” says Sample. “If you haven’t redone your architecture, you may find that a hybrid environment is actually worse from a performance perspective.”

Two knee-jerk reactions don’t add up to good planning

“I think cloud repatriation will continue to happen, but it will be more like a ripple than an ocean wave,” predicts Sample. “Companies will continue to move workloads to the cloud without being ready to do so. Then they’ll be faced with the motivation to pull back.”

Over time, clouds will become easier to use. They’re already becoming more flexible, and cloud portability is more practical. And as cloud technology improves, repatriation will become even less attractive than it is today.

“The motivation that would turn this into a tsunami just isn’t there,” says Sample. “I’m sure repatriation will continue to happen, but only in spots. And all too often, it’ll be the result of poor planning.”

Cloud Management