Why Some Workloads Are Coming Home: The Case for Cloud Repatriation

The Driving Force Behind Cloud Repatriation

Although only 8% to 9% of organizations intend to implement full workload repatriation according to IDC’s recent Server and Storage Workloads Survey, the cost and regulatory challenges driving repatriation are real.

“What we hear a lot is that the unpredictability of the cost for some workloads in the cloud has become untenable,” Gordon says. Teams also need to make room in their budget for AI infrastructure.

Organizations using “private AI” applications are likely turning to on-premises infrastructure, says Rob Tiffany, research director in IDC’s worldwide infrastructure research group and part of the cloud and edge services practice.

“They're going to have large language models or small language models running on their own gear, and training or fine-tuning those AI models with their own private corporate data,” Tiffany says. These companies are hesitant to share their LLMs with AI vendors.

FIND OUT: How to navigate cloud migration and modernization projects.

How Heavily Regulated Industries Are Moving Data Workloads

Data sets are governed by strict compliance or data residency regulations, potentially leading companies to repatriate and move IT networks back on-premises.

Compliance requirements are also a big factor, particularly in financial services. “Although there might be some experimenting in the cloud, some of these more regulated industries are thinking about how to keep production centrally in their data center, for security and availability,” Gordon says, which means the most valuable or most sensitive data could end up back in the data center.

Tiffany recommends keeping sensitive data in public cloud infrastructure and then running large workloads in Microsoft Azure or Amazon Web Services. Application Programming Interfaces make this integration between the public and private cloud possible.

How to Evaluate Predictable Workloads

Repatriation can also help teams handle traditional AI workloads that are resource-heavy, says Gordon.

“If it's a predictable workload that's running the same way all the time, it probably makes more sense to run it on-prem versus some burstable system where I need to burst to a lot more servers dynamically, because maybe e-commerce is doing a sale,” Tiffany says. 

In fact, some of the largest enterprises keep their critical infrastructure in the data center to be more efficient operationally and financially, says Gordon.

Click the banner below to gain efficiencies by keeping some applications on-premises.

How Industrial IoT and Retailers Are Repatriating at the Edge

The Industrial Internet of Things and retail are two areas where teams experience issues with latency, skill sets and power requirements that could impact repatriation.

“Since there's that skills gap, the one way to make repatriation easier is to embrace hyperconverged infrastructure that is similar to public cloud in the way it works and scales and how you manage it,” Tiffany says. “HCI behaves and looks like the public cloud, and so in terms of skill sets, people are more likely to be able to work with that,” Tiffany says.

Gordon notes that cruise ships or floating hotels have infrastructure that must be in a data center due to its “far-edge environment.”

Manufacturers preparing for Industry 4.0 may also want to keep compute, storage and networking on-premises. They may store their infrastructure on a factory floor or in an adjacent data center.

Repatriating cloud applications can also ease network disruptions — a benefit for IoT. Dell, for example, has designed its PowerMax storage arrays to prioritize resiliency and has worked with hyperscale providers such as Google to bring AI models like Gemini on-premises.

EXPLORE: The benefits of on-premises IT infrastructure.

How to Handle Cloud Repatriation

A tech partner such as CDW can help teams modernize on-premises infrastructure using HCI and software-defined data centers. Experts can also run cloud infrastructure assessments to determine which workloads are ideal for repatriation, and whether teams can run AI applications as is or need to repatriate them. NetApp and Pure Storage are two companies that offer high-performance storage options for AI workloads.

“Scale your compute and storage separately and don’t have lock-in to any single vendor,” Gordon advises.

READ MORE: How the Houston Texans tackled their hybrid cloud strategy.

“In a hybrid strategy, for starters, make sure those workloads are either using traditional virtual machines or containers like Kubernetes, so that they're more portable from the get-go,” Tiffany says. “Then make sure that you actually have a legitimate hybrid cloud infrastructure between your public and private clouds so that those workloads can move back and forth freely as needed.”

For example, a Microsoft database workload in SQL server, running on-premises, could sync data with an Azure SQL Server, according to Tiffany. This kind of “disaggregated model,” according to Gordon, allows teams a bit more flexibility as they restructure workloads.