The Cloud Bottlenecks No One Warns You About Until It’s Too Late

This Industry Viewpoint was authored by Jennifer Curry Hendrickson, Senior Vice President of Managed Services, DataBank

Enterprise IT leaders have spent the better part of a decade being told that cloud is the answer. For many use cases, it is. Yet, a pattern has emerged across the industry that deserves more direct examination than it usually gets: Organizations are moving workloads back out of public cloud at a rate that suggests the original decisions weren’t as sound as they appeared.

According to recent Barclays research, 86% of enterprise CIOs now plan to move at least some public cloud workloads back to private cloud or on-premises infrastructure. This trend isn’t a complete rejection of the cloud, though. Instead, it’s evidence that organizations are refining their strategies based on real-world experience.

Understanding why it’s happening, the true price of reversing course, and how to avoid the problem in the first place is worth the time of any infrastructure or network leader making decisions today.

When the Economics Stop Working

The initial promise of cloud was predictable, scalable, pay-as-you-go economics. What many organizations found instead were pricing structures that reward adoption and penalize scale. Providers offer calculators that appear straightforward, but actual costs diverge from projections as workloads grow. For example, autoscaling adds compute during demand spikes, but it also generates charges for load balancers, monitoring tools, inter-zone data transfers, and a growing list of supporting services that may not have been clear in initial planning conversations.

For network-intensive workloads in particular, the math changes significantly at scale. A deployment that looks manageable at $10,000 per month can reach $20,000 once variable charges accumulate. Finance teams trying to forecast IT spending find that public cloud bills fluctuate in ways that fixed infrastructure costs simply don’t. At some point, the total cost of ownership for public cloud exceeds what dedicated infrastructure would cost, often by a margin that’s hard to defend.

The compliance burden compounds the problem. Regulated industries — such as healthcare, financial services, and federal contractors — often discover that meeting HIPAA, PCI DSS, or FedRAMP requirements in a public cloud environment requires layering multiple third-party security tools on top of a provider’s baseline.

The shared responsibility model sounds straightforward in theory. In practice, customers inherit only limited controls and must build cloud-specific security expertise on top of whatever the provider covers. For many organizations, this overhead eventually costs more than the infrastructure it was meant to simplify.

The Exit Isn’t Free

Here’s where the conversation typically gets uncomfortable. Organizations that recognize these problems and decide to act often find that moving out is more complicated than moving in.

Data egress fees are the first shock. Ingress is typically free or minimal, a deliberate design choice that makes migration into cloud easy. Moving data out carries per-gigabyte charges that many organizations have already experienced at smaller scale during analytics processing or backup operations. When repatriation requires extracting years of accumulated data, those charges can reach tens or hundreds of thousands of dollars depending on volume.

The operational friction runs deeper than most expect. Applications built on managed cloud services often have deep dependencies on proprietary APIs, event formats, and scaling logic. What looked like a convenient abstraction during development becomes an architectural constraint during exit. Teams that need to refactor those dependencies may spend months on work that delivers no new capability, only the freedom to move.

For mission-critical systems, there’s no clean cutover. Organizations must maintain duplicate environments during transition, paying for both the infrastructure they’re leaving and the new environment they’re building until the latter proves itself under real-world conditions. A financial services firm can’t simply cut over a trading platform and hope for the best. In many cases, these  parallel operations extend both the timeline and the total cost well beyond initial estimates.

The Better Path Is Designed In, Not Bolted On

The organizations that avoid these problems made a deliberate choice early on: They designed their infrastructure for flexibility rather than convenience, keeping future options open rather than defaulting to whatever was easiest in the moment.

Vendor agnosticism isn’t about rejecting cloud providers or their ecosystems. It means making deliberate trade-offs about portability from the start — favoring containerization with Kubernetes over proprietary orchestration, open-source databases over managed services with limited export options, and infrastructure-as-code tools that work across environments. At the application level, it means avoiding tight coupling to proprietary APIs and documenting dependencies, so future teams understand the cost of changing course.

Workload placement deserves the same rigor. Public cloud is well-suited for applications with unpredictable demand, development and testing environments, and workloads requiring global distribution. The picture changes for stable production workloads, latency-sensitive applications, data-intensive systems generating significant egress costs, and heavily regulated environments. An honest total cost of ownership analysis — one that factors in data transfer, compliance overhead, and long-term flexibility — generally tells a different story than early projections did.

Vendor selection matters as much as architecture. The right infrastructure partner offers a full range of deployment options across colocation, private cloud, bare metal, and managed services under one roof, with cloud connectivity built in. These options give organizations the freedom to place each workload where it makes the most sense, without starting over. Any infrastructure partner worth serious consideration should offer transparent pricing, including candid conversations about worst-case cost scenarios.

Infrastructure Maturity Looks Like This

The rise of cloud repatriation isn’t a verdict on cloud computing.  It’s a sign that infrastructure strategy is maturing, with organizations learning, sometimes at significant cost, that no single model works for every workload at every stage of growth.

The lessons are learnable before the pain sets in. Portable architecture, deliberate workload placement, transparent partner relationships, and regular reassessment are not complicated in concept. They require discipline to maintain when cloud providers make proprietary services attractive and convenient. Yet the organizations that maintain that discipline preserve something valuable: the freedom to adapt when business requirements change, without facing a migration project they didn’t budget for and a bill they didn’t see coming.

# # #

About the author

Jennifer Curry Hendrickson is the Senior Vice President of Managed Services at DataBank, where she leads the company’s product and technology teams. In this role, she oversees the architecture, design, engineering, service delivery, and support of DataBank’s managed services and network offerings. With nearly two decades of experience in technology and customer operations, Jen has held senior positions at INAP, Zayo Group, Latisys, and Level 3 Communications. Her expertise in network and cloud services is instrumental in advancing DataBank’s infrastructure to support emerging technologies like AI and Web 3.0. Jen holds a Bachelor of Business Administration from the University of Notre Dame.

If you haven't already, please take our Reader Survey! Just 3 questions to help us better understand who is reading Telecom Ramblings so we can serve you better!