For nearly a decade, enterprise IT leaders have followed a cloud-focused approach, leaning on massive public cloud infrastructure to power their workloads. However, as the tech industry matures, the problems with cloud infrastructure not only become more apparent but also more expensive. As a result, IT leaders are looking beyond these traditional set-ups for more robust solutions.
I’ve been seeing this for many years, but recently the signs of an exodus of enterprises from the public cloud became hard to ignore. A recent Barclays study found that 86% of CIOs plan to enact cloud repatriation, moving part of their public cloud workloads to private or on-premises infrastructure. This number will likely grow even more in the coming years as leaders shift their mindset from cloud-first to cloud-appropriate.
This isn’t a full rejection of the model; it’s simply an adjustment based on lessons learned over years of running production environments in the public cloud. Most often, problems with egress fees, performance degradation, security risks, and compounding costs are key drivers in considering public cloud alternatives.
In the face of these persistent issues, solutions such as switching to fully dedicated servers – also known as “bare metal” – are top of mind for IT leaders seeking to tackle them all at once. However, the steps to a successful migration aren’t clear to most professionals. Leaders need to take care in planning their exit from the public cloud, and the best way to start is by measuring the whole impact the public cloud has on their operations.
See also: Scaling AI from Pilot to Production: A Roadmap for Enterprise Reinvention
Egress fees scale silently but compound heavily
While public cloud providers offer flexible pricing and scalable resources, their models often introduce hidden complexities and significant fees. This is particularly apparent when considering the complexities of network traffic charges.
While ingress is often free or only incurs a negligible cost, data leaving the network is heavily metered. Early in the cloud era, public providers lured enterprises with promises of cheap or free ingress, but egress charges began to pile up, especially for enterprises that ran data-intensive workloads, such as real-time bidding systems, analytics pipelines, or content delivery.
Overall, public cloud bills fluctuate in ways that fixed infrastructure doesn’t. At some point, the cost of renting infrastructure far exceeds that of building and managing it yourself, and the answer becomes obvious.
Workloads are too demanding for shared infrastructure
At its core, public cloud infrastructure is shared infrastructure. Your key capabilities are fighting for space on the same physical hardware as hundreds of other enterprises. This model is fine for many production workloads, but latency-sensitive, compute-intensive, or I/O-heavy applications can experience significant problems when the cloud gets too crowded, and resource contention takes over.
Storage is where this problem compounds the most. Cloud storage services often enforce IOPS caps that become heavy constraints during real production loads. Just because an application runs smoothly in a staging environment doesn’t mean it’ll work when fully deployed, and debugging these problems in a virtual environment is never easy.
For work where sub-millisecond latency matters, the virtualization tax isn’t just overhead; it’s a fundamental problem. Certain workloads were never designed to occur on shared infrastructure; they need dedicated, predictable hardware in the first place.
Security is not one-size-fits-all
In theory, the shared responsibility model of public cloud infrastructure seems sound. However, in practice, enterprises have only limited control over the underlying environment and are forced to design specialized security functionality on top of what’s provided.
For organizations in heavily regulated industries, such as healthcare or financial services, this creates significant drag as custom security tools are layered on top of the base infrastructure, necessitating cloud-specific documentation and specialized management.
Data residency complicates these issues even further. Regulations often stipulate where data physically resides and the jurisdictions through which it transits. This creates a key bottleneck for the data centers that are valid in the first place, which can cause an organization to make concessions on otherwise key capabilities.
Where enterprises need to start
Organizations that avoid these problems design infrastructure for flexibility, enabling future transparency and predictability rather than continuing to rely on what was easiest at the start.
For workloads where performance, cost predictability, and security are paramount, bare metal infrastructure may be the best option. Dedicated hardware circumvents virtualization overhead, completely eliminates performance bottlenecks caused by demand-heavy neighbors, and makes isolationary compliance a breeze.
Architecture is the first step. Selecting based on your current needs can be important, but planning for the future will be even more valuable, as most bare metal offerings don’t offer the same scaling flexibility as the major cloud providers.
Vendor selection is also key. The best partners offer a wide array of deployment options, locations, complementary solutions, and modern developer tools, allowing the freedom to customize workload placement on the fly. Vendors that offer transparent pricing should be prioritized even further, including real conversations about cost, not just estimations.
Infrastructure strategy’s final form
The current industry trends indicate that infrastructure strategy is finally maturing. Enterprises are realizing that the benefits of a fluid approach to cloud infrastructure are too high to ignore.
While cloud providers make access to proprietary services easy, hindsight is 20/20. Those who maintain discipline and don’t default to simple solutions preserve the freedom to quickly adapt their infrastructure as needed, while harvesting the operational benefits of optimal infrastructure.
Setting up hybrid and multi-cloud architectures becomes easier each quarter, especially when enterprises rely on specialized connectivity vendors that already have private backbones in place to interconnect multiple vendors and their native solutions.
The public cloud just isn’t cutting it anymore as a one-size-fits-all approach, but other solutions, such as switching to bare metal and combining it with network-as-a-service offerings, can provide the breathing room enterprises need, enabling workloads with sub-millisecond latency, eliminating virtualization overhead, and ensuring airtight compliance.
If your enterprise is experiencing heavy drag due to public cloud failures or your margins are suffering from its high costs, it may be time to take a new direction and reclaim the high performance and cost predictability your operation truly deserves.
