There are right and wrong uses for every type of cloud infrastructure. Dedicated servers have their own particular set of characteristics that make them right for some applications and wrong for others.
An awful lot of what you can read online about the benefits of dedicated servers over cloud servers, or vice-versa, is really fluff that could apply to either solution with the right configuration. So, when do the unique aspects of dedicated, physical servers transfer benefits to your application?
When do you need a dedicated server?
When you Need Predictable Performance
Virtual Private Servers (VPS') can all suffer from the noisy neighbour effect.
The cloud, where VPS' are created, is a multi-tenant environment in which multiple users or applications share the resources of the host infrastructure. These must all contend for access to the available bandwidth, CPU, I/O, etc. When one application becomes too greedy, it can negatively affect availability of those resources to others, resulting in their uneven performance.
In theory, the number of Virtual Machines (VM) on any one host and the specifications thereof can be managed to minimise the risk of impact from noisy neighbours, allowing enough headroom for peaks in demand for resources. In reality, it's all but impossible to fully protect against it in shared environments.
A dedicated, physical server is immune to the noisy neighbour problem since its resources are 100% dedicated to you.
If your own applications exceed the available resource within your dedicated server, or the network environment around it, it is still possible that you will experience uneven performance, but both the cause and the solution lie in your hands.
When you are I/O Intensive
Optimising your server environment for IOPS is essential in business where storage is used intensively and frequently. In Big Data, analytics or Internet of Things applications, for example, high I/O throughput it a must.
The exclusive access to server resources, interconnects and local storage that dedicated servers provide guarantees the maximum IOPs possible for the specification.
The absence of any hypervisor, contention or noisy neighbours ensures your access to IOPs is predictable and reliable.
When you Want Full Control
When operating on virtualised resources, the hypervisor sits between you and the host machine, limiting your choices and the extent of your control.
With a dedicated server you get admin level access to the bare metal machine. You have the ability to choose and install your preference of OS and configure it in any way you like.
You can even install your own hypervisor if you wish and create Virtual Machines on your own terms.
When you Need to be 'Always on'
Virtualisation has enabled dramatic changes to the way that IT infrastructure is costed out. Utility or on-demand pricing is now commonplace. In the case of public cloud giants like AWS or Azure, significant un-bundling of pricing has taken place, where, for example, bandwidth and CPU cycles can be billed separately.
These pricing innovations, however, tend to favour users with significant fluctuations in demand. In the case of always-on instances, dedicated servers often cost significantly less than like-for-like VMs.
If any of these describe critical aspects of your infrastructure requirements, you need dedicated servers.