Cloud Migration: The Complete Guide

Cloud Migration It involves moving applications, data, and infrastructure to a cloud platform. This typically involves moving from a company’s own data center or colocation facility to a public cloud such as AWS, Microsoft Azure, or Google Cloud. It can also involve moving from one cloud to another, or back to an on-premises environment. The latter is called repatriation.

Cloud Migration is different from application modernization. With migration, you change where an application runs; with modernization, you change the application itself. Many projects combine the two: first migrate, then modernize each application, or vice versa. You can read about how to decide which approach to take for each application in the guide at Application Modernization.

The most common reasons:

• the data center or the hardware needs to be replaced;
• VMware's license terms have changed since its acquisition by Broadcom;
• The load varies greatly, with peaks that you cannot handle efficiently in your own data center;
• You want to build AI applications. Gartner expects that by 2029, half of all cloud capacity will be dedicated to AI workloads, compared to less than 10% today;
• Teams want to roll out solutions faster and use managed services instead of managing everything themselves.

Yet the reality is often disappointing. Gartner expects that by 2028, 25% of organizations will be highly dissatisfied with their cloud adoption, and that by 2029, more than half will fail to achieve the expected results from multicloud. The causes are almost always the same: migrating everything one-to-one and running capacity 24/7, not assigning ownership of the costs to anyone, setting up the infrastructure after the fact, and forgetting about outbound data traffic and licenses. Section 12 shows how to prevent this.

In practice, hybrid is the norm: Gartner predicted in 2024 that 90% of organizations would be using a hybrid cloud by 2027. Multicloud requires more expertise and management than it appears. Choose it for a specific reason, such as a service offered by only one provider. As a general safeguard against lock-in, it is expensive. The difference between hybrid and multicloud is also explained in the frequently asked questions.

Start by taking inventory of applications, dependencies, usage, licenses, and owners. Then, for each workload, assess the following:

Start with workloads that have few dependencies and aren't business-critical. That way, your team can get familiar with the new environment before the heavy-duty systems are migrated.

That choice has changed over the past two years. Digital sovereignty is now on the agenda of executives and regulators, and Gartner expects European spending on sovereign cloud IaaS to rise by 83% to $12.6 billion by 2026. In July 2026, the Dutch government tightened its cloud policy: anyone using the public cloud must justify the risks and dependence on a single provider and have an exit plan. For critical tasks, the government advises against using providers subject to non-European legislation.

You make that decision on a per-dataset basis. A lot of data can remain with a hyperscaler in a European region, with encryption using your own keys and solid contracts. The most sensitive data requires a European or on-premises environment. A Data Sovereignty Assessment can map this out in a week.

A Landing Zone is a pre-configured base in the cloud: account structure, identities and roles, network, central logging, security settings, and cost guardrails. Set it up before migrating your first workload. Doing it afterward costs much more, because you’ll have to migrate existing workloads without interrupting them.

• Define the Landing Zone as code—for example, using Terraform, Bicep, or CloudFormation—with version control.
• Use your cloud platform's standard as a starting point: AWS Control Tower or Landing Zone Accelerator, Azure Landing Zones, or the Google Cloud blueprint.
• Manage tags and budgets from day one so you can see the costs per team or application.
• Link identities to your existing directory, such as Entra ID, and enforce MFA.

Learn more about SUE's approach on the Landing Zone page.

Since Broadcom’s acquisition of VMware, the licensing terms have changed, and many organizations are looking for an alternative. vSphere 8 and ESXi 8 will be supported through October 11, 2027. Gartner expects that by 2028, more than one-third of VMware workloads will be running on a different platform, and warns that a complete migration will take three years or longer.

Start with the applications you were already planning to modernize and migrate the rest in phases. At NXP, SUE chose Apache CloudStack, which reduced virtualization costs and provided strategic flexibility. Linux alternatives to vSphere are covered in the guide on choosing a Linux distribution, and virtualization on Kubernetes is covered in the guide on Kubernetes.

Plan for three to five years and factor in the costs of doing nothing, such as hardware replacement or a VMware renewal. For larger migrations to AWS, the AWS Migration Acceleration Program can cover a portion of the migration costs. After the migration, FinOps determines whether the business case pays off: at Colect, a three-day analysis resulted in a 43% reduction in monthly AWS costs. Learn more in the guide on reducing cloud costs.