Stop treating AWS like another data centre

An enterprise in the financial services sector anticipated cost savings when transferring a key application to the cloud. However, their planned "lift-and-shift" migration could result in an annual expenditure of $4.5 million. This revelation has prompted many businesses to question the efficacy of utilizing hyperscale cloud infrastructure.

Nonetheless, a distinct strategy managed to cut that bill to $300,000. The issue originated from a fundamental misinterpretation: the public cloud is not merely another data centre, and cost savings materialize when concentrating on an application's final state and business value. Cloud expenses continue to perplex companies, leading numerous to drastic measures like repatriation (48% of respondents in a 2025 Cloud Computing Survey from Foundry said they are repatriating due to costs).

Others merely tolerate the financial impact, justifying benefits elsewhere. However, most made the same error, squandering budgets by treating public clouds like AWS as another data centre, according to Jonathan Leibbrandt, Senior Solutions Architect at TrueMark. "When organizations move to the cloud, they want to provision it like for like, just as they have it in a data centre.

If they have a virtual machine running in their data centre, they think, 'We need to get the exact same virtual machine, same spec, and everything running in AWS.'" This approach is not optimal. When migrating applications to a public cloud provider, they retire the original servers. To avoid complications, they replicate the original server's hardware specifications.

However, this strategy overlooks the most critical distinction between purchased servers and cloud infrastructure. In data centres, you are investing in assets intended to last several years, considering peak capacity, future growth, and potential application needs before the next hardware refresh. Once you've purchased that capacity, it's no longer flexible.

You've made the investment, and you're responsible for it. If the server remains at 10% utilization most of the day, you've already paid for it. The cloud paradigm has shifted from buying capacity to consuming it. Previously, you had a three-to-five-year hardware cycle. Now, you don't have the same constraints. The mindset of procuring capacity persists even in a consumption model.

This mindset often seems simpler, as companies avoid the complexity of refactoring and rearchitecting established and customized applications. Migration projects involving hundreds of servers and cutovers often reinforce the like-for-like approach. But these choices result in additional costs. Leibbrandt explains the ramifications of this approach.

A company decides to migrate 2,000 virtual machines to AWS, creating 2,000 EC2 instances and running the applications in the same manner. However, this capital expenditure logic leads to rapidly eroding margins. If you have 2,000 servers running 24/7, you are paying for something that you might not even be using 24/7. What is the alternative?

In this scenario, Leibbrandt recommends transitioning to an event-driven architecture centered on serverless technologies. Infrastructure only spins up to complete a transaction and then shuts down again, minimizing idle costs. You only pay for those few milliseconds required to execute that transaction. This highlights the trap of treating public clouds like AWS as another data centre, as it is not.

It represents a different operating model from the predict-buy-sweat procurement cycle, necessitating a different approach. Leibbrandt advises companies to consider the following when planning a migration: Most businesses view the public cloud as another data centre, leading to significantly higher expenses. They should not, and if they adopt the correct approach from the outset, they will not.

They will realize substantial savings and an investment that genuinely evolves with them. Replicating a server may appear tempting, but is it the best choice? That is the question to answer when migrating to the public cloud.