The question on whether public cloud infrastructure is cheaper than running on-premises data centers keeps coming in client inquiries. Clients realize that most of the answers produced by the industry so far are skewed by the vested interests of whoever is coming up with those answer. Public cloud providers make their offerings look significantly more cost-effective than on-premises data centers. Hardware vendors promote the opposite view. Furthermore, within organizations themselves, internal politics continues to inevitably influence the results of any attempt to produce defensible calculations.
That’s why we decided to take a shot at answering this. I’m proud to announce that my research note “How to Develop a Business Case for the Adoption of Public Cloud IaaS” (paywall) is now available on gartner.com. The research provides guidance on how organizations should go about calculating TCOs and ROIs for their cloud adoption and migration projects. Gartner clients often struggle to quantify the cost savings that the cloud model can lead to as well as the potential for new revenue opportunities. As a results, clients often end up calculating cloud costs with the same buying patterns as they were using in their data centers, missing out on the optimization opportunities that public cloud infrastructure can offer. At the same time, clients struggle to quantify the necessary investments to skill up and operationalize cloud to take full advantage of the technology.
The research states that “cloud services can initially be more expensive than running on-premises data centers. [However, it also proves that] cloud services can become cost-effective over time if organizations learn to use and operate them more efficiently.” The statement is backed by an example of workload migration for 2,500 virtual machines from an on-premises data center to Amazon Web Services EC2. The example TCO (shown in the figure below) shows an initial uptake in cloud costs and a steady decline as soon as organizations learn how to apply cost optimization best practices (as described in this other framework). The chart also shows how on-premises costs may have a long tail as organizations take time to actually shut down their data centers.
While the savings on infrastructure costs over time may look appealing, organizations should bear in mind that the overall ROI may be still negative in the short term due to the hefty investments in transformation and the long tail of on-premises data center costs. Furthermore, the example in Figure 1 is based on a number of assumptions (available in the research for consultation) that will not be representative of all situations. As a consequence, organizations that want to conduct a similar exercise should be prepared to tailor the assumptions, being aware of their impact on the final business case result.
To know more about this topic or if you would like to discuss further, you can read the research note at”How to Develop a Business Case for the Adoption of Public Cloud IaaS” (paywall) or reach out to your Gartner representative to schedule an inquiry call with me. Looking forward to hearing your comments!
The Gartner Blog Network provides an opportunity for Gartner analysts to test ideas and move research forward. Because the content posted by Gartner analysts on this site does not undergo our standard editorial review, all comments or opinions expressed hereunder are those of the individual contributors and do not represent the views of Gartner, Inc. or its management.
Hi Marco, I like your blog, I am keen to understand your take, how mature do you think PaaS and DBaaS are today?
Sorry, I do not buy the chart. Every company that I know of that started using the cloud, it’s monthly bill goes through the roof. And it is so difficult to predict next month’s bill. Not to mention that now, ANY developer that makes a small dev error, can cost a company unbelievable amounts.
And we didn’t consider that the move is single sided. Don’t even think about going back. It’s impossible.
I would bet that most board of directors follow the hype. “We must move to the cloud, everybody moves to the cloud, did you move to the cloud already?”
Uncontrolled growth in cloud bills is indeed a problem for many organizations starting with public cloud adoption. However, that’s due to lack of financial management processes. Spend can (and should) be managed and developers’ errors can be prevented and/or mitigated. Repatriation as a way to address the inability of managing cloud spend is not something we’d even recommend considering.
I have been saying that from the beginning of so called cloud fashion. Honestly, moving to cloud is almost giving away your business to a different company and with a hefty top up money. Yes, the exit strategy of cloud systems is just bunch of data dumps. It will again require data modeling, formatting to bring it back to on premises and can cost from millions to 100s of millions of dollars.
Marco, this was an informative and well written article, thanks for sharing.
To me it reads that…. Most organization’s applications are built on redundant pools of virtual or physical servers that are both sized for peak load and that are run fully and continually. That said, moving these applications to the Public Cloud and operating them as is would naturally lead to increased infrastructure costs.
However, given the “pay as you go” nature of the Public Cloud, if organizations refactored their applications to use a platform that enabled them to automatically increase and decrease running application capacity in realtime response to increasing or decreasing user demand, it is highly likely organizations would be able to operate their applications in the Public Cloud at a cost lower than on-premises. An example of such a platform would be Docker containers and Kubernetes container orchestration.
Is that a fair summary? Thanks!
Yes, fair summary. With some caveats… in fact, scheduling instances and rightsizing don’t require any application refactoring. Enabling auto-scaling based on user demand, as you describe, would certainly help achieve further cost benefits. Using containers and Kubernetes could help implement auto-scaling policies, but it would not be a requirement. You can achieve auto.scaling using just cloud provider’s native services and capabilities and potentially no containers (although when using compute instances, you’d have to plan for much slower start time).
I enjoyed the article Marco. Would you mind sharing some detail on what was factored into the onpremise cost?
Hello – I would be happy to discus those details in an inquiry call!
Thanks for the insightful article. Did your study include any organizations that relied heavily on 3rd party, non-web based systems (e.g. ERP, CRM, etc.) that were still client installations? I ask because in many of the SMBs I have worked in/with were anxious to move to the cloud but found it difficult to migrate such 3rd party apps without needing added complexitiy of running RDP/Remote App/VDI or similar solutions. We often found the costs to be prohibitive due to such endevours and the need to increase internet bandwidth (expenses often overlooked when the internet was for external use such as email and not required for heavy operational use – especially if said 3rd party apps were not designed for internet or bandwidth optimization).
We also saw that many struggled with OpEx increases. While they didn’t like having CapEx for hardware, etc., most were wanting desktops and cheap servers to live 15-20 years and would only replace when they died (and typically without any DR or Backup solutions).
I’m just curious as to whether the same analysis spans across SMB domains as well as enterprise environments.
My opinion is that if when purchasing hardware for on-premises, you get all the quantities right, buying exactly what you need without excessive extra capacity and you can budget growth exactly in time when you need, then yes it would be definitely cheaper than cloud.
However, the reality is that on day one we end up purchasing capacity for 3 years even if we do not initially need it. The may reason being that we get better discounts from hardware vendors for initial purchase rather than on upgrades. And because we don’t want to be upgrading every few months.
And getting the quantities right?? Either we end up never using the capacity we planned for or the opposite we run out of resources after a year and struggle until we justify the requirement for more budget and get through the procurement and upgrade process.
With cloud it is easier to purchase exactly what you need and easy to adjust quantities up or down according to change in needs.
Hence although byte for byte cloud may cost more, if well managed can be cheaper because you only pay for what you need. Adding to that you can be more agile to increase capacity due to some sudden growth without having to wait for procurement, delivery, installation and configuration.
On paper it is all too easier, in reality it is a different story.
thanks for a great article. After studying public vs having a private data centre the public is costing more when there is a existing reasonably good setup.
The Public cloud can be cheaper for the following
1) for new startup business
2) For additional projects
3) For Business Continuity
4) For Promotion & Marketing Campaign
5) For Software Development & training.
Other than the above cost really shoots up leaving in you in a shock.