A recent article claims that Ahrefs, an SEO software suite company, was able to prevent $400 million in expenditures over three years by not leveraging cloud resources. Similarly, 37Signals, the makers of Basecamp, has begun a cloud exodus with the stated goal of saving seven million dollars in infrastructure costs over five years.
Efim Mirochnik, global DC lead for Ahrefs, explains that Ahrefs currently hosts on-premises within a data center located in Singapore. Mirochnik states that "Ahrefs has spent $122 million to support its on-premise infrastructure since 2017". Mirochnik then projects what the equivalent EC2-based infrastructure within AWS would cost. Two projections were made, one using on-demand resources, and one using three-year reserved instances paid all-upfront.
This projection indicates a potential spend (accumulated since 2017) within AWS of over one billion dollars as compared to the on-premises infrastructure being approximately ten times cheaper.
Reactions to the article were mixed with many claiming the author neglected to fully comprehend the total cost of ownership (TCO) of running on-premises. User serverhorror on Reddit expressed that:
For most people: they'll never have the, even the basic features, AWS provides out of the box (and if you don't need them it's absolutely fine to not pay for them by going to your own DC or just rent a rack).
User weehoey raised a warning to ensure that only the applicable benefits of investing in the cloud are weighed:
You are right that AWS provides a massive number of features. Features are not business outcomes. [..] Only the features that provide benefit given your use case matter.
However, many instead called out that the cloud is not always cheaper and each company must review their own use cases carefully. User jippen on Reddit shared this sentiment:
This article is super reductive, looks at two data points and extrapolates the world from it. Just because these were good decisions for your business doesn't mean they're good decisions for every business at every level of scale.
David Heinemeier Hansson, CTO at 37signals, shared a similar story to Ahrefs. Hansson predicts that 37signals stands to save approximately seven million dollars over five years by exiting the cloud and moving to on-premises hosting.
Adam Jacobs, CEO at System Initiative, shared on X that Hansson's assessment:
Makes complete sense to me that these numbers pencil out like this. In many ways, we’re paying for having forgotten how to rack compute, manage operating systems, and run networks.
Forrest Brazeal expands upon the idea that each company must assess the appropriate approach for their use cases and expertise. Brazeal explains that "not every company runs at Google scale, not every company has the competence (high) or growth aspirations (low) of 37signals". To elaborate on this point, Brazeal sketched a matrix to assess if the cloud or self-hosting would be a better choice.
User Indifferentchildren expressed a similar sentiment on Reddit: "Cloud is great for scaling, and cold-standby DR, but if you have decent-sized continuous loads, cloud can be a really expensive option."
The 2024 State of FinOps survey found that waste and cost reduction is now the highest priority with engineering enablement moving down the rankings. A recent CNCF microsurvey on FinOps found that Kubernetes usage has driven cloud spending up for 49% of respondents with the primary driver being overprovisioning.
The theme of Werner Vogels's 2023 AWS re:Invent keynote was frugality. Vogels outlined seven laws for frugal architectures and posited that cost should be viewed as a non-functional requirement when building.
With many businesses looking for ways to reduce costs and improve profits, there will be a continued focus on efficiently built infrastructures. As Kelsey Hightower highlights:
It might just turn out that the cloud was the best way to research and design better ways of managing our systems, and thanks to the open source community standardizing the APIs on top, we might finally have the blueprints we need to close the gap on-prem.