Cloud repatriation, like everything else in tech, is surely exaggerated, as this article says. Few companies I’ve chatted with move any significant number of application elements from the cloud back to the data center. What isn’t exaggerated is what even a small number of repatriations means in terms of how enterprises assess their use of the cloud. I’ve gotten some new comments on this for 2024, and they show a different side of things.
Overall, 411 enterprises offered comments on cloud application policies and practices, not including SaaS usage, so far in 2024. This group can be divided into segments based on the extent to which the companies believed they’d made good cloud decisions. Only 19 of the group said their decisions had been “consistently correct”. Another 182 said they had been “largely correct”. A group of 144 said they had made “some major errors”, and the remaining 66 said their policies had been incorrect overall. In that last group, 12 said that they’d backed at least some applications out of the cloud—repatriated them. None of the other groups had done that, and no enterprise left the cloud entirely.
In all, though, 348 (84%) of the companies said they’d made changes to some cloud applications to improve cost efficiency. Even in the “consistently correct” group, 15 (79%) had done so. In the “largely” group, 148 (81%) had made changes, in the “some major errors” group 140 (97%) made changes. In the final group, 45 (68%) had made changes, and that was the smallest percentage of all the groups, a point I’ll address below.
In terms of workloads, it’s harder to get good data without actual surveys, and I try not to survey but to rely on spontaneously offered comments. However, it looks like he group overall reduced their 2023 cloud workloads by just under 20%, but they added work that meant, in the net, they showed almost no change so far for 2024. Some of the gains in 2024 are in the area of generative AI usage, though, and enterprises in groups overall think their cloud workloads will decline slightly this year.
Every enterprise who made changes said their failure to consider the factors that increased cloud costs was a primary reason they had to make the changes. In the two groups most happy with their decisions, enterprises said they’d gained more knowledge and experience and had “tuned” their cloud usage accordingly. That same view was held by about half the “some major errors” group, but the other half said they had to make major changes, and all the companies who said their cloud approach was simply flawed said that. I want to focus on those who had to make major changes (113 overall) in the rest of this discussion.
The biggest “major” problem that this group of 113 reported, and one cited by 103 of them, was the myth of “moving to the cloud”. These companies expected to utilize the cloud by simply transporting applications there, without real changes, and all of them said that proved to be a bad approach. According to them, the key to cloud success was to make a cloud decision as a part of application modernization (appmod). Of the remaining 235 who had been tuning their cloud, 220 said they’d either cloudified via appmod or had built a new application to utilize the cloud.
The issue that the 113 major-changes group cited was that simple movement of applications to the cloud increased cost, on the average, by 221%, meaning the cloud was over twice as costly. A few reported increases as high as 500%, and the smallest cloud cost premium reported was 121%. Where a transported app was redone via appmod, the reported cost premium dropped from 248% to 112%.
So the cloud is more costly in most cases, even after fixes? Yes, but that can mislead you easily. The reason is that these comparisons relate to the prior cost of on-prem hosting, and the majority of appmod projects were aimed at improving QoE. Had the cloud not been used to provide scalability and resiliency, enterprises say the cost to do that in the data center would have been higher. That’s because you have to size resources for peak loads and availability, which means utilization overall is lower and costs higher. This same point was made by the users more satisfied with their cloud planning; the cloud saves enterprises money by handling peak loads and faults more efficiently, which is exactly what you’d expect from a pooled resource model.
Why didn’t all enterprises expect it, then? They all say that cloud hype was a factor; few stories about the cloud talked about anything other than the notion of universal cloud adoption. But hype has been a fixture in technology for at least 25 years. The more direct problem was that enterprises making cloud decisions normally had no staff experience with the cloud; they tried to hire only when they were actively doing cloud things. Early hiring was hampered by the simple fact that management had little ability to assess cloud skills among applicants. Even successful cloud users said that it was difficult to staff for the early cloud projects, but got easier as they gained an understanding of what skills they really needed.
What seems to separate enterprise “tuning” of applications from the major change group is “appmod awareness”. That can arise from an understanding that “moving to” the cloud has to be more like “rewriting for the cloud”, but more often it arises simply from the fact that the mission of an application or applications is changing, most often related to either remote work or direct customer access to application data. Companies who plan business-driven application changes will normally end up considering cloud hosting, but within the context of specific business needs. One CIO said “We don’t do cloud projects, we do business projects that include the cloud”, and that seems the best approach.
Cloud providers may be coming to realize that simply hyping the cloud at the risk of an unfavorable cost outcome isn’t necessarily wise. Of the 348 companies who made changes to improve cloud economics, 49 said the changes were suggested by their provider, though none said that the suggestion was spontaneous rather than in response to a user complaint on costs. There seems to be an increase in this cost-helpfulness, but I’d hope that eventually cloud providers would step up and tell users they were incurring unnecessary costs. If that doesn’t happen, the cloud may be tarnished unfairly. Used correctly, it remains a good strategy for most enterprises.
