feature Cloud-based infrastructure services date back to at least 2006, when AWS introduced its S3 storage platform, followed by Elastic Compute Cloud (EC2) instances. Since then, the cloud has grown into a $100 billion global industry, but some customers have begun to question the move to these services and begin bringing workloads back in-house.
The cloud market has been so successful, in fact, that it has proven largely resilient to global economic factors, with some tech companies reporting record losses this year. The growth rate of cloud spending has slowed — from 20 percent in Q4 2022, to 19 percent in Q1 this year, to 18 percent for Q2, according to Synergy Research — but it continues to grow.
Many enterprises were reluctant to buy into the cloud in the early days, citing concerns about security and losing control over the infrastructure they depended on to operate key applications and services.
Those days are gone, as can be seen by more statistics from Synergy, which show that a decade ago enterprises were spending more than $80 billion annually on their own datacenters, with less than $10 billion on cloud infrastructure.
Since then, spending on datacenters has grown by an average of 2 percent annually, while spending on the cloud has grown by an average of 42 percent annually, reaching $120.3 billion in 2022.
So cloud-based infrastructure has become an accepted part of the way organizations run their IT these days, and that’s unlikely to change, but perhaps some of the shine has taken off the allure of the cloud in recent years.
Some companies are finding that, far from saving them money, operating applications and services in the cloud can be just as expensive as owning and managing your own infrastructure for the purpose, and sometimes more.
Earlier this year, it was demonstrated by a company that calculated that keeping its infrastructure on-premises instead of using Amazon Web Services would save $400 million over three years, because Register Reported.
Another example is Basecamp project management developer 37Signals, which decided to abandon the cloud and return to on-premise infrastructure after being presented with a $3.2 million cloud hosting bill.
Part of the problem seems to be that the cloud is not easier to manage than maintaining your own infrastructure, but instead presents a different set of management challenges.
For example, there was what was called server sprawl, where users spin up new virtual servers for each new project, then forget to turn them off when they are no longer needed. Even if they are unused, those instances will still be metered by the cloud provider, incurring unnecessary costs.
A Forrester Consulting report commissioned by cloud software outfit HashiCorp claims that 94 percent of decision makers and practitioners surveyed said their organization has experienced one or more types of avoidable cloud costs.
Half of the respondents were affected by over-provision of resources and idle or underutilized resources, blamed on lack of skills or appropriate capacity to manage resources.
The issue was also highlighted by CAST AI late last year, which has some vested interest as it develops a platform to monitor customer resource usage across three major cloud platforms – AWS, Microsoft (in this case Kubernetes clusters). Azure and Google Cloud.
It claims that organizations allocate a third more cloud resources than they use on average, with organizations blaming a lack of visibility into their cloud usage as the main reason.
The CAST AI platform provides free analytics for organizations to determine how their cloud resources are provisioned, while paying customers have the option to let the platform take remedial action based on its findings.
Those actions can include freeing up unused resources, or moving workloads to spot instances — virtual machines that take advantage of unused spare capacity — which can save up to 60 percent, the company claimed at the time.
According to Everest Group, cost pressures are forcing organizations to take a closer look at their cloud investments. Cost has now become a top concern among cloud customers, and 67 percent of enterprises surveyed say they don’t see the value they expect from the cloud.
Abhishek Singh, Cloud and Legacy Transformation Leader at Everest, told us that the cost benefits of operating cloud-native applications and the way organizations are using the cloud are clear.
“For legacy or complex enterprise workloads, enterprises are realizing that picking up and moving applications from on-premises to the public cloud is a bandwagon, not a permanent solution,” he said. “Containerizing applications and porting them to the cloud can help you solve the infrastructure question, but the question of modern application architecture still hangs in the balance,” he added.
The whole cost question is “probably the worst-kept secret in enterprise technology,” Singh claimed. “Cloud isn’t cheap and doesn’t free you from redundancy, two cases for public cloud we were convinced 10 years ago.”
And a bigger problem is that cloud cloud environments mean that multiple lines of spending can change in oversight and management, making the multibillion-dollar commitment around cloud use less of a return on investment, Singh said.
“Hyperscalers made it seem like it was all self-service, when it really wasn’t – as seen in the booming businesses built by systems integrators (Accenture, Deloitte, PwC, TCS, Infosys),” he commented. .
It’s reminiscent of software licensing, where an entire ecosystem of software asset management providers sprung up to try to make money by obfuscating the byzantine process of ensuring software compliance, while dangling promised savings to customers.
Meanwhile, AI is now changing the cloud infrastructure landscape, especially with generative AI and Large Language Models (LLM) recently gaining the attention of many enterprise decision makers.
A recent report by Omdia found that investing in infrastructure for AI model training is now a top priority among datacenter operators. This means buying high performance servers equipped with expensive GPUs like Nvidia, which is displacing funds that would otherwise be used to refresh existing server fleets and invest in other new projects.
“One of the big questions around GenAI is the cost of running it,” Singh said. “Right now, it’s Nvidia and the cloud hyperscalers that are just making money.”
“But once the cost equation is optimized, the second question for enterprises will be: Do I keep my privileged data in the cloud or bring it back and run the model in a walled-garden environment? That’s a future conflict that cloud players will have to deal with,” he said. ®