After years of marching to the cloud migration drumbeat, CIOs are increasingly becoming circumspect about the cloud-first mantra, catching on to the need to turn some workloads away from the public cloud to platforms where they will run more productively, more efficiently, and cheaper.

“‘Cloud exit’ became a big theme in 2023 and there’s good odds it’ll turn into a real trend for 2024. The savings are just too big to ignore for a ton of companies,” says David Heinemeier Hansson, the Danish developer of Ruby on Rails and co-owner and CTO of 37signals, which completed a six-month total exit from the cloud last June. “Enough people are realizing that the cloud marketing spiel doesn’t necessarily match their reality.”

And it’s that reality of accumulated experience in the cloud that has many CIOs rethinking their platform-centric approach to cloud in favor of a workload-specific one. The resulting infrastructure of choice — a combination of on-premises and hybrid-cloud platforms — will aim to reduce cost overruns, contain cloud chaos, and ensure adequate funding for generative AI projects.

David Linthicum, former chief cloud strategy officer at Deloitte, says many enterprise CIOs who got caught up in the race to the cloud are now fixing their “misadventures” by seeking out the ideal platforms for various applications — whether that is in a private cloud, on an industry cloud, within their own data centers, through a managed service provider, on the edge, or orchestrated in a multicloud architecture.

“The most common motivator for repatriation I’ve been seeing is cost,” writes Linthicum, who conjectures that “most enterprise workloads aren’t exactly modern” and thus not best fits for the cloud.

C R Srinivasan, EVP of cloud and cybersecurity services and chief digital officer at Tata Communications, sees many enterprises “getting more nuanced” with their cloud use and strategies in an effort to balance performance, costs, and security.

“As businesses look to leverage artificial intelligence a lot more, they are and will relook at the workloads and place them on the right infrastructure, be it in the public cloud or the edge or bringing them back to their own private cloud or servers in-house,” Srinivasan says. “Such decisions are largely driven by the need to maximize performance and business benefits while not losing track of costs.”

John Musser, senior director of engineering for Ford Pro at Ford Motor Co., agrees.

“It’s a form of rightsizing, trying to balance around cost effectiveness, capability, regulation, and privacy,” says Musser, whose team found it more cost effective to run some workloads on a high-performance computing (HPC) cluster in the company’s data center than on the cloud. “Even though we’ll often do it in the cloud, doesn’t mean we should always do it in the cloud.”

The public cloud pivot

That mindset is catching on as CIOs look to apply lessons learned from their initial push to the cloud.

“Any organization of size, dealing with diverse technology, is doing their company a disservice if a public cloud–only strategy is their ultimate end goal,” says Brian Shields, SVP and CTO of Boston Red Sox and Fenway Sports Management.

“Like many complex businesses, we are an evolving hybrid model that maintains compute and storage capabilities in the public cloud, on-prem, with our co-location partner, and industry cloud partners,” Shields adds.

This refinement of thinking about the cloud comes as hefty AI costs loom on the horizon. For CIOs who need real-time access to data, such as for manufacturing or industrial controls, loading data on the edge is a better solution than the public cloud.

“Edge provides processing of real-time computation, for instance computer vision and real-time computation of algorithms for decision-making,” says Gavin Laybourne, CIO at Maersk. “I send data back to the cloud where I can afford a 5-10 millisecond delay of processing. “

At the CDO Summit in December in Boston, Mojgan Lefebvre, chief technology and operations officer at Travelers, noted that the cloud has the scalable and adaptable infrastructure for various needs, as well as access to more advanced AI tools such as large language models.

But “importantly, this reliance on cloud technology does not necessitate a complete migration of all assets to a cloud-based environment,” Lefevre said.

Payroll giant ADP, for instance, uses AWS for most of its net-new applications, as well as Microsoft Azure and Cisco Cloud, but “we still have a lot of load running in our data centers,” says Vipul Nagrath, head of product development at ADP and the company’s former CIO.

Some CIOs are opting to host workloads in private clouds, such as HP Enterprise’s Greenlake or Dell’s APEX platforms, to achieve greater security, and lower costs, than they would in the public cloud.

Richard Semple, CIO of Williamson County, Texas, where Samsung’s sprawling new chip-making facility is under development, considered all the public clouds for the government’s growing digital infrastructure. In the end, he opted for the security of retaining data on premises but on a private cloud engineered by Dell.

Reassessing, one workload at a time

For those CIOs already deep into the cloud, taking a hard look at all aspects of an application before adding yet another to their cloud estate is becoming more the norm than simply pushing forward.

“We don’t go into the cloud unless we know there are savings and we keep measuring to ensure,” says Jamie Holcombe, CIO of the US Patent & Trademark Office. “I know from experience that ‘chatty’ applications are often the most expensive in the cloud, so we either re-factor or keep on-premise.”

Not all government CIOs are moving workloads off the cloud or feeling the need for repatriation. “I am 100% in the cloud and would not have it any other way,” says Gerard Caron, CIO of the International Trade Administration.

And while repatriation is a real trend, it’s not yet universal.

“It just shows CIOs are actually thinking about where they want to platform their application portfolios,” says Steve Randich, CIO of the Financial Regulatory Authority (FINRA), a private company. “The cloud makes sense in some but not all cases.”

As for FINRA, the cloud remains central.

“In FINRAs case it would cost double to build the infrastructure internally that we use every day on AWS,” Randich says. “Plus, we would lose the flexibility to quickly ramp up and ramp down infrastructure based on expansion and contraction of transaction volume. It may very well be that many organizations have highly predictable, stable volume. Not FINRA.”

Whether any given workload is best suited to the cloud is a matter of context. Wiser and more experienced, CIOs today are being more intentional about determining this to ensure, on a case-by-case basis, applications are hosted in the context that matters most.