There is a growing urgency to adopt cloud services to buttress today’s rapidly growing inventories of high-end analytic and AI applications. This is fueled by massive volumes of data that are spilling over the capacities and capabilities of on-prem systems. It works the other way, as well: AI is paving the way for more robust cloud services, offering ever-expanding real-time automated provisioning and scalability.
These days, just about every application—such as Salesforce analytics—is native to the cloud, of course, so an on-prem deployment may no longer even be a realistic consideration. For example, “When data warehouse technology is exclusively SaaS-based, there is no open source or on-premise equivalent,” said Joseph Morais, technical champion for Confluent. “Such architecture would be extremely difficult, if not impossible, to replicate in the vast majority of data centers.”
Ultimately, then, it’s about AI serving to run better AI. AI-boosted cloud services can be employed to run AI and analytics faster and reach more users. At the same time, there are situations in which on-prem makes more sense, at least at this time.
This is where IT and data managers need to make a decision: Is it worth the possible long-term costs to move or access AI and analytical applications in the cloud? How are cloud-based applications and data to be governed? What are the security implications? Does the potential higher latency of cloud-based interactions put any real-time business processes at risk?
Adding to the migration decision process is the fact that the world of data has become exponentially more complicated. “There has been a boom of tools, engines, databases, and systems, where before we only had Hadoop and Hive,” said Sergio Gago, CTO of Cloudera. “Now we have hundreds of database types for big data. Companies need to have a holistic view of their data and how it changes and adapts dynamically. For example, after a merger or an acquisition, workflows and datasets will continuously move from one cloud provider to another, or to different data centers.”
CLOUD VERSUS ON-PREMISES
Many, if not most, data and analytics functions are now supported on the cloud. Almost thirty-eight percent (37.8%) of enterprises seek improved analytics query performance by moving to the cloud, a survey of 259 data executives published by Unisphere Research has found. Close to 40% (39.8%) seek improving self-service data analytics for users through cloud services.
Nearly 34% (33.6%) are turning to cloud to enable broader analytics, business intelligence (BI), and AI/machine learning (ML) use cases (“2025 Market Study: Modern Data Architecture in the AI Era,” July 2025).
At this point, “I’d be hard-pressed, categorically, to find capabilities that exist solely on-premises that you can’t already find in the cloud today,” noted Luis Flynn, senior product marketing manager at SAS. “But just because you can, doesn’t mean that you should.”
Most analytics workloads—“including relational, graph, vector, and AI-driven analytics—are now cloud-ready,” agreed Kevin Kline, senior staff technical marketing manager at SolarWinds. “Cloud platforms provide the scalability, flexibility, and GPU resources needed for compute-intensive tasks such as machine learning or LLM training.”
However, speed in terms of latency still makes the case for on-prem systems. “Although many companies have moved their data to the cloud, some situations still require on-premise storage and analysis,” said Morais. “An example of this could be low-latency scenarios, such as real-time failure detection, where even a few milliseconds of delay could cause a significant and negative impact on the business.”
Manufacturing is another environment in which low-latency on-prem systems are still needed and where “real-time failure detection is critical,” Morais added. “Sensors on machines must quickly spot problems to prevent major breakdowns. If this data were sent to the cloud for processing, the time it takes for the data to be sent and then processed could take too long, potentially leading to significant financial loss or safety hazards.”
Another case for remaining on-prem is the handling of “highly sensitive or regulated data that requires strict governance,” Kline pointed out. “For example, many critical government systems remain air-gapped from the internet. And the cloud isn’t ideal for some real-time streaming workloads, where ultra-low latency is critical.”
The scale and nature of the organizations also shape cloud versus on-prem decisions for analytics and AI applications and data. “Most data and analytics functions can now run in the cloud, but large organizations continue to rely on hybrid or multi-cloud setups,” said Scott Gnau, VP of data platforms at InterSystems. “Some data and workloads are staying on-premises because of intellectual property policies or to ensure compliance in tightly regulated industries.”
Ultimately, a hybrid approach is called for. “Hybrid cloud offers the power of choice and flexibility,” said Prakash Pattni, managing director for digital transformation at IBM Cloud. “Organizations must become hybrid by design—meaning enterprises need to place workloads and data where they will perform best.”
A hybrid approach helps maintain a balance between using “cloud for scale and agility, while retaining on-premises infrastructure for compliance or performance-sensitive workloads,” Kline pointed out.
It also pays to be picky about where data gets moved. “If re-platforming an existing application to the cloud, avoid a traditional lift-and-shift migration where all data is migrated,” advised Steve Leeper, VP of product marketing at Datadobi. “Instead, review the existing datasets and migrate the most meaningful data to the cloud while archiving old, outdated, orphaned, and irrelevant data.”
“Data managers need to understand the uniqueness of their business, IT processes, and applications” in order to make such decisions, said Pattni. The goal is “to strategically develop a workload placement strategy, deciding which data should reside in which environment. These decisions can be made based on factors such as cost, performance, lifecycle, compliance, and business requirements.”
He added, “Hybrid by design doesn’t mean scrapping a company’s existing IT infrastructure and starting over—it means making intentional choices about what outcomes data managers want and the technology to help you get there.”