ABSTRACT: As cloud adoption accelerates, not all analytics workloads are heading in the same direction. This blog explores three strategic options for data and IT leaders.

Read time: 4 mins.

Sponsored by Exasol

The rise of cloud computing seems inexorable as organizations embrace hyperscale platforms. But not everything flows to the cloud. Like eddies in a stream, some workloads are circling back on premises as organizations make nuanced decisions about flexibility, control and risk. This blog, the first in a series, explores three options for today’s data and IT leaders:

• Migrate analytics projects to the cloud to gain flexibility
• Move analytics back on premises or to private clouds (i.e., “repatriate”) to regain control
• Keep them on premises to avoid risk

We profile the ideal workloads and define the tradeoffs for each option while helping optimize price-performance. We focus on analytics projects, an especially fluid area as data and IT teams modernize to support AI innovation.

State of cloud adoption

The topic is critical because many teams must exercise all three options at once. They continue to migrate some workloads to the cloud, as evidenced by the rising revenue of AWS, Azure and Google Cloud. But they repatriate others: 86% of CIOs are doing so, up nearly 2x in five years, according to a recent survey by Barclays. And most maintain still other workloads on premises.

Let’s examine each option, including the motivations and ideal workloads for each, to understand what works where.

Migrate to the cloud 

More than 20 years after the launch of AWS, cloud migrations remain compelling for analytics projects that need flexibility. Organizations gain economic flexibility when they rent cloud storage and compute on demand rather than buying systems on premises. They shift capital expenditure (CAPEX) to operating expenses (OPEX), which can save money because they don’t pay for idle time.

The cloud also provides project flexibility: data and IT teams can start, stop and change projects on a dime, spinning resources up or down as the business requires. And they gain access to advanced tools: consider the analytics and AI/ML ecosystems now available through a hyperscaler such as Azure or a cloud lakehouse such as Databricks. Cloud platforms such as these make sense for analytics projects with evolving data sources, AI/ML models or compute requirements.

And yet… the tradeoffs have become stark. Technical debt, thanks to custom code or data gravity, raises the cost and complexity of migrating to the cloud. Data sovereignty requirements loom large given a volatile geopolitical environment, forcing some organizations to keep their data on premises. Some data and IT teams won’t accept the security risks of a migration or cloud hosting. Perhaps the greatest concern is price-performance on the cloud: spin up extra compute clusters to support quarter-end reporting or Cyber Monday ecommerce sales, and you might have a budget-busting cloud compute bill at the end of the month. You might even find you’re overpaying for the performance of predictable analytics workloads.

Repatriate back on premises or to the private cloud

Given such factors, a growing majority of organizations now repatriate select projects to regain control. The repatriation option provides more predictable price-performance than the public cloud because costs are fixed rather than variable. And by managing analytics in their own datacenters, data and AI teams can allay concerns about sovereignty and security. They also might use private cloud offerings such as Exasol’s data warehouse service to offload IT administrative tasks while maintaining flat usage costs.

Repatriating workloads can prove compelling when control matters more than flexibility. A European firm might move its data warehouse back on premises to address sovereignty regulations or keep personally identifiable information (PII) such as customer credit card numbers behind its own firewall. A US firm, meanwhile, might calculate that it can save money by hosting BI projects on its own servers or training new AI/ML models on a private cloud service. Even if they have ultra-high performance requirements, they can utilize infrastructure in a predictable, stable and cost-effective fashion.

Keep workloads on premises

Sometimes the best strategy is to maintain the status quo and avoid risk. Most Global 5000 organizations, born long before the cloud, still have heritage applications, servers and storage systems on premises. Their custom integrations and specialized processes, the result of long-accumulating technical debt, make migrations infeasible. Data and IT teams might have predictable and stable analytics projects, similar to those described in the last paragraph, that are best left as is on premises. They also might find it easier and cheaper to train new AI/ML models on premises so they don’t have to move all the necessary database tables, emails and documents to the cloud. Indeed, many AI/ML projects, while requiring advanced techniques and high volumes of data, have predictable workloads and performance requirements. Scenarios like these make the risk of a public cloud project unnecessary and unwise.

Chart your course

To chart the right course, data and IT leaders must assess flexibility, control, and risk for each workload.

• Migrating to the cloud offers scalability, flexibility and access to AI tools but requires cost vigilance and introduces security considerations.
• Repatriating restores control, makes costs predictable and keeps data sovereign, which helps compliance-sensitive sectors.
• Keeping workloads on-premises remains the lower-risk option for stable, high-performance and slow-changing analytics projects.

Leaders should conduct workload-specific evaluations, align decisions with business objectives, and create a hybrid strategy that optimizes price-performance while mitigating risk. The second blog in our series will explain where AI adopters host their workloads—and why—and the third blog will recommend best practices for navigating hybrid cloud environments that span all three scenarios. To learn more in the meantime, check out these blogs from our sponsor Exasol about repatriation and comparing cloud and on-premises options.