The race to the cloud, accelerated by the pandemic, left many organizations running workloads in hyperscaler environments — often with consumption-based pricing that’s difficult to predict. That model is beginning to change, with production-scale AI adoption reshaping how and where workloads run. As organizations move beyond AI pilot projects, they’re increasingly prioritizing hybrid and on-premises infrastructure. According to a recent benchmark study from Cloudian, 93 percent of enterprises are now engaged in some form of AI workload repatriation, whether actively moving workloads or evaluating a shift away from public cloud. Of course, even as some workloads move away from public cloud, overall demand for hyperscaler services continues to accelerate. According to McKinsey, AI remains the primary growth engine for data centers in the United States, with hyperscalers expected to capture 70 percent of the capacity in the U.S. market between now and 2030. As Cloudian noted, “This does not signal an abandonment of cloud – rather, it reflects a maturing understanding that different AI workloads have different infrastructure requirements, and that on-premises deployment is increasingly the right choice for many of them.” Key Drivers of AI Repatriation Data Sovereignty: Cloudian’s survey revealed a growing tendency to migrate AI workloads involving sensitive data, mission-critical operations or regulatory compliance. This trend coincides with rampant “shadow AI,” or the unauthorized use of cloud-based AI tools. For example, employees may run sensitive data through tools such as ChatGPT without formal oversight or governance. In the study, 74 percent of respondents viewed shadow AI as a critical or significant data security concern. Almost one in four reported incidents where employees ran confidential data into cloud AI tools. And almost half of respondents have implemented controls to restrict sensitive data from flowing into AI cloud services. When asked how they would deploy new AI applications, 91 percent of respondents said they would choose on-premises infrastructure, a private cloud or hosted environment or a hybrid approach while processing sensitive data on-premises. Meanwhile, only 8 percent would select a public cloud with enhanced security, while just 1 percent would use public cloud with standard configuration. Cost predictability: While many enterprises initially migrated to the cloud to reduce costs, consumptionbased models have made spending more difficult to predict. Forty percent of enterprises reported that actual cloud By Garald Baldino AI workload repatriation gains steam as organizations seek improved security, cost, predictability and performance Balancing Back AI & AUTOMATION AI Workload Repatriation from Public Cloud (Past 24 Months) Repatriated significant workloads from cloud 26% Moved some workload or in process 53% Actively evaluating repatriation 13% No plan to move from the cloud 5% Not applicable 2% Source: Cloudian’s 2026 Enterprise AI Infrastructure Survey 14 CHANNELVISION | SPRING 2026
RkJQdWJsaXNoZXIy NTg4Njc=