68% Move AI Workloads Off Public Cloud As Enterprises Rethink Infrastructure
In the News
68% Move AI Workloads Off Public Cloud As Enterprises Rethink Infrastructure
As artificial intelligence moves from pilot projects to production systems, a growing share of Indian enterprises are moving some workloads away from public cloud environments. A global survey by data platform company Cloudera found that 68 percent of Indian respondents had moved at least some AI workloads from public cloud to private cloud or on-premises infrastructure over the past year.
Separately, 79 percent of Indian respondents said AI had changed their data storage and architecture practices altogether. Globally, 66 percent of respondents said they had moved AI workloads back to private or on-premises infrastructure in the past year, pointing to a wider shift towards more selective infrastructure decisions as AI deployments scale.
Cloud-first To Workload-first
Technology leaders and analysts say the movement does not represent a wholesale rejection of public cloud. Instead, enterprises are now assessing individual AI workloads based on their specific requirements, with hybrid architectures combining public cloud, private cloud, and on-premises environments emerging as a more pragmatic model.
According to Biswajeet Mahapatra, Principal Analyst, Forrester, Indian enterprises are shifting away from a cloud-first or cloud-only approach as AI initiatives enter production. “Enterprises are evaluating infrastructure through the lens of business risk and workload characteristics rather than simply infrastructure cost,” he said.
Forrester's research in India shows hybrid cloud has become the dominant operating model, driven by the need to balance innovation, governance, cost, resilience and sovereignty. AI is accelerating this shift because training, fine-tuning, inference, retrieval-augmented generation and agentic AI workloads have materially different infrastructure requirements.
The same shift is being seen in enterprise architecture more broadly. Workloads that require strict control over data, predictable performance or sustained compute are assessed differently from experimental or highly elastic applications.
Cost, Compliance Drive Choices
Cost remains an important consideration, but enterprises are weighing it alongside security, compliance, data sovereignty, latency, resilience and operational control. The infrastructure decision has become more technical as organizations move from experimentation to scaled production and deal with different forms of AI workloads.
Regulatory obligations and auditability are particularly relevant in financial services, healthcare, government and telecommunications, according to Deloitte. Enterprises are also placing greater emphasis on architectural portability, ensuring they retain flexibility and optionality as technology, provider and business requirements evolve.
Amit Maheshwari, Head of Cloud & DevOps Practice, EPAM India, identified four imperatives that determine where production-grade AI workloads land: cost predictability, regulatory compliance, data gravity and vendor independence. “Instead of all-or-nothing cloud mandates, business and technology leaders are adopting a disciplined, workload-by-workload placement strategy,” he said.
Maheshwari also points to the economics of continuous inference as a factor in infrastructure decisions. “Running continuous, high-throughput inference 24/7 on cloud GPUs can cause operational budgets to spiral out of control,” he said.
Read the full article here.
Discover how EPAM delivers end-to-end AI development and cloud services to help enterprises automate, optimize and scale with production-ready solutions here.
