New Delhi, Aug. 13 -- Data limitations, regulatory complexity and cloud economics are pushing Indian enterprises to redesign the infrastructure supporting their AI ambitions. Nearly seven in ten Indian organisations believe their data architecture needs a significant overhaul to support future artificial intelligence requirements, according to a new Cloudera survey. Governance hurdles and shifting workload economics are also prompting enterprises to reassess their reliance on public cloud infrastructure.
The findings point to a widening gap between companies' AI ambitions and the technology foundations needed to realise them. Infrastructure designed for conventional analytics must now support data-intensive models, automated decision-making and stricter requirements around security, privacy and auditability.
The Cloudera-commissioned survey found that 79% of Indian respondents said AI integration had significantly or moderately changed their organisations' data storage and architecture practices. This included 22% who reported significant changes.
Meanwhile, 68% said their existing data architecture needed a substantial overhaul to meet future AI requirements. The India findings are based on responses from 100 enterprise architects, cloud infrastructure leads and data architects. They form part of a global survey of 1,500 technology leaders conducted by Wakefield Research in June.
The results suggest that access to AI models is no longer the only determinant of enterprise readiness. Companies also require reliable data pipelines, scalable storage and governance systems that can manage information consistently across different technology environments.
"Indian enterprises are shifting their focus from mere AI experimentation to effectively converting AI-led innovation into tangible business impact," said Mayank Baid, regional vice-president for India and South Asia at Cloudera.
"As organisations deploy more AI applications, they are realising that success depends on having the right data foundation, strong governance, and the flexibility to run workloads across hybrid environments," he added.
"The biggest constraint isn't AI-it's enterprise readiness," Rishi Aurora, managing partner at IBM Consulting India and South Asia, said in an earlier interview with TechCircle. Aurora identified legacy processes, technical debt, fragmented data and skills shortages as the principal obstacles to scaling AI. "Scaling GenAI and agentic AI requires simultaneous change-modernising systems, strengthening data foundations, redesigning workflows, and preparing the workforce," he said.
Governance becomes an architectural concern
Governance is emerging as one of the biggest constraints on AI deployment. About 71% of Indian respondents in the Cloudera survey said AI integration had made data governance more complex and difficult to maintain. As many as 91% said their organisations had delayed or cancelled at least one AI project during the preceding 12 months because of data governance, compliance or regulatory concerns.
The findings reinforce how governance is changing from a largely policy-led exercise into an infrastructure requirement. Enterprises need to track data lineage, consent, access rights and model outputs as information travels across clouds, data centres and business applications.
The challenge becomes more pronounced with agentic AI, where systems can independently retrieve information, make decisions and initiate actions. "As agent adoption scales, ungoverned data doesn't just slow things down-it multiplies risk," Gaurav Pathak, senior vice-president for metadata and AI products at Informatica, wrote in a TechCircle opinion article.
Traditional governance tools frequently depend on employees reviewing dashboards and approving decisions. Autonomous systems, however, require data-quality checks and governance controls to operate at the point of access, before an action is taken. This is particularly important for banks, insurers, healthcare companies and government agencies, where AI-led decisions must meet requirements around privacy, auditability, explainability and human oversight.
Workloads find new homes
Indian enterprises are also reconsidering where their AI workloads should run. About 68% of respondents said their organisations had moved at least some AI workloads from public cloud to private cloud or on-premises infrastructure over the past year. A quarter reported making significant shifts.
The movement does not necessarily signal a retreat from public cloud. It reflects a more selective hybrid strategy, under which companies place workloads according to cost, latency, performance, security and data-residency requirements.
Faiz Shakir, managing director of sales for India and SAARC at Nutanix, had earlier told TechCircle that enterprises increasingly recognised that "one size does not fit all", as different applications require different deployment models.
AI is also altering storage economics. Inference logs, telemetry, synthetic datasets and model checkpoints create expanding volumes of information. Some of this data may have little immediate utility but could become valuable for model training, validation or retrieval later.
"AI does not just consume information; it amplifies it," Owais Mohammed of Western Digital wrote in a TechCircle opinion article. Retaining such data indefinitely, however, can drive up infrastructure costs, forcing enterprises to balance its potential value against storage economics.
For CIOs, the emerging reset is therefore about more than adding computing capacity or moving workloads out of public clouds. It requires deciding which data should be retained, where sensitive workloads should run and how governance can remain consistent across environments.
The companies that resolve these questions will be better positioned to move AI beyond isolated deployments. Those that do not may find that their biggest obstacle is not the capability of the model, but the architecture beneath it.
Published by HT Digital Content Services with permission from TechCircle.