India’s Enterprises are Scaling AI. Is the Infrastructure Ready?

retained text

Indian enterprises are investing in artificial intelligence, but deploying AI agents across business operations presents a challenge that goes beyond the models themselves: whether existing digital infrastructure can support them.
AI agents need to retrieve information from enterprise databases, connect with business applications and carry out tasks across systems. Yet many organisations still rely on legacy applications, fragmented data environments and technology architectures that were not designed for software capable of taking actions across multiple systems.
Consider an agent tasked with processing a customer order. It may need to retrieve customer details from a customer relationship management (CRM) platform, verify inventory against an enterprise resource planning (ERP) system and initiate the next step in the workflow. If the applications cannot exchange information reliably or return conflicting records, the process can fail even when the model performs as intended.
The gap between AI ambitions and infrastructure readiness is already evident. Dell Technologies’ Modern Enterprise Readiness Study, conducted with Vanson Bourne, found that 67% of Indian organisations believe their current data centre environments are not ready to support AI workloads at production scale.
“The biggest hurdle in moving an AI agent from pilot to production is rarely the model itself. It's the reality of integrating with fragmented enterprise data, legacy applications, security controls, and business processes that have evolved over decades,” says Vishal Kappanna, Engineering Director at Tesco Technology.
For enterprises, the challenge is no longer simply building an agent that can complete a task. It is ensuring the surrounding systems can support that task reliably, repeatedly and at scale.
The Legacy System Challenge
Replacing entire technology stacks is neither practical nor necessary for most enterprises. Existing systems contain years of business logic and continue to support critical operations. Instead, companies are introducing application programming interfaces (APIs), middleware and orchestration platforms to connect legacy applications with AI systems.
However, connecting an agent to an application does not automatically make a workflow reliable. An agent operating across several systems must contend with inconsistent records, different access permissions and dependencies between applications. A failure midway through a transaction can require retries, recovery mechanisms or human intervention.
“Most enterprises are not replacing legacy systems to enable AI agents. They’re wrapping and modernising them through orchestration platforms, getting data out through events and APIs, and gating access behind governance layers and agent identities—modernising the seams, not the systems,” Kappanna notes.
This approach allows organisations to retain existing applications while exposing the capabilities agents need. But it has limits. Connecting an older system to an orchestration layer does not necessarily resolve the constraints of its underlying architecture. Some applications may still require deeper modernisation to support the demands of production workloads.
Vijay Yellapragada, Consulting AI and Data and US Commercial Leader at EY GDS, highlights the difference between demonstrating an agent’s capabilities and making it dependable in a live environment.
“A pilot demonstrates that an agent can perform a task. Production requires proving that the enterprise can trust the agent to perform that task repeatedly, securely, and at scale,” Yellapragada explains.
That requires enterprises to account for system failures, process exceptions and transaction volumes that may not emerge during controlled experiments. Integration, testing and recovery mechanisms become as important as the agent itself.
Fragmented Data Creates Another Bottleneck
Even when agents can connect to enterprise applications, their performance depends on the quality and accessibility of the information those systems contain.
Large organisations often store data across on-premises infrastructure, cloud platforms and distributed repositories. Records may be duplicated, outdated or inconsistent across applications, making it difficult for an agent to identify which information it should use.
“The biggest challenge is often not the AI model itself, but the data behind it. Most organizations have data spread across on-premises environments, cloud platforms, and edge locations, many of which were never designed to support AI-driven workflows at scale,” says Premalakshmi PR, Managing Director and Area VP, India and SAARC at NetApp.
NetApp's AI Data Engine is designed to help organisations discover, classify and activate data across distributed environments without moving or duplicating it. However, making data discoverable does not automatically resolve problems with its accuracy, consistency or suitability for a particular task.
These limitations matter when agents use information to initiate actions rather than simply generate responses. An agent working across finance, customer service or supply chain systems could act on an outdated record and carry that error into subsequent steps.
Enterprises therefore need to address data quality and access alongside integration. Without reliable information, connecting more applications to an agent may expand the number of systems it can interact with without making its decisions more dependable.
Infrastructure Must Scale With AI
The infrastructure requirements of an AI experiment differ from those of a production deployment. A pilot may involve a small group of employees and a limited dataset. A production system must accommodate more users, larger workloads and higher demands on reliability.
This places pressure on compute, storage and networking, as well as the connections between AI applications and existing enterprise systems. Organisations must ensure that their infrastructure can support these workloads without disrupting the applications already running their businesses.
“Agentic AI is changing the demands placed on enterprise infrastructure. Systems built to support people and conventional applications must now accommodate software that can interpret information, make decisions, and initiate actions across business workflows,” says Venkat Sitaram, Senior Director and Country Head, Infrastructure Solutions Group, India, at Dell Technologies.
Dell’s customer deployments illustrate how organisations are addressing different infrastructure requirements. Lakshmikumaran & Sridharan Attorneys is using the Dell AI Factory to run an in-house AI assistant on-premises, retaining control over its proprietary knowledge and client data. Zoho is leveraging the platform to develop privacy-first, multimodal enterprise and agentic AI capabilities. Omega Healthcare has modernised its infrastructure with Dell Private Cloud while continuing to support existing systems.
These examples reflect different priorities, from retaining control over sensitive information to developing AI capabilities and maintaining existing operations. They also underline that infrastructure readiness is not a single upgrade. It depends on the workloads an organisation needs to support, the systems it already operates and how effectively those components work together.
Modernisation Becomes The Deciding Factor
For Indian enterprises, moving AI agents into production will depend partly on how well they can build on existing digital investments. Organisations with established API platforms and modern integration architectures may have a head start. Others may need to improve data quality, connect older applications through secure interfaces or strengthen the infrastructure supporting their workloads.
The task is not necessarily to replace every legacy system. It is to identify where existing systems constrain AI deployment and modernise those parts of the technology stack. That could mean improving how applications exchange information, making enterprise data more accessible or introducing the infrastructure needed to support larger workloads.
These requirements are interconnected. Better computing capacity cannot compensate for unreliable integrations, while access to more data will not help if the information is inconsistent. An agent may work in a controlled pilot but struggle in production when it encounters the dependencies and exceptions of a live business environment.
As enterprises expand their use of AI agents, the deciding factor will not be the model alone. It will be whether their existing digital infrastructure can support agents as they move from isolated tasks to connected business workflows at scale.

document fetched 2026-10-30 02:57