AI is changing the role of enterprise infrastructure. For Rampal Singh, SVP & Global Business Head, Hybrid Cloud Services and AI Factory, HCLTech, the shift is deeper than a new generation of hardware.

Infrastructure has traditionally been responsible for performance, uptime and capacity. It mattered to the business, but it often operated as a back-office function. AI changes that relationship because infrastructure performance and economics can directly influence how quickly the business can scale new capabilities and differentiate in the market.

“This is the first time in the history of IT infrastructure where IT infra has become a strategic priority for the business,” said Singh. 

That creates a new challenge for infrastructure leaders. They need to understand the language of the business while also managing a technology stack that is becoming more complex.

Why pilots break in production

The difference between a successful AI pilot and a production deployment often comes down to scale.

Singh noted that pilots are typically small, which keeps the associated complexity manageable. In production, the number of use cases, users, models and dependencies can grow rapidly.

That means infrastructure decisions made for a pilot may become constraints later.

“You need to have long-term thinking. You need to think ahead at least five years,” said Singh. 

That means designing the architecture, selecting products, integrating the stack and building the team with future demand in mind. Scalability, performance, availability and speed to market all become harder to manage if production requirements were not considered from the beginning.

The weakest link defines the system

AI infrastructure also needs to be treated as one system rather than a collection of individually optimized components.

“If there is one component which is not functioning as per the desired outcome, the performance of the whole stack [is at risk],” said Singh. 

A powerful compute layer cannot compensate for poorly designed networking. Expensive GPUs can remain underutilized if storage cannot deliver data quickly enough. Even when the physical infrastructure is well designed, an inefficient model can prevent the environment from being used effectively.

The goal is cohesion across compute, networking, storage, data and models.

That systems view becomes especially important because of the cost of AI infrastructure. Underutilization is not simply a technical problem. It directly affects the business case.

The foundation needs to become autonomous

Agentic AI introduces another change: workloads are becoming more persistent, dynamic and autonomous.

Singh argued that the infrastructure underneath them needs to evolve in the same direction.

“If the foundation itself is not autonomous, then you can’t expect the workload running on top of it to be autonomous,” he said. 

Traditional operating processes will struggle to meet the scalability, performance and response-time requirements of modern AI environments. Singh pointed to the need for a new level of AIOps that can anticipate failures, support remediation and approach capacity planning differently.

Rather than simply applying existing infrastructure processes to AI, enterprises need to reimagine how the environment is operated and consumed.

Singh expects humans to remain involved, but increasingly by exception rather than as the default mechanism for managing every infrastructure event.

AI Factory as an operating layer

This is where Singh places the HCLTech AI Factory.

“AI Factory is not really a product. It’s not even an infrastructure.  Its the operating model to deliver shared capabilities to thrive in AI era,” he said. 

The distinction is important because buying capacity alone does not create the capabilities needed to use it effectively.

Singh described a common problem: organizations invest in infrastructure and later discover that they have built capacity without the supporting capabilities needed to consume it. Data pipelines may be incomplete, GPUs may remain underused and governance may be added too late.

HCLTech AI Factory is intended to bring those capabilities together through reusable components, standardized architecture, governance, economic control and FinOps.

“AI Factory is a next-generation operating model, so that people, process, tools and technology come together and deliver the capabilities to the business to thrive in the modern era,” added Singh.