“Everybody wants to jump into AI, but many organisations are still not ready from a governance perspective.” – SK Venkataraghavan, Lenovo India

In an exclusive interaction with EFY’s Saba Aafreen, Lenovo’s S. K. Venkataraghavan discusses the challenges of moving enterprise AI from pilots to production, with a focus on governance, hybrid AI, infrastructure and edge computing.

Q. How has the enterprise AI conversation evolved from experimentation to production over the past year?

A. From what we are seeing, India is not behind global markets in enterprise AI adoption. The challenges and priorities are largely the same, with little difference between Indian CIOs and their global counterparts. Awareness of AI has increased significantly across enterprises, so the conversation has moved beyond whether to adopt it. The focus is now on where to start, how to invest, how to prepare the organisation, and how to deliver measurable business value and returns on investment.

Q. How is Lenovo approaching enterprise AI as organisations move towards production deployments?

A. Lenovo’s approach to hybrid AI is guided by its “Smarter AI for All” vision and aligned with the India AI Mission. Its strategy has three pillars: personal AI, enterprise AI, and the integration of the two. Personal AI is delivered through Lenovo and Motorola devices using the KIRA assistant. Enterprise AI, meanwhile, combines infrastructure, edge computing, storage and AI platforms. Together, these form an architecture designed to support enterprise AI deployments while addressing business value and return on investment.

Q. Which industries are moving beyond AI pilots into production first, and what is driving that shift?

A. India has become a key market in Lenovo’s AI strategy, with adoption growing across AI PCs, AI infrastructure and AI as a Service. The company has developed agentic AI solutions by combining its own intellectual property (IP) with a broad independent software vendor (ISV) ecosystem. Several proof-of-concept projects are progressing to production. Manufacturing, supply chain, and wealth and asset management are among the sectors leading enterprise AI adoption.

Q. What is driving enterprise demand for AI PCs, AI infrastructure, and agentic AI?

A. Lenovo is seeing strong demand for AI PCs, AI infrastructure and agentic AI, although the drivers differ. AI PCs are gaining adoption as users increasingly value on-device AI processing, higher productivity and greater data privacy. In enterprises, demand for AI infrastructure and agentic AI is particularly evident in manufacturing and supply chain operations, where organisations are using AI to improve efficiency, reduce defects, optimise inventory and address operational bottlenecks.

Q. Despite growing AI investments, why are so many organisations still unprepared to scale AI responsibly?

A. Everybody wants to jump into AI, but many organisations are still not ready from a governance perspective. Questions about AI governance, ownership and investment management remain unresolved. As a result, the focus is moving from individual pilots to organisation-wide platforms supported by change management, capable implementation teams, strong governance and suitable technology partners. The gap is less about willingness to invest and more about readiness to scale AI responsibly.

Q. What prevents AI pilots from becoming production deployments in most enterprises?

A. The short answer is that most failures are organisational rather than technological. They often stem from unclear objectives, choosing the wrong problem to solve, selecting unsuitable tools or a lack of organisational readiness. AI initiatives therefore need to be driven at the CXO level and backed by management. In the early days of enterprise AI adoption, many companies also launched projects out of fear of missing out. Combined with a shortage of skilled talent, these factors help explain why many pilots never reach production.

Q. Is building AI actually the easy part compared with operationalising it across an enterprise?

A. Yes. Integration, change management and organisational readiness are proving far more complex than building AI models. The challenge is not whether AI is developed in-house or sourced externally, but whether enterprises can integrate it into existing systems, align their processes and drive adoption across the organisation. This is a global challenge, not one limited to India. As a result, AI decisions are increasingly moving beyond IT teams to business leaders and boardrooms. Success depends on organisational buy-in, data readiness, talent availability and the ability to move from pilots to production at scale.

Q. What critical mistakes do enterprises make when taking AI into production?

A. Governance remains the biggest blind spot, followed by security and a lack of clarity about what the organisation wants to achieve with AI. Many initiatives still lack clearly defined business objectives. Data readiness is another critical factor. Organisations must decide whether to invest in improving data quality or risk poor outcomes by building AI systems on fragmented datasets. Talent and governance responsibilities must also extend beyond the IT function. Controls over user access, AI spending and human oversight are becoming increasingly important. Even with human-in-the-loop systems, approving AI-generated outputs without proper validation can compromise the reliability of enterprise AI over time.

Q. How important is data readiness in determining whether an AI initiative succeeds or fails?

A. Data has been one of the biggest challenges in the AI journey. Many organisations could not get started because they lacked a clear data policy or had information scattered across systems in formats unsuitable for AI applications. AI readiness therefore begins with a clear data policy, effective governance and a well-defined business problem. At the same time, newer AI solutions are reducing this barrier by working with existing enterprise data rather than requiring a complete data warehouse. This can help organisations accelerate deployment without waiting for their data infrastructure to be fully consolidated.

Q. Beyond technology, what organisational capabilities do enterprises need to build for AI at scale?

A. This is one of the biggest challenges facing enterprises. A widely cited estimate suggests that the industry will need around 2 million AI-trained professionals over the next five to seven years, highlighting the scale of the skills gap. In response, organisations are making AI training mandatory across job levels, often linking it to performance and continuity. They are also expanding structured upskilling initiatives internally and through industry partners. There is no quick fix, so enterprises are working with their existing talent pools and building capabilities gradually through internal programmes and industry collaboration.

Q. As AI matures, are enterprises moving towards repeatable AI architectures instead of custom deployments?

A. Enterprise AI is in transition. Many organisations are still building custom architectures for specific AI workloads, but deployments are increasingly progressing from proof-of-concept projects to production. As these deployments scale, larger enterprises are adopting reference architectures to improve scalability and simplify long-term operations. The result is a mix of custom-built and standardised hybrid AI environments.

Q. How does infrastructure strategy change once AI moves from proof of concept to production?

A. In the early stages of enterprise AI adoption, most initiatives remained at the proof-of-concept (POC) stage, with attention focused largely on the application or service layer. Enterprises could experiment relatively easily using hyperscalers or free trials. However, as organisations move from POCs to production, infrastructure requirements and total cost of ownership become critical considerations. Assumptions made during pilots often change as deployment scale, supply constraints and actual costs become clearer. This is making infrastructure providers more relevant to enterprise AI decisions, particularly when assessing hardware costs alongside the wider solution. Consequently, architecture discussions are increasingly focused on production requirements, cost, scalability and choosing the right mix of GPU and non-GPU systems for different AI workloads.

Q. What role does edge computing play in making enterprise AI practical at scale?

A. Enterprise computing has increasingly moved towards a unified compute fabric across infrastructure layers, and AI has accelerated this development rather than introduced it. Most modern server environments are already hybrid by design. In this context, edge computing is becoming an extension of the same compute continuum that connects edge, cloud and data centre environments. Its role is determined by workload requirements rather than being treated as a separate architecture.

Q. How should enterprises decide which AI workloads belong at the edge and which belong in the cloud?

A. The decision is not simply about balancing edge and cloud. In distributed environments such as renewable energy, edge processing is essential because turbine data must be captured and acted on locally. Edge systems handle real-time and latency-sensitive workloads, while cloud or data centre systems are selected according to cost, scale and workload criticality. The focus is therefore on distributing AI workloads efficiently and ensuring effective data flow between edge devices and central systems. The final arrangement depends on industry requirements, use cases and cost trade-offs rather than a fixed architecture.

Q. Why is edge computing becoming an important starting point for enterprise AI deployments in India?

A. We are seeing many enterprise AI deployments begin at the edge rather than at the core platform. In markets such as India, where much operational activity takes place outside centralised facilities, edge computing is becoming increasingly relevant. Renewable energy provides an example: wind turbines generate data in the field, and real-time insights are important because maintenance can be expensive. Consequently, many AI proof-of-concept projects are starting with edge and IoT-generated data. This is influencing how enterprises approach AI deployment across distributed environments while maintaining consistency.

Q. When does hardware acceleration become essential for enterprise AI deployments?

A. It depends on the use case, but hardware acceleration becomes critical for real-time AI at the edge. Where an industry or solution needs to run substantial workloads locally, investment in acceleration is necessary to support successful deployments. Although there may be arguments for both approaches in theory, many industries are increasingly opting for more powerful edge systems to handle performance-intensive workloads.

Q. As enterprises consolidate more AI workloads, what challenges will emerge in scaling these deployments?

A. The primary challenge today is not performance but deployment maturity. Most enterprises are still running relatively focused AI workloads, so scalability challenges remain limited. However, as organisations begin consolidating manufacturing, supply chain, HR and other functions on shared AI platforms, maintaining consistent performance will become more complex. Increasing volumes of data generated at the edge, security requirements and demand for on-premises deployments are also expected to shape the next phase of AI infrastructure design.

Q. How mature is adoption of ‘Everything as a Service’ models in Indian enterprises today?

A. Adoption is fairly mature now. It began gaining momentum during the COVID-19 pandemic, when enterprises realised that infrastructure could be consumed differently, particularly as hyperscalers and cloud providers expanded rapidly. Since then, adoption has remained steady, with infrastructure as a service becoming widely accepted among enterprises, including mid-market and large businesses. Many organisations now run a significant share of their workloads on cloud-based or service-based models, while highly intensive systems, such as SAP and certain manufacturing and financial workloads, continue to run on dedicated infrastructure. Cloud adoption is no longer a major point of debate; it is now an established part of enterprise IT planning. 

Loading form…

Industry's Buzz

Learn From Leaders

Startups