Lenovo ISG India managing director Srinivas Rao told TNIE that 99% of Indian organisations plan to increase AI investment, with budgets expected to rise by 19% year-on-year, as businesses expand deployments and shift from pilot projects to production.
The findings are from Lenovo’s CIO Playbook 2026, which points to growing investment across generative AI, agentic AI, cloud services, on-premises infrastructure and AI security.
Rao said spending was no longer limited to large technology companies and financial services firms, with manufacturing, healthcare and retail also committing resources to AI infrastructure.
“What’s changed over the past 12 months isn’t just the amount being spent, it’s who is spending and on what,” he said.
He expects investment to continue growing over the next year, as more proof-of-concept projects move into production and tax incentives for data centres reduce the cost of adding capacity.
Indian enterprises are also adopting a mix of their own infrastructure, cloud services and edge computing, rather than relying on a single deployment model. Around 80% of Indian enterprises are standardising on hybrid models, according to Lenovo’s research.
Rao said the choice between owning infrastructure and using GPU-as-a-service depended on the workload. Cloud-based GPU services suited experimental projects and workloads with unpredictable demand, while owning or co-locating infrastructure could be more cost-effective when inference volumes were high and steady.
Inference, the process of running trained AI models to produce results, is becoming the main driver of infrastructure spending. Lenovo’s CIO Playbook projects that inference will account for 75% of AI computing by 2030. It also estimates that inference costs over a model’s lifecycle can be up to 15 times higher than training costs.
Generative AI remains the largest category of AI used by volume, including applications such as content generation, coding assistance and customer service. However, interest in agentic AI, which can plan tasks, use tools and carry out multi-step workflows, is expected to double over the next 12 months, Rao said.
The growing use of AI is also changing data-centre requirements, particularly for cooling and power. India’s data-centre cooling market is growing at more than 27% annually, according to figures cited by Rao. While air-based cooling remains the majority of the installed base, he said new facilities designed for high-density GPU deployments were increasingly turning to liquid cooling.
Lenovo has been manufacturing servers at its Pondicherry facility. Rao said the company expected to add local capacity as demand grew, particularly from hyperscalers, AI cloud providers and enterprises building their own AI infrastructure. He did not provide figures for current local production or planned capacity additions.
Rao said only around half of AI proof-of-concept projects currently reach production, highlighting a gap between experimentation and implementation.
Over the next three years, on-premises systems and edge infrastructure are expected to retain a role in workloads requiring low latency, sustained computing capacity or tighter control over data. Public cloud will continue to serve experimentation, model evaluation and workloads requiring flexible capacity, Rao said.
