GPU shortages, rising costs emerge as key risks to India’s AI compute ambitions, says Economic Survey

Budget Economic Survey 2025

• India's AI expansion faces bottlenecks from GPU shortages and rising costs
• Survey warns financing alone won't solve hardware and infrastructure gaps
• Agent Based Modelling proposed to identify policy intervention points

Did our AI summary help?

India’s push to rapidly expand data centres and AI compute capacity could face significant bottlenecks due to rising costs, infrastructure constraints and global shortages of critical hardware such as GPUs, the Economic Survey released on January 29 said.

The Survey said that while sufficient compute capacity is essential for training and developing cutting-edge artificial intelligence models, surging global demand for GPUs and limited supply of key components such as high-bandwidth memory chips and storage are pushing up costs. This could make the financial viability of large-scale data centre expansion increasingly challenging for India.

“Even if the sovereign, domestic investors, or financial institutions are willing to finance data centre expansion, plans may need to be put on hold until GPU supplies are secured,” the Survey noted, highlighting that higher demand from international buyers could choke the availability of GPUs for domestic players.

According to the Survey, the interaction between financial constraints, infrastructure readiness and external supply dependencies will determine the pace and scale of India’s AI compute expansion.

It cautioned that simply making financing available may not be sufficient if other bottlenecks such as grid availability or hardware access remain unresolved.

To analyse these risks, the Survey has proposed an Agent Based Modelling (ABM) framework, allowing financing, electricity grid capacity and access to hardware to interact through feedback loops. The objective is to identify scenarios where “coordination failures” can emerge, even when demand for AI infrastructure remains strong.

The government clarified that the exercise is a policy stress test rather than a forecasting tool. It does not attempt to estimate optimal compute capacity or exact investment needs. Instead, it seeks to uncover structural vulnerabilities in India’s AI infrastructure ecosystem and identify leverage points where policy interventions could be most effective.

The Survey also underlined the limits of price-based or subsidy-led approaches in isolation. When financial constraints, power infrastructure gaps and global hardware shortages coincide, subsidies alone may not accelerate capacity creation.