India Weighs Allowing Domestic GPU Providers to Export AI Compute

The central government is evaluating a proposal to permit domestic graphics processing unit cloud operators to sell computing capacity to foreign commercial clients. This policy consideration emerges amid mounting official anxiety over subdued domestic uptake, even as Indian artificial intelligence startups and academic researchers face severe hurdles accessing affordable hardware. Under the 10,372 crore rupee IndiaAI Mission, authorities empanelled fourteen infrastructure firms including Jio Platforms, Tata Communications, and Yotta Data Services to provide subsidized computing resources. Allowing overseas sales is intended to help private infrastructure vendors recoup massive capital investments in high-end computational clusters while the domestic AI ecosystem matures. However, industry participants argue that weak uptake does not reflect absent demand, but rather stems from structural misalignments in pricing, hardware architecture, and bureaucratic access.

Under the Uttar Pradesh Data Center Policy, data center units receive a 100% electricity duty exemption for 10 years and a 50% concession on transmission and wheeling charges for 25 years.

The IndiaAI Compute Portal approved 237 end-users for subsidized compute support, with approximately 93.18 lakh GPU hours sanctioned across government entities, startups, MSMEs, researchers, and academic institutions.

Under the government's voucher system, Indian startups obtain access to high-performance GPUs at a subsidized rate of approximately ₹115 per hour.

Hardware Fragmentation and Frontier Model Constraints

India's deployed computational infrastructure remains heavily fragmented across multiple independent data centers, making it suitable for inference tasks rather than centralized training. Foundation models demand thousands of high-end processors interconnected via ultra-high-bandwidth fabrics to operate as a coherent distributed system. Furthermore, only a minimal fraction of domestic installations comprise cutting-edge silicon such as Nvidia B200 processors required for demanding training runs.

Commercial Misalignments and Developer Economics

Application startups frequently find it economically advantageous to consume commercial application programming interfaces from international providers like OpenAI and Anthropic rather than leasing dedicated infrastructure. Sasank Chilamkurthy questioned the claim that India lacks commercial compute demand, emphasizing that adoption bottlenecks are fundamentally driven by pricing. Because the IndiaAI Mission relies on private operators rather than state-owned assets, escalating global rental rates incentivize providers to prioritize high-paying corporate accounts over subsidized startups. Manoj observed that regional GPU rental expenses have surged by 50 percent within a six-month window, discouraging providers from locking into fixed-rate subsidized commitments.

H100 SXM rental costs $2.69 per GPU per hour on Jarvislabs.

Administrative Friction and Developer Migration

According to researchers and entrepreneurs, cumbersome verification procedures and bureaucratic documentation frequently deter early-stage developers from utilizing subsidized capacity. Krishna Sangeeth KS recounted spending half an hour filling out subsidized compute applications for an experimental project budgeted under 10,000 rupees. After being prompted to submit formal annual turnover records and meet rigid corporate eligibility thresholds, Sangeeth abandoned the subsidized portal in favor of Amazon Web Services. He explained that acquiring equivalent cloud capacity on global commercial hyperscalers required merely submitting an instant quota expansion request.

Industry Perspectives and Transitional Frameworks

Sameer Redij argued that the Indian compute landscape is undergoing a structural transition from raw capacity build-out to productive workload optimization. Redij emphasized that although the national mission provides subsidies reaching up to 40 percent, affordable hourly compute must be paired with networking and engineering support. Shashi Shekhar Vempati noted that general compute availability has expanded substantially compared to acute shortages experienced two years ago. Vempati highlighted targeted funding via the Anusandhan National Research Foundation and the Research, Development and Innovation Fund as vital channels for academic researchers and startups.

In 2024, the Indian Cabinet approved the IndiaAI Mission with a budget of over ₹10,300 Crore to build national compute capacity.