MeitY is looking to add 25,000 GPUs (primarily NVIDIA B200) to the IndiaAI Compute Portal, which currently offers around 40,000 GPUs at subsidised rates. While on one hand, this move signals the government responding to growing demand, the reality is disappointing. Launched in March 2025, the portal has just over 300 users and is heavily underutilised (only 15,114 GPUs in use). This may simply be a result of bureaucratic hurdles delaying access to a key resource required for fast innovation.
This brings us back to the fundamental question of how India can democratise access to compute. Let’s first look at access to the IndiaAI Compute Portal. Eligible users include DPIIT-recognised startups with demonstrated AI experience, researchers with a minimum h-index of 5 or equivalent citations, MSMEs, and government bodies. To gain access, applicants must register with supporting documents like startup certificates, submit detailed project proposals, and await approval. Small requests under 5,000 GPU-hours get monthly auto-approvals, but larger allocations require review by the Program Monitoring and Evaluation Committee (PMEC).
The platform provides up to 40% subsidies through public-private partnerships, but it operates as a siloed system with fixed quotas, limited GPU types, no standardised open APIs, and no requirements to share resulting models or datasets openly.
Government subsidies for compute create a fundamental trade-off by eliminating financial barriers but introducing administrative ones. While the portal sets prices below market rates, subsidised compute is limited, and hence, access to the same must also be limited to minimise wastage. This is managed through measures such as rigorous eligibility vetting. Consequently, the cost of the resource shifts from capital expenditure to time and agility. While developers gain affordability, they become restricted by bureaucratic approval cycles.
Moreover, the portal remains isolated, lacking peering or global open standards, which limits users to technical silos of this centralised infrastructure.
In contrast, open compute clusters rely on a mix of private and public clouds with open characteristics. This includes standardised APIs for self-service access across providers like AWS, Azure, or local operators. Users often receive vouchers or credits redeemable on any compliant platform, with reciprocity rules mandating that outputs like models and datasets be released openly. Hardware predominantly follows Open Compute Project (OCP) designs for cost efficiency and interoperability. With minimal bureaucracy, startups gain access to resources instantly via familiar cloud dashboards, can scale without quotas, and peer with global networks.
Such open compute clusters minimise friction, embed developers in interoperable ecosystems, and ensure that every subsidised hour also contributes to building national commons. By issuing redeemable vouchers usable on private clouds with open APIs and reciprocity rules, the IndiaAI Mission could blend subsidies with instant access and fast innovation without building redundant silos.