Critically evaluate the role of public compute infrastructure in nurturing India's AI start-up ecosystem.

Compute is the costliest input in frontier AI. The IndiaAI Mission, approved by the Cabinet in March 2024 with an outlay of ₹10,371.92 crore over five years, treats it as a public utility through the IndiaAI Compute Capacity pillar [1]. This has measurably lowered entry barriers for start-ups, though access alone does not guarantee a competitive ecosystem.

Enabling role — the case in favour

• Cost subsidisation: over 38,000 GPUs onboarded on a common compute portal are offered to start-ups and academia at subsidised rates, converting a prohibitive capital cost into a variable one [2].
• Demonstrated output: Varya, a distilled indigenous video-generation model by Avataar launched in June 2026, uses distillation to cut inference from 50 steps to 4, delivering video at roughly 50 paise per second [3].
• Ecosystem depth: 12 teams shortlisted for indigenous foundational models and 30 India-specific AI applications approved show compute translating into research pipelines, not idle capacity [2].
• Strategic autonomy: domestic frontier models reduce dependence on foreign video and language models, complementing the Startup Financing pillar [3].
• Public–private hybrid: government-pooled compute with private model development keeps the state an enabler, not a builder [3].

Critical limitations

• Import dependence: the pool rests largely on imported accelerators; without the Semiconductor Mission maturing, sovereignty is at the software layer alone.
• Absorption gap: capacity has expanded faster than demonstrated demand, risking under-utilised subsidy.
• Skill and data bottlenecks: talent and quality Indian-language datasets constrain outcomes even where GPUs are free.
• Safety externalities: cheap video generation amplifies deepfake risks, making the Safe & Trusted AI pillar a precondition, not an afterthought [2].
• Concentration risk: benefits may cluster among a few well-networked firms.

Public compute has credibly shifted India from consumer to producer of AI, but it is a necessary rather than sufficient condition. Pairing subsidised GPUs with datasets, FutureSkills talent and enforceable trust safeguards will convert infrastructure into a durable, inclusive innovation ecosystem.