Power, Compute, and Sovereignty: Why India Must Build Its Own AI Infrastructure in 2026
As frontier models demand ever-increasing compute for both training and inference, a question that once seemed theoretical has become urgent national policy: where should AI infrastructure live, and who controls it?
For India — home to the world's largest digital public infrastructure stack and a rapidly growing AI startup ecosystem — the answer will shape economic competitiveness, data sovereignty, and strategic autonomy for the next decade.
The Compute Gap India Must Close
India's current AI compute capacity sits far below what its ambitions require. While the country trains engineers at scale and deploys AI applications broadly, the GPU clusters that power frontier model training and high-throughput inference are overwhelmingly located in the United States and, to a lesser extent, Europe and East Asia.
| Region | Estimated AI Compute Capacity (2026) | Share of Global Total |
|---|---|---|
| United States | ~55 exaFLOPS | ~65% |
| China | ~18 exaFLOPS | ~21% |
| Europe | ~6 exaFLOPS | ~7% |
| India | ~1.2 exaFLOPS | ~1.4% |
| Rest of World | ~4 exaFLOPS | ~4.6% |
This asymmetry means that Indian enterprises, research institutions, and government agencies are running sensitive workloads on infrastructure they do not own, in jurisdictions whose laws may not protect their data or their citizens' privacy.
The Sovereignty Imperative
Three categories of AI workload make a strong case for domestic infrastructure:
1. Government and defence AI. Decision-support systems for national security, border management, and public safety cannot be processed on foreign-owned compute — not for legal reasons alone, but for basic strategic hygiene.
2. Healthcare and financial data. India's 1.4 billion citizens generate extraordinary volumes of sensitive health and financial data. The DPDP Act creates explicit data localisation obligations that hyperscaler shared infrastructure may not fully satisfy for the most sensitive categories.
3. Competitive intelligence. Indian enterprises using foreign AI APIs for product development, customer analytics, and supply chain optimisation are sending commercially sensitive signals to infrastructure they do not control.
What a Sovereign AI Supply Chain Looks Like
Building genuine AI sovereignty does not mean autarky — India should remain deeply integrated with the global AI ecosystem. It means ensuring that the critical path of AI development — foundation model training, sensitive inference, and strategic AI research — runs on infrastructure that India controls.
The components:
-
Domestic GPU capacity: The government's IndiaAI Mission targets 10,000+ GPUs in a shared compute pool. The private sector — Adani, Yotta, Reliance Jio — is building toward 100,000+ GPUs in India-based data centres by 2028.
-
Open-weight foundation models: Sarvam AI's work on large-parameter, multilingual open-source models provides a foundation that Indian enterprises can fine-tune without dependency on foreign API providers.
-
Domestic semiconductor research: India's semiconductor mission is a 10-year project, but early investments in chip design capability reduce long-term supply chain risk.
-
Academic compute access: IITs, IISc, and premier research institutions need subsidised access to frontier compute to keep Indian AI research competitive with US and Chinese peers.
The Policy Actions That Matter Most
The countries that control AI infrastructure in 2030 will set the norms for how AI is used — not just by their own citizens, but across their spheres of influence.
India's 2026 budget amendments, data centre tax holidays, and the IndiaAI Mission are the right signals. Execution will determine whether they become the right outcomes. The three policy levers with the highest near-term leverage:
-
Power allocation certainty for data centres — the single largest barrier to greenfield investment is unpredictable grid access.
-
Fast-track land clearance for large-scale compute campuses, particularly in Tier 2 cities where land cost and political resistance are lower.
-
Mandatory data residency for government AI procurement — creating a guaranteed demand base that justifies domestic infrastructure investment.