India Bets $200 Billion on AI Sovereignty — and the Real Trade Isn't What You Think
New Delhi, February 19, 2026. Inside Bharat Mandapam, where 250,000 attendees have crowded a summit that dwarfs every prior AI gathering on earth, Indian Prime Minister Narendra Modi stood shoulder-to-shoulder with OpenAI's Sam Altman and Anthropic's Dario Amodei for a unity pose that instantly went viral — not for the solidarity, but for its absence. Altman and Amodei raised separate fists rather than join hands. "I wasn't quite sure what was happening," Altman said afterward. The moment, sharp and unscripted, was an accidental metaphor for an entire industry.
The India AI Impact Summit 2026 (February 16–21) is the fourth in a global series following Bletchley Park, Seoul, and Paris — but the first hosted in the Global South. That geography is the point. Where prior summits wrestled with safety and regulation, New Delhi is asking a harder commercial question: who builds, owns, and profits from AI infrastructure at planetary scale?
The Capex Is Real — But Read the Fine Print
Strip out the pageantry and the summit reduces to one signal: India is treating AI as infrastructure, not software. The numbers are staggering in aggregate. Reliance pledged ₹10 trillion over seven years into AI data centers and edge computing, anchored by a 120MW facility in Jamnagar coming online by late 2026. Adani committed $100B into renewable-powered AI data centers by 2030. Google CEO Sundar Pichai announced a $15B AI hub in Visakhapatnam with gigawatt-scale compute, fiber-optic connectivity via America-India Connect routes, and a $30M Google.org challenge for AI in government services across 18 languages. OpenAI formalized a partnership with Tata Consultancy Services, beginning at 100MW of dedicated capacity — badged under the "Stargate" project — with optionality toward 1GW. Nvidia is building what it calls India's largest gigawatt-scale AI factory alongside L&T, and separately partnering with three cloud providers for Blackwell Ultra chip deployments via Yotta Data Services' planned $2B+ AI hub.
Critically, "100MW now, potentially 1GW later" is optionality, not a financed gigawatt. Investors should interrogate what percentage of announced megawatts carry signed anchor tenants, what power-purchase agreement prices look like, and who bears GPU procurement risk. The Economist has already called India's broader plans "fanciful," citing limited domestic chip manufacturing. That critique has merit — but it targets the wrong layer.
Where Value Actually Accrues
India is not competing to win frontier model training. It is competing to dominate inference and deployment at scale, in many languages, for the Global South. That distinction changes the investment calculus entirely.
On the model side, Sarvam AI launched two LLMs — a 30B-parameter model for real-time applications and a 105B-parameter mixture-of-experts model with 128K context — both supporting 22 Indian languages and benchmarking above Gemma-27B and Mistral-32B. The government-backed BharatGen released Param2, a 17B-parameter multilingual model built with Nvidia for governance, healthcare, and education. PM Modi's "MANAV" framework promises 7,500+ shared datasets and 270 models for startups nationally.
But sovereign model IP is not where the moat lives. The durable advantage belongs to whoever controls latency, local language quality, data residency compliance, and workflow integration — a distribution and operations moat, not an academic benchmark moat. That framing explains the Anthropic-Infosys deal: Claude models deployed enterprise-wide, beginning in telecom via a dedicated Center of Excellence, with Infosys shares jumping 5% on the announcement. Indian IT's bull case is not building frontier models. It is operationalizing AI inside messy, regulated enterprises — with the bear case being margin compression as automation eats into delivery costs.
The Real Governor: Power
Every gigawatt-scale ambition in this summit eventually collides with the same constraint: cheap, reliable electricity. This is why Adani and Reliance — conglomerates with deep renewable energy positions — are structurally advantaged in the AI factory narrative in ways that pure-play software companies are not. AI data centers are electricity-to-revenue converters. Investors should treat India's AI sovereignty story as a multi-year energy and infrastructure capex cycle first, and a software boom second.
The voluntary "New Delhi Frontier AI Impact Commitments" framework, deliberately non-binding, is pro-growth near-term. It is also a medium-term tail risk: one high-profile misuse incident — deepfakes, election interference, autonomous agent failure — could trigger the abrupt regulatory tightening that India is currently sidestepping. Bill Gates canceled his keynote over unrelated controversy, a reminder that reputational risk travels fast in this arena.
For investors, the clearest edge is boring execution over celebrity narrative: signed MW, contracted GPU delivery, PPA pricing, utilization rates. Everything else is a fist in the air.
NOT INVESTMENT ADVICE
Sources: India AI Impact Summit official page: https://impact.indiaai.gov.in
About the summit: https://impact.indiaai.gov.in/about-summit
Sessions schedule (Feb 19): https://impact.indiaai.gov.in/sessions?date=2026-02-19