What India needs to do to take advantage of the AI race

In Part 1, I argued that India’s “application-first” AI strategy is a dangerous delusion. The economic math of generative AI is brutal: 85% of financial value sits in the semiconductor and infrastructure layers, leaving a mere 15% sliver for applications. India’s historical advantages of labour arbitrage and scale engineering headcount are obsolete against hard Intellectual Property (IP), chips, and infrastructure. To avert a new era of AI colonisation, India must execute a radically reimagined sovereign AI blueprint.

The trillion-dollar chasm

The Indian government has taken meaningful early steps via the India Semiconductor Mission, ₹1 lakh crore in RDI (research, development and innovation) allocations, and the ₹10,000-crore IndiaAI Mission. Yet, the rest of the world operates on a scale that renders these efforts modest. The chasm is stark: while OpenAI and Anthropic target trillion-dollar valuations and raise hundreds of billions, India’s flagship success, Sarvam AI, recently raised $234 million, the budget its competitors spend on a single cluster training run.

Geopolitical dynamics compound the danger. US and Chinese export controls can weaponise access to chips, models (e.g., Anthropic Fable 5 and Mythos 5) and technologies. If India relies entirely on Silicon Valley tokens or foreign open-weight architectures, our AI economy can be disabled by a single regulatory pen stroke in Washington DC or Beijing.

The radically reimagined sovereign AI blueprint: five asymmetric bets

Symmetric war—building the exact same stack later and with a fraction of the capital—is a losing strategy. India cannot out-spend its way to sovereignty, it must out-think it. The next era of AI will reward whoever can power, manufacture, and feed intelligence at population scale. India must leapfrog using five asymmetric bets (see figure below).

Bet I – Energy first: Megawatts, not GPUs – building the AI-native power grid

The binding constraint on sovereign AI is electricity, not silicon. While America’s grid is power-starved and congested, India is building its energy system now and can make it AI-native. India must aspire to be the cheapest place on Earth to run a watt of compute by:

• Carving AI compute power out of expensive industrial cross-subsidy tariff regimes.

• Designating gigawatt-scale AI Power Parks that co-locate renewables and storage behind the meter.

• Vaulting from 1.6 GW of AI data-centre capacity to 50 GW of dedicated AI power by 2032.

Bet II – Advanced packaging and frontier silicon, not the ‘licensed’ fab

Sovereignty will not be won by attempting to out-fabricate TSMC at the two-nanometre frontier. That is a two-hundred-billion-dollar race, we would enter a decade late and lose. The asymmetric path is competing on advanced packaging and chiplets, which govern how dies communicate, where modern AI performance is increasingly bottlenecked not by transistor count but by data transfer speeds between chips. Whoever masters heterogeneous integration controls the performance envelope of the resulting accelerator, and that mastery can be built without competing for ever-smaller lithography nodes dominated by ASML and the North & East Asian fabs that wield them. This controls the performance envelope without entering the Node War. Under the ISM 2.0, India must promote frontier silicon technology such as silicon photonics, ultra-wide-bandgap power transistors and cutting edge packaging which dramatically enhance AI performance (per unit of compute) particularly at the emerging frontier of inference.

Bet III – Data sovereignty: the population-scale data moat

With 1.5 billion internet-savvy people and 22 languages, India possesses the world’s largest real-world training environment. As frontier models approach the “data wall”—the point at which homogeneous internet text is exhausted and only diverse, real-world multimodal data can drive further capability gains—India’s position shifts from latecomer to gatekeeper. We must convert this latent advantage into deliberate strategy: a National Sovereign Data Trust that holds the world’s largest consented multimodal corpus across health, commerce, agriculture, mobility, and language. This trust would treat data not as a commodity to be exported but as sovereign capital to be leveraged and selectively strategically deployed.

Bet IV – Physical AI and the sovereign agent stack

The frontier is shifting from text to the physical world. As models saturate the written word, the next leap belongs to physical AI and world models—systems whose true bottleneck is not text scraped from the internet but real-world data and real-world environments. Here India can move first: a national drive that trains robots on the floors of Indian factories, homes of the aspirational Indian consumer and in the fields of Indian farms would generate exactly the diverse, multimodal data the next generation of models demands. The vehicle enabling this is a sovereign Agentic Public Infrastructure—India’s DPI transformed into an API. Rather than replicate yet another text-generating model, India could lay an open-source, government-backed AI agent layer directly atop its proven digital rails—Aadhaar, UPI, ONDC—a network of action-oriented micro-agents embedded in public and state systems and open to enterprises at home and abroad. Foreign players wishing to commercialise their technology may train on anonymised, state-curated Indian data, surrendering compute equity in return. The India Stack would become the deployment rail for sovereign models; and a Sovereign Compute Commons would pool a share of hyperscalers’ compute capacity for startups and researchers.

Bet V – Asymmetric capital and the third bloc

India will not be able to match Silicon Valley dollar for dollar. Two moves change the game. The first is financial: a Sovereign AI Fund that treats AI-energy-compute infrastructure as a strategic national asset class – anchored by government equity (e.g., NPS, LIC) and amplified by sovereign-backed compute bonds that crowd in private capital. The second is geopolitical: India should stop petitioning for a seat at the American or Chinese table and instead build a “third table” of its own. This coalition would be anchored by India, joined by Gulf, Norwegian, Singaporean sovereign wealth actors and the wider Global South. India contributes the one thing this bloc cannot assemble without New Delhi: the world’s deepest pool of AI talent, a 1.5-billion-people market, and democratic legitimacy.

One machine, one mission: the case for Bharat future

Scattered efforts across uncoordinated ministries achieve nothing. India needs a new single national vehicle, Bharat Future, driven by two bodies:

• Mission authority: Vested with cabinet rank within the Prime Minister’s Office, holding single-window powers over land, power, and clearances to operate at a wartime tempo.

• Sovereign AI holding company: State-anchored but commercially run to hold equity in Power Parks, silicon ventures, and the data trust.

Bharat Future requires a chief executive from the global Indian diaspora, international co-investors on its board, and a hard 15-year sunset clause so it builds the machine and hands it to the market.

Path towards leadership

Make no mistake about the scale of this ambition. This blueprint demands a paradigm shift—deploying hundreds of billions of dollars over the next five to seven years with a clear theory of victory. It demands that our policymakers wield sovereign instruments rather than incremental incentives; that our universities pivot from consuming software to creating hard IP; and, above all, that our conglomerates and industrial titans step forward with massive private risk capital into power, silicon, models, and data. The prize justifies the audacity. AI is as foundational to this century as electricity was to the last. By executing these five asymmetric bets under a single national mission, India can transcend the role of a mere consumer and emerge as a genuine pole of the global AI order, securing a self-reliant Viksit Bharat, well before India celebrates a hundred years of freedom from colonial servitude.

(Part 2 of a two-part series. The author is Chairman of Hyperion Ventures Corporation, a global AI-led manufacturing platform focussed on semiconductors, rare-earth magnets and electronic components. Views are personal.)