India made identity, payments, and data free. Intelligence is next. No other country has deliberately built all three as a digital public infrastructure (DPI) stack at India’s scale. Others have developed only pieces of this. Estonia has a world-class digital identity system, but nothing comparable to UPI. Brazil’s Pix is an excellent free payments rail, but it stands alone. Singapore has Singpass and SGFinDex, while Europe has open banking and data portability. India’s genuine claim to global leadership in public digital infrastructure lies in this unique combination. Its distinctiveness is not any single component, but the integration of Aadhaar, the Unified Payments Interface (UPI), and DEPA/Account Aggregator as interoperable public digital rails.
Between September 2016 and 2019, the cost of a gigabyte in India fell from about $4 to under 30 cents, making it among the cheapest in the world. The result was not a telecom story. It was a civilisational one: 500 million people came online in half a decade, powering India’s digital economy such as payment volumes, the startup ecosystem, the direct-benefit transfers reaching the last village. Aadhaar enrolled 1.4 billion people and turned identity verification from an expensive paper process to a low-cost API call. UPI made digital payments effectively free, processing around 20 billion transactions a month at near-zero cost. India made data free not by subsidising it, but by letting one player absorb the fixed costs of a nationwide 4G network, price at marginal cost, and force every incumbent to match or die.
India should not be in the capital race to the bottom which we would lose to hyperscalers, and it is the intelligence equivalent of nationalising the telecoms. India crashed the price of data by engineering conditions and competition. So, the target is inference, not training: relentlessly drive down the cost of using a model. It released spectrum, created a demand shock, and let ferocious competition do the rest. The state built the road and set the rules; the market crashed the price. A billion people came online not because data was gifted, but because its cost fell below the threshold of thought. And the case for making it free — or as close to free as a gigabyte of data — is not utopian. It is the same case, run on the same playbook that worked three times before.
This is the model India should now apply to its fourth foundational utility: intelligence itself. Intelligence is no longer a luxury software product; it is foundational infrastructure.
The trade running against India
The global artificial intelligence (AI) race is currently operating under an extractive economic model, and India — despite its immense demographic dividend — is on the losing side of the trade. India serves as the world’s digital quarry. India supplies an outsized share of the world’s raw intelligence. Its engineers build and fine-tune the frontier models of Silicon Valley. Its digitised public life — languages, documents, transactions, behaviour — becomes training data.
Its universities and diaspora furnish a disproportionate slice of the research talent behind every major laboratory. Each day, Indians generate vast data; Indian engineers build Silicon Valley’s frontier models; and Indian workers label datasets to make them safer. Yet, Indian startups must rent that intelligence back as dollar-priced API tokens, subject to export controls and hosted on distant servers. India exports talent, data, and usage, then imports the finished intelligence at high-margin, per-token prices on terms set in San Francisco. The pattern is familiar. Ship out the cotton, buy back the cloth. India recognised that trap in textiles centuries ago. It should recognise it in intelligence now.
India’s AI token economy
India’s national AI token economy should be built on the following pillars. First, making intelligence free is raw compute power. Through the IndiaAI Mission, backed by an outlay of about ₹10,372 crore, the government is building a public-private partnership model — not a state-run data centre monopoly. By empanelling private cloud providers and aggregating demand, it has onboarded over 38,000 GPUs, with plans to scale to 1,00,000. Eligible startups and researchers can access compute for about ₹65 per GPU hour, a fraction of global market rates.
India’s grid planning does not yet treat AI load as a category. The single highest-leverage intervention available to the government is boring: fold compute into the National Electricity Plan, fast-track transmission to data-centre clusters, and dedicate renewables generation, including nuclear, to inference the way it once dedicated coal linkages to steel. Cheap electrons are the new cheap spectrum.
Second, cheap compute is useless without models to run on it. Currently, the most capable foundation models are closed source and owned by western tech giants. If Indian startups build their businesses on top of these proprietary APIs, they remain economically subservient. A single pricing change or policy shift in San Francisco can wipe out an entire sector in Bengaluru. India must mandate that foundational AI models become public goods.
The government must leverage its most valuable asset: linguistic and demographic data. The state should aggregate anonymised public data (legal rulings, agricultural data, educational curricula in all 22 official/scheduled languages) and make them available exclusively to researchers building open-source models. Further, any AI model developed using state-subsidised compute or public datasets must be released under an open-weights license. This mirrors the UPI philosophy: the government builds the rails, makes the protocol free, and lets private companies (Google Pay) compete on the user experience. By flooding the market with highly capable, open-source Indic LLMs (Large Language Models), India commoditises the “intelligence” layer. When foundational models are free and open, the economic value shifts from the model creators to the application builders — exactly where India’s strength lies.
Third, the unified intelligence interface (UII) or distribution. How does a rural farmer or a Tier-3 developer access this free intelligence? An API gateway built on the principles of DPI, a UPI for AI. UPI’s genius was not free payments as such; it was interoperability. A common protocol let any app talk to any bank, and switching costs fell to zero. Intelligence needs the same: an open, standardised interface through which any application can call any model — sovereign or private, open or proprietary — with shared standards for identity, consent, billing and safety.
Just as UPI abstracts away the complexity of inter-bank transfers, a national API gateway would abstract away the complexity of AI inference. Startups, government agencies, and educational institutions could plug into this gateway to access a suite of open-source models (translation, vision, text generation) for fractions of a penny.
To kickstart the ecosystem, the government could offer a national “freemium” model. Verified Indian startups or students through Aadhaar-linked digital identity could get monthly free API tokens, subsidised by the state. When startups scale and become profitable, they move to a commercial tier, paying slightly more for compute, which subsidises the free tier.
Additionally, a tenth of the fertilizer subsidy could be diverted to tokens for research and development institutions and schools. When intelligence becomes nearly free, sectors constrained by scarce human capital transform. A rural doctor, aided by AI trained on Indian data, can treat more patients with fewer errors; India’s 300 million students get 24X7 personalised AI tutors in local dialects; a Karnataka farmer can settle a crop-insurance claim by voice in Kannada, with AI accessing land records.
The next great equaliser
In sum, free identity and cheap data did not make Indians rich; they made mass participation frictionless. Free intelligence can do the same. Nobody foresaw that near-free data would create the world’s largest payments system, short-video platforms, and QR-code vegetable vendors.
Near-free intelligence could be just as generative if the state primes it with token entitlements for students and teachers, citizen services in local languages, and affordable tools for 60 million small businesses. India’s leadership in the AI economy is contingent on developing an open-source ecosystem with open weights no-kill switch models that expand competition, choice, opportunities and innovation. The goal is not better chatbots; it is making cognition no longer a class privilege.
Srivatsa Krishna is an IAS officer. The views expressed are personal
Published - July 31, 2026 12:16 am IST