India 2035 Research
AI as Public Infrastructure
How should Indian governments build, buy and govern AI so that it serves citizens at population scale?
The India 2035 scenario asks whether India should act. These reports ask how, one decision at a time. Each takes a single question that governments face now, tests it against primary evidence, and treats the strongest objections as part of the argument.
Policy reports
3 publishedSubsidise, Own or Rent?
Designing the next phase of India's AI compute programme
India rented its way to one of the cheapest public AI compute pools in the world, and outbuilt its own target more than fourfold. The next constraint is not GPUs. It is allocation, co-payment design and fixed prices in a rising market. As the government moves to five-year, ring-fenced contracts, those design choices will decide whether the next phase works.
Consent at Population Scale
Governing public data for AI in India
India's digital public infrastructure has produced one of the world's richest records of how a developing economy works. Using that data to build AI suited to Indian languages and conditions is now official policy direction. The legal and institutional framework for doing so safely does not yet exist, and India's own evidence shows how data systems built for efficiency can exclude the people they are meant to serve.
Pilots Are Not Policy
A procurement framework for AI in Indian state governments
Indian states are adopting AI faster than they are building the means to buy it well. The constraint is mostly not the technology. It is the route from a promising pilot to a contract the state can scale, own, audit and exit. Procurement rules designed for laptops and taxi services, awarded at the lowest price, are the wrong tool for systems whose performance is only known after testing and which change after delivery.
In progress
Next in the seriesHow we work
- Primary sources, footnoted. Every figure links to an official release, government order, court record or named report.
- Objections answered in the text. Each report has a section written as though the strongest critic were reading.
- Practical outputs. Recommendations name an owner, an instrument and a timeline, so they can be acted on.
- Neutral tone. The evidence does the persuading.
Review, contribute, or share evidence
If you work in state government, public procurement, privacy law, compute infrastructure or school education and would like to review a draft or correct the record, write to rahul@india2035.ai.