India’s DEPA AI training framework could become one of the most important pieces of digital infrastructure in the country’s artificial intelligence push. The government is proposing a system that would let startups, researchers and companies train AI models on valuable datasets held by hospitals, banks and public agencies without simply handing over the underlying raw data.
India has invested heavily in computing capacity and domestic AI models, but access to useful, high-quality data remains another major constraint. A model can have GPUs and skilled engineers and still struggle if the information needed for training is fragmented, legally sensitive or locked inside institutions that cannot safely share it.
DEPA AI Training Framework Tries to Unlock Data Safely
The draft Data Empowerment and Protection Architecture for AI Training, or DEPA Training, was unveiled by MeitY Secretary S Krishnan at a Bengaluru roundtable on October 30. The framework was prepared with technology think tank iSPIRT Foundation and is being described as digital public infrastructure for AI.
Its central idea is relatively straightforward: developers should be able to use data for legitimate AI training without automatically receiving unrestricted copies of sensitive records.
According to the framework, training could take place inside secure data environments. Privacy-preserving techniques such as differential privacy would help reduce the risk that individuals can be identified from the information used. Machine-readable electronic contracts would define what a developer is allowed to do with a dataset.
Some of India’s most useful potential AI datasets are also among its most sensitive. Healthcare records can improve medical models. Financial information can help build better credit and fraud systems. Government datasets can support public-service applications. But unrestricted transfers of such information would create obvious privacy, security and governance concerns.
Why Data Is Becoming as Important as Compute
The AI race is often discussed in terms of chips and data centres. India has been expanding access to GPUs through the IndiaAI Mission, while companies and states are investing in new AI infrastructure.
Data is the other half of the equation.
Large technology companies have access to enormous proprietary datasets and the resources to negotiate private data arrangements. Smaller Indian startups and research groups generally do not. If valuable institutional data remains inaccessible, the market can tilt toward companies that already have scale.
The DEPA AI training framework is an attempt to reduce that imbalance without treating privacy as an obstacle to innovation.
The document says the architecture could help unlock as much as $500 billion in AI-driven value for India’s GDP by 2030. That figure should be treated as an estimate rather than a guaranteed economic outcome. The more important question is whether the framework can create practical, trusted access to datasets that developers currently cannot use.
India Already Has a Digital Public Infrastructure Playbook
India has repeatedly used shared digital rails rather than relying on one company to own the entire system.
UPI created interoperable infrastructure for payments. Aadhaar became a large-scale identity layer. Account Aggregator infrastructure created consent-based mechanisms for financial-data sharing.
DEPA Training tries to extend a similar philosophy into artificial intelligence.
That does not mean AI training data can be handled exactly like a payment instruction. Model training creates different risks. Data can influence a model long after a training session ends, and researchers still debate how reliably sensitive information can be removed or prevented from resurfacing.
The Privacy Test Will Decide Whether the Model Works
The strongest version of DEPA would give Indian AI developers access to useful information while keeping data providers and individuals in control. The weakest version would create another layer of compliance without making enough quality data genuinely usable.
Clear rules will be needed around consent, accountability, audits, permitted purposes, model outputs and what happens when something goes wrong. Institutions will also need confidence that participation will not expose them to uncontrolled legal or cybersecurity risk.
The framework’s use of secure environments and privacy-preserving techniques is therefore significant, but implementation will determine whether those safeguards work in practice.
This is especially important as India moves toward a broader AI regulatory framework. The government has indicated that consultations on AI legislation are coming, while regulators such as the Reserve Bank of India are separately developing controls for AI use in financial services.
What It Means for India’s AI Ambition
India does not need to recreate every advantage held by American or Chinese AI companies. It needs infrastructure that makes its own strengths usable.
One of those strengths is scale. India has enormous volumes of financial, healthcare, language, commerce and public-sector data. Much of it, however, cannot simply be copied into an AI training pipeline.
The DEPA AI training framework is an effort to turn that constraint into an infrastructure problem that can be solved.
If it succeeds, Indian startups could gain access to richer training environments without requiring ownership of the underlying datasets. Hospitals, banks and government bodies could participate without surrendering complete control of sensitive information.
That is a more consequential proposition than another chatbot launch. Compute gives AI models the ability to learn. Trusted access to relevant data determines what they can learn from.
India has spent years building digital rails for identity, payments and financial data. The next test is whether it can build equally credible rails for artificial intelligence.
