India’s government is proposing a national framework to let startups, researchers and companies train artificial intelligence models on data now locked inside hospitals, banks and public agencies.
The Data Empowerment and Protection Architecture (DEPA) for AI Training, a draft framework steered by the Ministry of Electronics and Information Technology (MeitY), says it “provides the potential to unlock $500 billion in AI-driven value for India’s GDP by 2030.”
The plan would route model training through secure data environments, differential privacy and machine-readable electronic contracts, keeping providers’ raw data out of developers’ hands.
MeitY Secretary S Krishnan on Friday unveiled the DEPA for AI Training Handbook at a roundtable in Bengaluru. Speakers included Principal Scientific Adviser Prof. Ajay Kumar Sood, Infosys co-founder Kris Gopalakrishnan, HDFC Bank’s Anjani Rathor, iSPIRT Foundation co-founder Sharad Sharma, Emergent AI executive chairman Manav Garg and Ashwani Sharma of Sarvam.
The framework, known as DEPA Training, is described in the report as India’s Digital Public Infrastructure for AI. It was prepared with the iSPIRT Foundation, a technology think tank.
The report says the consequences of inaction could be severe. It warns that without such a framework, innovation would be stifled for Indian startups and researchers. It also warns of increased market concentration favouring large entities. It ties the framework to a ₹20-25 trillion credit gap for micro, small and medium enterprises (MSMEs), an estimate it attributes to an expert committee constituted by the Reserve Bank of India.
The ‘Data Deadlock’
The report calls the core problem a “Data Deadlock.” Data required to solve the next generation of problems sits divided across domains. It is safeguarded behind different institutional and regulatory walls.
“India is data rich. The critical challenge is to convert this wealth into real world solutions,” the report says.
It cites four barriers. They are privacy risks, institutional and business risks, a lack of trusted and accessible frameworks, and a cost-benefit analysis that often tilts against sharing. Removing names and ID numbers is not enough, the authors argue. “Simple anonymization won’t work due to re-identification risks and utility loss,” the report states.
The framework is distinct from DEPA Inference, which operates through the Account Aggregator model in finance and handles consented sharing of individual data. DEPA Training instead targets bulk, de-identified data used to build models in the first place.
How it would work
The architecture rests on four technical pillars. They are a Data Exchange Platform, Electronic Contracts, a Secure Data Environment and Differential Privacy.
Training Data Providers, such as hospitals, banks and public agencies, would register datasets. Training Data Consumers, including startups, researchers and enterprises, would request access for a specific use case. Intermediaries called Data Collabs would aggregate and standardise data across sources.
The report sets out two options. In the first, data providers anonymise data, and consumers can download it or use a secure data environment. In the second, the Data Exchange Platform handles anonymisation and encryption, and consumers get access only inside the secure environment.
Under the proposed Data Sharing Procedure, personal and sensitive data would not be shared in raw form. Trained models would face bias, security and explainability checks before deployment, under guidelines from the competent government authority.
“In this model, compliance is not a barrier to innovation; it is the feature that unlocks it,” the report says.
The report envisions a DEPA-based grid in the future where data flows safely and freely between producers and consumers, powering a thriving AI economy built on security and trust.
Sector opportunity
The report pegs India’s financial services landscape at $550 billion. It says DEPA could potentially extend formal credit to 400 million underserved individuals and MSMEs. It notes that MSMEs contribute nearly a third of India’s GDP.
It also points to RBI action against major banks over IT deficiencies. The report says “many incumbent institutions are burdened with outdated legacy systems that are technically incapable of supporting modern, secure data environment.”
In healthcare, the report describes a $600 billion-plus market. It cites a Penn Medicine and Intel project across 71 international institutions that achieved over 99 per cent of the accuracy of a hypothetical model trained on centralised data.
In education, it cites 250 million students and a $29 billion EdTech sector.
Governance and rollout
Oversight would rest with a consortium of participants, including data providers, consumers, technology service providers, domain experts and civil society. The competent government authority would certify models and could suspend or revoke non-compliant activities.
The roadmap starts with pilots in finance and healthcare, with regulatory sandboxes in parallel. Later steps include cross-border agreements, capacity building and incentive models such as royalty or access-fee structures.
“Inaction is not a neutral stance; it actively allows risks to grow and opportunities to wither,” the report says.
