India’s Next AI Breakthrough Could Be Data: Inside the DEPA Framework for AI Training 🇮🇳🤖
India’s next big artificial intelligence advantage may not come from a new GPU, a larger data centre, or even a more powerful large language model.
It could come from something every AI system fundamentally depends on:
High-quality data.
The Indian government has introduced a draft Data Empowerment and Protection Architecture (DEPA) for AI Training, a framework aimed at enabling startups, researchers, and companies to train AI systems using protected institutional data while putting safeguards around how that data is accessed and used.
Steered by the Ministry of Electronics and Information Technology (MeitY) and developed with support from the iSPIRT Foundation, the framework addresses one of the most difficult questions facing India's AI ecosystem:
How can valuable data be made useful for AI innovation without compromising privacy, security, or institutional control? 🔐
The ambition is significant.
The draft handbook estimates that unlocking data responsibly for AI could potentially contribute to $500 billion in AI-driven value for India's GDP by 2030.
But the bigger story is about solving what the framework calls India's “Data Deadlock.”
What Is the “Data Deadlock”?
AI systems learn from data.
The more relevant, representative, and high-quality the data, the greater the potential to build useful AI systems.
India has enormous volumes of data distributed across sectors such as:
🏦 Banking and financial services
🏥 Healthcare
🏛️ Government and public services
🏢 Enterprises
📊 Public institutions
The problem is that much of this information cannot simply be collected and handed over to AI developers.
And for good reason.
Banking records can contain highly sensitive financial information.
Healthcare datasets can contain confidential medical information.
Government databases may contain protected citizen information.
There are also regulatory requirements, security concerns, commercial sensitivities, and questions around consent and permitted use.
As a result, organisations may possess valuable datasets but remain unable or unwilling to make them available for AI development.
The DEPA handbook describes this situation as a Data Deadlock.
The data exists.
AI developers need it.
But privacy, regulatory and institutional barriers prevent the two sides from connecting safely.
DEPA for AI Training attempts to create a structured bridge between them.
Why Data Access Matters for India's AI Ambitions
The global AI race is often described as a competition for computing power.
But compute is only part of the equation.
A simplified AI development equation looks more like:
Compute + Algorithms + High-Quality Data + Talent = Powerful AI Systems
India can invest heavily in GPUs and AI infrastructure, but models still need relevant datasets to learn patterns that reflect Indian businesses, institutions, languages, healthcare systems, financial behaviour and public services.
Without appropriate datasets, AI systems can struggle to understand local contexts.
This makes access to trustworthy data a strategic AI capability.
The challenge is ensuring that access does not mean uncontrolled exposure.
That distinction is at the heart of the proposed framework.
The Four Pillars of DEPA for AI Training
The framework proposes four major building blocks designed to make institutional data more accessible while maintaining safeguards.
1️⃣ Data Exchange Platform
The first pillar is a Data Exchange Platform.
Instead of organisations exchanging sensitive datasets through informal or fragmented mechanisms, a structured platform could help connect:
Data providers → Approved AI users → Secure processing environments
Hospitals, banks, public agencies and other institutions could potentially participate as data providers.
Startups, researchers and companies could seek access for approved AI use cases.
The important idea is not simply moving datasets from one organisation to another.
It is creating a governed mechanism for controlled data access.
That could make data collaboration more transparent and scalable.
2️⃣ Electronic Contracts
Data sharing is not only a technical problem.
It is also a legal and governance problem.
Who can access the data?
What can they use it for?
How long can they access it?
Can they redistribute it?
What happens after the project ends?
Who is responsible if the agreed conditions are violated?
The framework proposes electronic contracts to help establish clear rules between participating organisations.
These contracts can define important conditions around:
🔹 Purpose of data use
🔹 Access permissions
🔹 Responsibilities
🔹 Restrictions
🔹 Compliance requirements
🔹 Data-use conditions
This could make institutional AI collaborations more accountable.
Instead of relying purely on trust, access can be governed by explicit agreements.
3️⃣ Secure Data Environments
This is potentially one of the most important technical ideas in the framework.
Traditional data sharing often follows a simple model:
Organisation → Copies dataset → Sends dataset → External party processes it
That approach can create obvious security and privacy risks.
A Secure Data Environment changes the model.
Instead of freely transferring sensitive raw data, approved users can potentially perform analysis or AI development inside a controlled computing environment.
Conceptually:
Bring the AI workload to the protected data instead of unnecessarily moving protected data to the AI developer.
Access can then be restricted, monitored and governed.
This approach can reduce unnecessary exposure of sensitive datasets while still allowing useful computation.
For industries such as healthcare and finance, this could be particularly important.
4️⃣ Differential Privacy
Even when obvious identifiers are removed from datasets, privacy risks do not automatically disappear.
In some cases, information can potentially be inferred or individuals could be re-identified by combining different attributes.
That is where differential privacy becomes important.
Differential privacy is a mathematical privacy technique designed to reduce the ability to determine whether information about a particular individual influenced an output.
One common approach involves carefully introducing statistical noise while preserving useful aggregate patterns.
The goal is to achieve a difficult balance:
Protect individuals while preserving enough statistical value for useful analysis and AI development.
This makes differential privacy particularly relevant when working with sensitive datasets.
Why Finance Could Be One of the Biggest Beneficiaries
The initial rollout is expected to prioritise pilots in finance and healthcare.
Finance provides an interesting example of why improved data access could matter.
Millions of people and small businesses may have limited traditional credit histories.
A lender using only conventional credit records may struggle to evaluate these customers.
With appropriate safeguards and permissions, broader datasets and AI-based analysis could potentially help institutions understand financial behaviour more effectively.
According to the framework, the approach could help extend formal credit opportunities to as many as 400 million underserved individuals and MSMEs.
That could have implications for:
💳 Credit accessibility
🏪 MSME financing
📊 Risk assessment
🏦 Financial inclusion
🤖 AI-powered underwriting
The key challenge will be ensuring that increased data use also comes with strong protections against misuse, unfair profiling and inappropriate access.
Why Healthcare Is Another Important Starting Point
Healthcare contains some of society's most valuable — and most sensitive — datasets.
Hospitals and healthcare organisations generate information that could potentially support AI applications including:
🩻 Medical imaging analysis
🧬 Disease research
📈 Predictive healthcare
💊 Drug and treatment research
🏥 Hospital resource optimisation
🧠 Clinical decision-support systems
But healthcare information requires particularly strong privacy and security safeguards.
This creates a difficult trade-off.
Locking healthcare data away completely can limit research and innovation.
Making it freely available can create unacceptable privacy risks.
The goal therefore should not simply be “more data sharing.”
It should be responsible, controlled and privacy-preserving data access.
DEPA for AI Training is attempting to create an architecture around that middle ground.
The $500 Billion AI Opportunity
One of the biggest numbers associated with the proposed framework is $500 billion.
The handbook estimates that responsible data unlocking could help generate approximately $500 billion in AI-driven value for India's GDP by 2030.
Why could data have such a large economic impact?
Because data is not valuable only to technology companies.
AI trained on high-quality sector-specific information could potentially improve productivity across:
🏦 Banking
🏥 Healthcare
🏭 Manufacturing
🚚 Logistics
🌾 Agriculture
🏛️ Public services
🛒 Commerce
📚 Education
When organisations can build AI systems using more relevant and representative datasets, they may be able to automate processes, improve predictions, personalise services and make better decisions.
In that sense, data infrastructure can become economic infrastructure.
The Critical Balance: Innovation vs Privacy
The potential is enormous, but so are the risks.
Making institutional data available for AI development raises serious questions.
How will access be authorised?
Who determines whether an AI use case is legitimate?
How will misuse be detected?
How will sensitive information be protected?
How will AI systems be prevented from learning or exposing information they should not?
How will organisations demonstrate compliance?
How will citizens retain meaningful protections?
These questions matter because AI innovation cannot be sustainable if it destroys trust.
The strongest version of a framework like DEPA would therefore need to support two objectives simultaneously:
Enable innovation.
Protect individuals and institutions.
Neither objective should be treated as optional.
Why Startups Could Benefit
Large technology companies often have access to significant proprietary datasets.
Startups usually do not.
This creates a structural disadvantage.
A young AI company may have excellent engineers, a strong model architecture and an innovative product idea — but still lack the high-quality domain data required to build the system.
A well-designed data-access architecture could potentially lower this barrier.
Instead of needing to own enormous datasets, startups could potentially gain controlled access to approved datasets under defined rules.
That could encourage innovation in sectors where data has traditionally been difficult to access.
From Digital Public Infrastructure to AI Infrastructure
India has already built large-scale digital systems supporting identity, payments and digital services.
The AI era introduces a new infrastructure challenge.
Future digital infrastructure will need to support not only transactions but also data-intensive intelligence.
That requires several layers:
Compute Infrastructure
↓
Cloud & Data Centres
↓
Connectivity
↓
Trusted Data Infrastructure
↓
AI Models & Applications
DEPA for AI Training focuses on one of the most difficult layers in this stack — trusted access to data.
Interestingly, this also complements India's broader AI infrastructure push.
Building AI-ready fibre networks and data centres solves the compute and connectivity problem.
Building privacy-preserving data frameworks addresses the data-access problem.
India will need both.
“Inaction Is Not a Neutral Stance”
Perhaps one of the most striking statements in the handbook is:
“Inaction is not a neutral stance.”
That sentence captures the strategic challenge facing governments around the world.
Moving too quickly with sensitive data can create privacy and security risks.
But refusing to develop responsible mechanisms for data access also carries consequences.
AI development will continue globally.
Countries capable of combining compute, talent, high-quality data and responsible governance may develop significant technological and economic advantages.
The challenge is therefore not simply choosing between sharing data and protecting data.
The more important question is:
Can we design systems that make data useful while keeping it protected?
That is the problem DEPA for AI Training is attempting to solve.
Final Thoughts: Data Could Become India’s Next AI Advantage
The AI race isn't only about who owns the most GPUs.
It is increasingly about who can combine:
Compute + Connectivity + High-Quality Data + Talent + Responsible Governance
India has enormous quantities of valuable data distributed across healthcare, finance, government and industry.
But data locked inside institutional silos cannot automatically become an AI advantage.
It needs mechanisms for secure access, clear governance, privacy protection and responsible use.
The proposed DEPA for AI Training framework represents an attempt to build that missing layer.
If implemented effectively, the biggest outcome may not simply be making more datasets available.
It could be creating an ecosystem where AI innovation and data protection can progress together.
And that may prove far more valuable than simply building bigger AI models.
Comments