On 20 August 2026, the IndiaAI Mission held a workshop on AIKosh, the national datasets and models platform, at the India Habitat Centre, New Delhi, bringing together government, industry, academia and startups to discuss data governance, harmonisation and the responsible use of AI-ready data.
The framing came from Shri Saurabh Vijay, CEO of the IndiaAI Mission, speaking on "Data as National Wealth: India's Imperative for Sovereign AI", who said: "AIKosh is the platform India has provided for the ecosystem. If we are to match the best in the world, our collective ownership of this initiative is what will take it to that global standard."
The workshop was inaugurated by Dr Saurabh Garg, Secretary, Ministry of Statistics and Programme Implementation (MoSPI), who stated the technical problem exactly: "The defining challenge we are solving alongside the IndiaAI Mission is ensuring that data is discoverable, understandable, interoperable, and securely accessible across all ministries."
Those four words are the substance of the whole event, and worth unpacking before anything else.
Why data, and not chips or models
The intuitive assumption about AI capability is that it is decided by computing power or by model architecture. Both matter. But the constraint that binds a country like India is neither.
A model is a compressed representation of the data it was trained on. A system trained overwhelmingly on English-language, Western text will be fluent about the things that text describes and weak about everything else — Indian languages, Indian legal and administrative categories, Indian crop patterns, Indian disease profiles, Indian road conditions. No amount of computing power fixes a data gap; it only makes the gap more efficiently reproduced.
That is what "sovereign AI" means in this context, and it is a narrower and more defensible claim than the phrase sometimes suggests. It is not autarky. It is the position that a country needs models trained on its own data, running on infrastructure it can rely on, for applications where local specificity decides whether the output is useful or dangerous.
Now the four properties, which describe why national data is not automatically usable:
- Discoverable — you can find out that a dataset exists. Much government data is held by an agency that never publishes a catalogue.
- Understandable — the fields, units and collection method are documented. A column labelled "yield" is useless without knowing the unit, the year and the measurement protocol.
- Interoperable — datasets from different ministries can be joined, because they use compatible identifiers and formats. This is the hardest one, and it is why harmonisation was on the agenda.
- Securely accessible — available to legitimate users under controls that protect privacy and sensitive information.
India generates enormous administrative data. Very little of it satisfies all four conditions at once. AIKosh exists to close that gap — a national repository of datasets and models, so a startup or a researcher does not have to negotiate separately with each ministry.
Who was in the room, and why that matters
The participant list tells you what kind of problem this is being treated as.
Presentations came from Shri Mohammed Y. Safirulla K (Director, IndiaAI Mission), Shri Sudeep Shrivastava (COO), and jointly from Ms Shikha Dahiya (Joint Director) and Shri Swadeep Singh (General Manager, Data Science). The thought-leadership panel — on lawful data sharing, privacy-preserving use, and the case for and against sharing specific categories of data — included Google DeepMind, UIDAI, the World Bank, IIT Jodhpur, EkStep and CoRover.ai, moderated by the Gates Foundation.
Keynotes followed from Shri Nand Kumaram (CEO, National e-Governance Division) on unlocking data for AI in governance; Ms Tanusree Deb Barma (Deputy Director General, UIDAI) on scaling data unlock for digital public goods; and Shri Abhishek Upperwal (SoketAI) on the needs of the open data ecosystem.
The presence of UIDAI and the framing around digital public goods connects this to India's established model. Digital Public Infrastructure (DPI) — Aadhaar for identity, UPI for payments, DigiLocker and the account aggregator framework for documents and consent — is India's most exported policy idea. AIKosh applies the same template to data: build a public rail, let private innovation run on top of it. That is a genuinely coherent strategy and it is the point a strong answer should make.
Note also the deliberate inclusion of the case against sharing certain data categories. Not all government data should be open — health records, security-related data and personally identifiable information are governed by the Digital Personal Data Protection Act, 2023, which sets out consent, purpose limitation and the obligations of a data fiduciary. A workshop that lists "the case for and against" is acknowledging that the default is not "release everything."
The IndiaAI Mission in one frame
The IndiaAI Mission, approved in 2024 under MeitY, is structured around several pillars that are worth knowing as a set, because a question can ask which of them a given initiative belongs to:
- IndiaAI Compute Capacity — building GPU capacity accessible to Indian researchers and startups.
- IndiaAI Innovation Centre — indigenous foundation models, including for Indian languages.
- IndiaAI Datasets Platform — this is AIKosh.
- IndiaAI Application Development Initiative — AI applications for problems of public importance.
- IndiaAI FutureSkills — expanding AI education and training.
- IndiaAI Startup Financing — funding for deep-tech AI startups.
- Safe and Trusted AI — responsible-AI frameworks, governance and tooling.
Alongside it sits BHASHINI, the national language translation mission, which addresses the same underlying gap from the language side: Indian-language data is scarce in global training corpora, so it has to be built deliberately. For candidates, science and technology items of this kind sit within the wider NDA general knowledge preparation, and the physics of the compute layer connects back to core topics like work, energy and power when questions turn to data-centre energy demand.
🔑 Revision block
The event. 20 August 2026 — IndiaAI Mission workshop on AIKosh at the India Habitat Centre, New Delhi, with government, industry, academia and startups, on data governance, harmonisation and responsible use of AI-ready data.
The framing. "Data as National Wealth: India's Imperative for Sovereign AI" — Shri Saurabh Vijay, CEO, IndiaAI Mission. Inaugurated by Dr Saurabh Garg, Secretary, MoSPI.
The four properties of usable data. Discoverable · Understandable · Interoperable · Securely accessible. Interoperability across ministries is the hardest, and the reason harmonisation was on the agenda.
What AIKosh is. The national datasets and models platform under the IndiaAI Mission — the IndiaAI Datasets Platform pillar. A single repository so researchers and startups need not negotiate ministry by ministry.
Why data, not compute. A model is a compressed representation of its training data. Training on largely Western, English text leaves gaps in Indian languages, law, crops, disease profiles and road conditions that no amount of computing power fixes. That is the defensible meaning of sovereign AI — not autarky, but models trained on one's own data for applications where local specificity decides usefulness.
Who participated. IndiaAI Mission officials Safirulla K (Director), Sudeep Shrivastava (COO), Shikha Dahiya, Swadeep Singh. Panel: Google DeepMind, UIDAI, World Bank, IIT Jodhpur, EkStep, CoRover.ai, moderated by the Gates Foundation. Keynotes: NeGD, UIDAI, SoketAI.
The DPI link. India's Digital Public Infrastructure model — Aadhaar (identity), UPI (payments), DigiLocker and account aggregators (documents and consent). AIKosh applies the same template to data: a public rail with private innovation on top.
The limit. Not all data should be open. Health records, security data and personally identifiable information are governed by the Digital Personal Data Protection Act, 2023 — consent, purpose limitation, and the duties of a data fiduciary. The workshop explicitly discussed the case against sharing some categories.
The Mission's pillars. Compute Capacity · Innovation Centre (foundation models) · Datasets Platform (AIKosh) · Application Development · FutureSkills · Startup Financing · Safe and Trusted AI. Approved 2024 under MeitY. Companion mission: BHASHINI, for Indian-language translation.
🎯 Practice MCQs
Q1. AIKosh is best described as: (a) the national datasets and models platform under the IndiaAI Mission (b) a supercomputing facility (c) a language translation app (d) a digital payments system → (a).
Q2. The IndiaAI Mission functions under: (a) MeitY (b) MoSPI (c) the Department of Science and Technology (d) NITI Aayog → (a) — the Ministry of Electronics and Information Technology.
Q3. Which of the following is NOT among the four properties of AI-ready data cited at the workshop? (a) Monetisable (b) Discoverable (c) Interoperable (d) Securely accessible → (a) — the fourth is 'understandable'.
Q4. The workshop was inaugurated by the Secretary of: (a) MoSPI (b) MeitY (c) DPIIT (d) UIDAI → (a) — Dr Saurabh Garg.
Q5. India's national language translation mission is: (a) BHASHINI (b) AIKosh (c) DigiLocker (d) UMANG → (a).
Q6. Personal data protection in India is currently governed by the: (a) Digital Personal Data Protection Act, 2023 (b) IT Act, 2000 only (c) RTI Act, 2005 (d) Aadhaar Act, 2016 → (a) — with consent, purpose limitation and data-fiduciary obligations.
Q7. 'Sovereign AI', as used at the workshop, refers primarily to: (a) models trained on a country's own data and infrastructure (b) a ban on foreign AI models (c) military use of AI (d) state ownership of all AI companies → (a).
Q8. Which body's Deputy Director General spoke on scaling data unlock for digital public goods? (a) UIDAI (b) NeGD (c) CPCB (d) TRAI → (a).
Q9. India's Digital Public Infrastructure includes which of the following? (a) Aadhaar, UPI and DigiLocker (b) AIKosh, BHASHINI and PMJAY (c) GSTN, ICI and IIP (d) NavIC, GAGAN and IRNSS → (a).
Q10. The pillar of the IndiaAI Mission that AIKosh belongs to is: (a) IndiaAI Datasets Platform (b) IndiaAI Compute Capacity (c) IndiaAI FutureSkills (d) Safe and Trusted AI → (a).
Q11. The reason computing power alone cannot fix a data gap is that: (a) a model is a compressed representation of its training data (b) GPUs are expensive (c) data centres consume power (d) algorithms are proprietary → (a).
Q12. The IndiaAI Mission was approved in: (a) 2024 (b) 2020 (c) 2016 (d) 2026 → (a).
📋 How this gets asked (PYQ pattern)
Technology-policy items reach the NDA paper through four doors. The platform-matching item — AIKosh, BHASHINI, DigiLocker, UMANG and UPI matched to their functions, where AIKosh and BHASHINI are the newest and therefore the likeliest distractors for each other. The ministry item — which ministry runs a given mission, with MeitY, MoSPI and DST regularly confused. The legislation item — the Digital Personal Data Protection Act, 2023 and the vocabulary of consent, purpose limitation and data fiduciary. The DPI item — the components of India's digital public infrastructure, asked as a set.
The fresh 2026 hook is the AIKosh workshop and the phrase "Data as National Wealth", alongside the four data properties, which convert neatly into a statement-type question. The likeliest single item pairs the IndiaAI Mission's ministry with a description of AIKosh — both halves true. As always, this describes the recurring pattern, not any exact past question.
Preparing for the NDA? Technology questions are won on institutional mapping — which mission, which ministry, which Act — rather than on technical depth. Build that map with our NDA general knowledge notes, follow the daily NDA current affairs, and train with our ex-officer faculty in the upcoming Cavalier courses in Delhi.
✍️ Written by Col D.N. Sharma — Defence studies & technology faculty at The Cavalier. Reviewed by the Cavalier Faculty Desk. The Cavalier, founded by ex-Army officers, has trained NDA/CDS/SSB aspirants since 2001 (Facebook · YouTube).
Source: PIB / Ministry of Electronics & IT, 20 August 2026. Mission structure cross-verified with independent sources.