Views of Abhishek Singh: “We are entering the era of agentic AI and physical AI”

India is entering a new phase of its digital journey, where artificial intelligence (AI) is being positioned as the next layer over digital public infrastructure (DPI) instruments such as Aadhaar, Unified Payments Interface (UPI) and DigiLocker. The government’s IndiaAI Mission is addressing key bottlenecks around compute, funding, Indian foundation models, AI-ready data sets, skills, and safe and trusted AI guardrails. There is also a clear push to make services easier to access through voice and Indian languages, and to use AI for real outcomes in areas such as agriculture, healthcare, education and governance. Abhishek Singh, IAS, Additional Secretary, Ministry of Electronics and Information Technology (MeitY), and Director General, National Informatics Centre (NIC), spoke about these priorities and the way forward at various industry events during the past year. Edited excerpts from some of his addresses…

 

India is among the fastest growing major economies. We are at the $4 trillion mark and we aspire to become a $30 trillion economy by 2047 to achieve a developed country status, what our prime minister calls “Viksit Bharat”. Technology, and now AI, can be a major enabler of this journey.

Our DPI, the India Stack, has been a key part of this progress. It includes Aadhaar for identity, UPI for digital payments at scale and DigiLocker for paperless exchange of documents at scale. I see AI as a way to turbocharge this DPI. With AI, we can take these services to the remaining 500 million people who are still not connected, many of them in rural areas. If services work through voice and Indian languages, they become far easier to access. A farmer can ask where to get fertilisers, what pesticide to use, or where to sell produce. People can ask for basic support in healthcare and education. Similarly, small businesses can use AI to optimise operations, improve efficiency and increase incomes.

Improving talent competitiveness

India has traditionally done well in cognitive skills and software services, and many of our professionals have built global careers. But those skills are now challenged by AI bots. We have to think about how we compete when bots can do tasks faster and at a lower cost. That is why we need to empower our software professionals and students with AI skills.

Leveraging the IndiaAI Mission

The government approved the IndiaAI Mission and consulted industry, academia and researchers to identify bottlenecks in adopting AI. Compute availability was a key constraint, with access earlier limited to around 600 graphics processing units (GPUs), which is not enough to train foundation models. Funding for start-ups was also a limitation. We needed Indian foundation models, AI-ready data sets and stronger guardrails for responsible, safe and ethical use.

The mission is designed around seven pillars – compute, foundation models, data sets, application development, future skills, safe and trusted AI, and start-up financing. We are also making compute available through a private sector-led incentive mechanism, with the government under writing a significant share of the cost. As a result, we have been able to empanel around 40,000 GPUs and make them available at a lower price of about Rs 65 per GPU per hour – below a dollar and lower than many global prices.

Building Indian foundation models

India needs its own foundation models trained in Indian languages and Indian context. If we rely only on global models, our data can go to servers outside India, and the results can also reflect bias if models are not trained on our culture, heritage and diverse inputs.

We are supporting Indian companies and institutions to build Indian foundation models, including initiatives to create large language models and smaller models for domains such as healthcare and material science. Alongside, we are building a data set platform called AI Kosh, which already hosts thousands of data sets.

Application development and safe AI

We are also supporting application development to solve societal problems in areas such as agriculture, healthcare, education and governance. On skills, we are funding projects and creating data labs to build capacity across the ecosystem.

Similarly, we are focusing on safe and trusted AI. We have set up an AI Safety Institute and are working on tools for bias mitigation, ethical AI certification, privacy
preservation, detection of deep fakes and watermarking. We are also supporting start-ups through financing structures, including a fund of funds to co-invest in these start-ups.

Widening access to AI

We are running a wide set of international and partner-led engagements to gather inputs for the India AI Impact Summit 2026. Alongside, we have planned regional group meetings and working group meetings across the country so that we involve every one in the planning and keep the focus on outcomes.

In terms of deliverables, we are working towards a charter that looks at how we can widen access to AI resources such as compute, data sets and algorithms for the wider world, and how we can build an AI commons repository. This repository can include open use cases that are made available to the larger global community, as well as reskilling principles, competency frameworks and courses. We are also working on a playbook for equitable workplace transition, a framework for capacity building in the public sector, an AI for science network, and AI impact awards that can become an annual initiative.

IP ownership

We must also grapple with copyright and intellectual property (IP) issues in AI. Some questions that arise include who owns the IP when a model is trained on publisher content, whether revenue sharing should happen and who carries the liability when something goes wrong. We need a balance that supports innovation while protecting original creators. At the same time, we have to address harmful issues such as misinformation and deep fakes, and think through regulation with this lens.

AI in legislatures and governance

In governance, the NIC has supported the National e-Vidhaan application, and almost 20 state assemblies have adopted it and become paperless. In Parliament, we have Digital Sansad, which is used actively by both houses. Translation across languages is a key need. Human translators play a role, but AI can make translation smoother, make documents available electronically, and support the analysis and summarisation of bills and papers.

The NIC has also developed a range of AI services that can be integrated into governance platforms. These include chatbots, speech-to-text, face verification, document image analytics, geospatial services, translation tools such as Matra and Panini, summarisation through Saranshi, masking of personally identified attributes through Nibhrith and services such as Anveshika and Ligra for generative AI (GenAI).

Identifying gaps

AI has been around for decades and it first started as a discipline in 1956. But after GenAI tools became widely available in late 2022, many people became AI users. As we move forward, we are entering the era of agentic AI and physical AI, and many even speak about artificial general intelligence.

India has strengths, including our engineering base and our ability to build large-scale products. But we also face gaps. Many data sets are not AI-ready because they do not follow metadata standards and are not API enabled. Compute is expensive and concentrated, and we need guardrails to manage bias and ensure ethical use.

Using AI in public services

We are deploying AI across public services. We are using tools for translation, including through Bhashini, which is being used by various departments and states. In e-Office, dictation can help officers create notes in multiple languages. In law and
justice, a legal research assistant can support officers by summarising large petitions and connecting key points with
relevant judgments.

For verification systems, face authentication is being used on platforms such as DigiYatra and Jeevan Pramaan. In citizen services, multilingual chatbots and document analytics can help people access services and understand information.

For practical city governance, AI can help identify potholes, garbage dumps, streetlights that are not working and even traffic violations, by using moving cameras and sending location-tagged alerts. In agriculture, voice-based advisory can help farmers.

Developing BharatGen

As we launch the BharatGen suite of applications, I see the work on Param 1, the ASR models, Indic models and data sets as a step towards building in Indian languages. The IndiaAI Mission is a whole-of-India approach, bringing together the government, industry, start-ups and researchers to catch up with the best in the world.

To build models and applications, we need talent, compute and data sets. We have talent, but many engineers moved to the West because India could not always offer compute, data sets and financial support. With national missions, this is changing and we are seeing more talented Indians returning to work on challenging projects.

We are expanding compute availability further and focusing on data sets across domains such as health, agriculture, education, manufacturing and logistics. When data sets are available, entrepreneurs can build models and applications, creating a productivity boom that makes people and businesses more efficient.