No country is fully sovereign in AI: Maya Sherman

Sovereign artificial intelligence (AI) is largely a function of geopolitics, with elements of technology thrown in, says Maya Sherman, Co-Lead of the AI literacy project of the Global Partnership on Artificial Intelligence (GPAI) under the Organisation for Economic Co-operation and Development. Sherman served as the Israel government's former Innovation Attaché in India. Her work, which spans discussions on AI policy to end users of the applications, gives her a hands-on understanding of how AI is transforming societies and livelihoods. Excerpts from an interview:

The understanding of what is sovereign AI is dynamic: there is no global consensus on its definition. How would you best describe it in the current context?

I would call it the capability to run AI infrastructure and the AI stack overall, usually including an overview of the AI value chain as a whole: each nation operating without any dependence on other factors or actors. Of course, very complex, different definitions are happening today. Importantly, we are seeing a shift in how we used to define excellence in this space.

So today, to define excellence in the AI space, we're speaking about nations' ability to have AI sovereignty. Previously, we evaluated nations based on their AI services and applications, and on how they were transforming their public sectors.

The shift is geopolitical. We're seeing an intensification over the past few months following the AI Impact Summit in New Delhi. However, it was the launch of the Chinese DeepSeek in 2023 that made countries sit up and think: 'Oh, we should have our indigenous models, too.' This realisation accelerated the discussion about AI infrastructure and stacks, moving beyond products and services.

Beyond the bluster, how many countries have, or can have, completely sovereign AI systems?

We have yet to see a nation that is fully sovereign in AI. The Brookings Institution notes that in the AI era, what nations do is manage dependencies. This is very interesting. All of a sudden, from speaking about full sovereignty, nations are shifting their thinking about sharing resources — in areas like storage, technology, or access to materials.

There are pragmatic reasons pushing nations in this direction. For instance, data centres need vast tracts of real estate. Several U.S. States are opposed to large data centres on their land, due to concerns about high water consumption, noise, and other issues. So not every State or nation will trade off land for AI supremacy.

The other example is of Meta, a huge private company that is building AI. It recently began limiting tokens (the basic unit of data that an AI processes) issued to employees to run the models. Running AI [infrastructure] is incurring huge expenses, and Meta has begun cutting costs. This is relevant because we are talking about a Big Tech company that does not know how to handle the costs of models it potentially created.

Even big nations and companies are figuring out that they have limitations. Going back to the question, we have yet to see a nation that is able to do everything together.

We'll probably see more tech blocs coming up around these domains; different nations will specialise in providing different elements for an AI system. Pax Silica, launched by the United States in 2025, is one such. Even (the) BRICS (bloc) is coming together for AI. The European Union recently proposed a new definition of AI sovereignty and its implications within its own territory. Europe, incidentally, is an entire continent, not one nation. It will be interesting to see how Europe will navigate between its regulatory supremacy and its ambition to become AI-sovereign.

AI sovereignty has become a talking point in India, spurred by both the desire to insulate itself from access denial and also to develop models that are inclusive of its immense diversity. How would you view India's efforts?

As a foreign expert who has lived in India, I feel that the discussions around AI infrastructure and stacks are not new there. From the beginning, India spoke of AI in its infrastructural meaning.

India has a large and diverse population, so discussions on how to run an AI infrastructure that involves the community, on investments, and on understanding energy requirements have always been present there. There is no significant thematic change compared to other nations whose views have recently changed. Also, India, I feel, has always tried to work on its own market independently, meaning that India is creating for India.

It is important for India to work on minimising gaps in the socio-economic fabric and connections with rural communities. Many core non-technical issues can pose obstacles to sovereignty.

India has launched its AI Mission. However, it should be ready to anticipate that locals may not agree to issues like setting up data centres in their region, as we are seeing in the U.S. There can be a clash between the need for AI infrastructure and AI sovereignty as India expands its footprint.

How effective do you think tech blocs can be?

At present, the blocs themselves are evolving, and it is hard to say what they will actually give. The idea is to connect like-minded nations. India could benefit if some of its issues could be resolved more quickly by allied companies. Similarly, the bloc could serve as a conduit for India to take its products, such as BHASHINI, to foreign markets.

Countries are naturally wary of how much they can collaborate. The ecosystem is still evolving. Pax Silica was formed only a few months ago.

BRICS is a very interesting bloc, not just touching AI, of course. It talks about tech co-operation. Yet, it has China as a member – a country because of which other nations are discussing sovereignty.

Announcements like the one by U.S. President Donald Trump about limiting access of other countries and nationals to new products being developed by the AI tech company Anthropic trigger the fear of being left out, don't they?

It is a fascinating case. How much can a President actually limit the work of a fully private company? It shows that sovereignty in AI has also become a tool of control for governments. Data, even more so AI data and infra, are suddenly political assets for governments. Let us see how this evolves. Anthropic is still a private company. I believe we will see more nations locking in battles against big tech companies.

Apart from sovereignty, could you elaborate on the priorities that each nation puts on advancing its AI landscape?

Europe is clearly focusing a lot on AI regulation. India brings the perspective of developing Digital Public Infrastructure and an open-source approach to AI. The U.S. is strong on Big Tech and does minimal regulation. China doesn't usually announce what it is doing till it has something like a DeepSeek moment.

Israel is bringing in Deep Tech start-ups. As a small nation, we know how to operate very fast. But it takes time to build compute power; we don't have large models yet. Our approach is similar to that of the U.S., with less regulation and strong investments, and is led by the private sector.

How do you see sovereign AI evolving with time?

Sovereign AI is, at its core, geopolitics, with elements of technology. The thinking about it will change as geopolitics change. During times of peace, when countries have like-minded partners and the global chain operates efficiently, the aspect of sovereignty in a given area becomes less important. During wars, nations become insular and think of doing everything themselves. But when a crisis becomes bigger than individual nations, they come together again. We saw such collaboration during the COVID-19 pandemic. We are seeing it now during the energy crisis, as nations get together to find alternatives. On the other hand, though the climate crisis is looming large, we do not have countries sitting together to find immediate technological solutions. It is because many nations still do not feel they are impacted by climate change.

When it comes to AI, there is no doubt that countries are realising they cannot work alone; they need like-minded partners. AI discussions for such connections are intensifying.

See also:

Made-for-India AI
AI for India
More for less