As AI reshapes global power, India must build sovereign technological capabilities while navigating an increasingly fragmented and competitive geopolitical landscape.

On the 9th of June, 2026, Anthropic released Fable 5. This model represented a new level of capability: a ‘Mythos Class’ model with unprecedented intelligence. Upon release, it became massively popular overnight, with most users admitting that it blew away the competition. 3 days later, on the 12th of June, I woke up to the news that Fable had been abruptly disabled by Anthropic worldwide. The United States Government had intervened. The Department of Commerce had issued an emergency export-control directive restricting foreign national access for national-security reasons. Unable to filter nationality at such short notice, Fable was simply pulled entirely, with its users, primarily programmers, scientists, and engineers, left in the lurch.

This marked the first time that a government treated a model as dangerous enough to qualify it for export control, with the same mechanisms usually applied to weapons and aerospace technology. It’s unlikely to be the last. And it’s already not the only place that this technology is restricted. The AI stack is vulnerable in 3 layers as given below:-

1) Restriction of model APIs. This was the case with Fable 5. Frontier AI is intelligence on tap, and one possibility is that the US and China will simply stop making their most advanced models openly accessible. The rest of the world would be relegated to using AI that is 6 months behind the frontier, made accessible as an enormous revenue stream. But with AI, a gap such as that is an uncrossable chasm, guaranteed to grow due to exponential progress at the frontier.

2) Restrictions in AI infrastructure. This is what China is currently battling. AI training requires GPUs from Nvidia and AMD, and these GPUs are the most export-controlled objects on Earth. The US already ranks countries by tiers for exactly this, and India is categorised as tier-2: quantity caps and licensing requirements for any large-scale GPU deployment.

3) The third vulnerability is with embodied AI. China already dominates most robotics manufacturing. On its current path, India might well find itself reliant but unwilling to buy Chinese robots, or forced to buy much more expensive Western robots.

So when faced with a future where AI and its adjacent technologies are likely to be wielded as geopolitical weapons, what should India do? Here’s two answers that I don’t think are true strategic bets:-

1) Should India train its own Sovereign AI models? This mostly comes down to incentive. When faced with the daunting task of competing with American and Chinese frontier AI, Indian labs will prefer to address the non-competitive domestic market, post-training models to cater to ethnic languages rather than aim for frontier AI and AGI. And frankly, odds are that they’d be right to. Chasing an exponentially moving frontier from 2 years behind is a recipe to spend billions to stay 2 years behind. What India should keep is the expertise to post train and fine-tune open weights models.

2) Should India focus on building compute muscle? Large datacenter buildouts from Microsoft, Google, and Reliance are already underway, with ~$260B pledged at the India AI summit. The strategic move here is making sure a baseline quantity of that compute is revocation-proof and running open-weights models. The irony here is that the only models worth doing this for are the Chinese AIs, which Delhi will hate. But open weight models, regardless of origin, running on Indian soil are far more sovereign than American commercial AI APIs. Those can be switched off from Washington. We know, because we watched it happen.

Both of these answers only de-risk what India consumes. The third is different. It’s the only one where India could become something others depend on.

An Asymmetric bet – Build the full stack of embodied AI. Not all at once, and not end to end from day 1. Start by assembling robots with motors and components sourced from non-Chinese suppliers. As the products get good, swap parts out for your own: the higher level electronics first, companion computer baseboards, flight controller baseboards, then ESCs and sensor boards, then your own motors, with non-Chinese or even some Chinese magnets. An Indian-designed robot with a few Chinese magnets in it is much better than an entirely Chinese robot. Buy Japanese reducers. Import Jetson’s for the brains. Sovereignty here doesn’t require zero dependencies, it requires no single point of failure controlled by a possible adversary. Today this isn’t just theory, it’s what the Indian drone industry has been doing since Chinese imports were restricted: indigenising upward, board by board.

Why robotics? Because robotics hasn’t had its GPT moment yet. The ‘models’ that run robots today are small, fragmented, and built on open weights backbones, and post-trained. And when the GPT moment does come, it won’t be won by whoever has the most GPUs. It will be won by whoever has the data to feed the training pipelines, and the robot fleets to deploy the models on. Massive fleets of robots operating in the messy real world. India can’t out train San Francisco. But India can plausibly out-deploy everyone on Earth except China, and it’s the only country of its scale that the West trusts. This would still be an enormously difficult and complex undertaking, but being a strategy that could get us ahead of the curve for the first time makes it worthwhile. A massive deployed Indian fleet is a seat at the table when robot intelligence actually arrives.

If you’re cynical about this at all, you might ask me if any of this matters if superintelligence arrives. Maybe. If a singularity dissolves all physical constraints overnight, then this is all moot, and most, if not all of the world, would be at its mercy. Nothing addresses that future. But if superintelligence arrives and the world of atoms stays slow, where factories and power plants take years to build, then owning the physical layer has leverage. And if AI ends up improving step by step rather than all at once, then this definitely holds true as well.

So the Indian strategy has to move from an even hedge across all domains to a few well placed bets, and it has to move fast. The endgame is trade: robots and manufacturing capacity that the West needs, exchanged for the things we can’t make, advanced chips and guaranteed access to closed frontier models. Today, if another Fable gets pulled, India must hope. In the world where this strategy works, A model pull sees open-weights fallbacks already running on our own soil, and robots and manufacturing withheld.

India doesn’t need to win the intelligence race. Intelligence is about to need hands, and nobody has won the hands race (yet).