Pageviews:
Introduction
By early 2026, India had become one of the largest AI markets on earth. Reports put its weekly ChatGPT users at close to a hundred million, among the biggest user bases of any country. India had also spent two years rapidly expanding its own computing power, assembling nearly four times as many high performance GPUs as the government’s own IndiaAI Mission had originally set out to acquire.
Then, in April 2026, Anthropic revealed a new model called Claude Mythos, capable of finding dangerous software vulnerabilities faster than almost any human expert alive. Rather than release it broadly, Anthropic launched Project Glasswing, giving early access to the model to roughly forty organizations built to defend the world’s most critical software, among them major cloud providers, chipmakers, banks, and open source foundations. According to an analysis by the Carnegie Endowment, no Indian organization was part of that initial circle, despite India’s size and its fast growing compute base.
That is a strange outcome for a country that had just spent two years building exactly the kind of computing muscle everyone insists AI power requires. It suggests that having a lot of something, users, GPUs, market size, is not the same as having a say in who gets access to the technology that matters most when it matters most. The rest of this story is about why.
When a Tiny Part Stops a Giant Factory
In 2021, some of the world’s largest carmakers, companies each worth tens of billions of dollars, had to shut down assembly lines and cancel entire production runs. The cause was not a shortage of steel, workers, or capital. It was a shortage of small, specialized computer chips, some worth only a few dollars each, that only a handful of factories on earth knew how to make. Ford, Toyota, and General Motors could not simply place a bigger order and fix the problem, because the few foundries that made those chips had no spare capacity to sell, no matter how large the customer waving a checkbook happened to be.
That is the essence of leverage inside a supply chain. Being a big buyer gives you influence over price when many sellers are competing for your business. It gives you far less when only a handful of sellers can make the thing you need, and they know it. The carmakers had scale. The chip foundries had scarcity. Scarcity won.
The Company That Made Itself Impossible to Route Around
ASML understood this logic before most of the industry did. In the 1990s, it was a distant third place lithography maker, smaller than Nikon or Canon, and dependent on the German optics company Zeiss for lenses it could not build on its own. Rather than compete on price, it partnered with Zeiss on new lens designs, then organized a research consortium with European partners to solve a lithography problem most of the industry considered impractical, using extreme ultraviolet light to print microscopic circuit patterns onto silicon. Intel eventually backed the effort, partly because it wanted more than one supplier for something so critical, but ASML’s technology proved impossible for competitors to match once it worked.
Today, ASML is the only company on earth that makes machines capable of printing the most advanced computer chips, and every major chipmaker, including firms with far larger overall revenues, has no choice but to come to it. ASML did not need to become the biggest company in its industry to get there. It became the one nobody could work around, and that is a very different kind of power than simply having the biggest budget.
Two Indian Missions, Two Different Bets
India’s own recent history offers two experiments in this same question, and they point in different directions. In 2021, the government launched the India Semiconductor Mission, aiming to get chip factories physically built on Indian soil. That effort is still developing rather than finished. Micron’s assembly and testing plant in Gujarat and Tata’s packaging facility in Assam are approaching commercial scale, and the Tata led fab in Dholera has signed an early agreement with ASML to bring some of its lithography equipment into India. None of this yet gives India anything close to ASML’s leverage. But each step moves India further into a part of the supply chain that is genuinely hard to replace.
The India AI Mission tells a different kind of story. Its early target was ten thousand GPUs. By late 2025, India had assembled closer to thirty eight thousand, comfortably beating that goal. That is a real achievement, and it gives Indian researchers, startups, and companies far more computing power to work with than they had before. The mission was never only about buying hardware either, part of its funding also went toward helping Indian teams build their own foundation models rather than simply license someone else’s. But the GPUs themselves were designed and manufactured elsewhere, running on technology India does not control. Access to compute is not the same as ownership of the layer that makes compute possible in the first place, and the Claude Mythos episode is a reminder of what that gap can cost.
Final Thoughts
None of this means India was wrong to buy GPUs, or that building AI applications on top of foreign infrastructure is a mistake. Both are reasonable ways to participate in a fast moving industry, and both are already paying off. But the two missions point to a difference worth sitting with. One expands what India can access. The other, slowly and unglamorously, is trying to build something the rest of the world cannot easily get anywhere else.
That same distinction, between access and control, is echoed in who stood inside the circle when Anthropic handed out access to Claude Mythos, and it will keep shaping who gets a say as AI’s supply chain grows more contested. The real question for India is not how many GPUs it can buy next year, or how many people use its AI models. It is whether there is some material, process, or capability in this stack that the rest of the world will eventually need badly enough that it cannot simply route around India.