What if India did not need to chase giant global LLMs, but instead built smaller, sharper AI specialists tuned to its legal, financial, and cultural landscape? A sovereign platform built on open source intelligence and local computation is already rewriting expectations. This is the rethink powering India’s next AI leap…
India’s ambition to build a sovereign AI ecosystem is colliding with hard constraints: limited GPU access, shallow research capital, and a dependence on foreign hyperscalers. Few CEOs state this as bluntly as NxtGen’s A. S. Rajgopal, who says, “We don’t have the money to compete with the world. Risk capital is very little in India.” Yet, he argues, this shortage is forcing a different kind of innovation. One not dependent on replicating Western AI models but on designing India-specific alternatives.
The first problem, as he describes it, is structural dependence. Indian enterprises have become deeply tied to Amazon Web Services (AWS), Azure and global Artificial Intelligence (AI) Application Programming Interfaces (APIs), raising what he calls the risk of “digital colonisation.” His worry sharpened after experimenting with Google’s Takeout tool. “They knew places I didn’t even remember going,” he recalls, calling it a personal turning point in understanding the sovereignty challenge.
NxtGen’s answer is M, an AI platform built entirely around open source, India-trained models and GPU infrastructure physically located in India. “We can’t beat ChatGPT, you can’t beat free with free” A. S. Rajgopal, admits. Instead, the goal is to build domain-precise Indian agents rather than a universal model. One such agent is a legal AI built on all Indian laws and every court interpretation since 1969. The idea is that every citizen should know what the law actually entitles them to.
Another innovation is architectural: all models inside M run in an isolated, non-Internet-connected environment. “If I ask the model today’s date, it can’t tell,” A. S. Rajgopal notes. The design prevents models from both sending and accumulating external knowledge. An attempt to blunt risks associated with open source AI. The platform breaks tasks into tightly scoped domains to avoid hallucinations. An expert should only answer what that expert is supposed to answer, framing M as a mixture of specialists rather than a monolithic intelligence.
Even the infrastructure has pushed NxtGen into new territory. The company is deploying Rs 3600 crore(36 billion) worth of GPUs using liquid cooling, forcing them to build talent pipelines almost from zero. There is no institute teaching this. One has to learn abroad and bring it back. Looking ahead, A. S. Rajgopal, sees a looming power bottleneck as India’s AI ambitions scale. The company is in discussions to deploy a small modular nuclear reactor (SMR) by 2030 to directly power future data centers. “Once India crosses 10 gigawatts of AI load, we need a different source. The best is nuclear,” he says.
The broader message he wants India to absorb is patience and pragmatism. “If we try to build what the West built in the same way, we will fail. We need to get there with brains, not muscle”.
A fuller version of the conversation, expanding on the technologies, constraints, and strategic calls shaping this effort, will be released soon.
