Sarvam AI has generated a lot of buzz around India’s sovereign AI ambitions, but the question many analysts ask is: is it too small to compete with global AI leaders? Training frontier models like Google’s Gemini, ChatGPT of OpenAI or Anthropic’s Claude, requires tens of thousands of GPUs and billions of dollars. Even companies like Meta and Microsoft spend massively on AI infrastructure – and they keep pushing what AI can do in reasoning, code, and text plus image work.
In contrast, Sarvam operates with far smaller capital and compute infrastructure. As of early 2026, this Bengaluru-based firm has raised a total of $53.8 million in funding, with key investors being Lightspeed, Peak XV Partners and Khosla Ventures. What happens if future funding slows, global models improve faster, or talent gets poached? Can it run mission-critical systems reliably and match or exceed global benchmarks at scale? What is its overall business viability?
“It is encouraging to see startups stepping forward with confidence and capability,” says Ashank Desai, chairman, Mastek. That said, sovereignty goes beyond innovation – it requires scale, resilience and long-term institutional strength. “A young company can be bold and visionary, but national digital backbones must be designed to withstand economic cycles, geopolitical shifts and rapid technological change. The ambition is admirable; the challenge is durability,” he adds.
Frugal Edge
Take for instance, two of its large language models (LLMs) – Sarvam-30B and Sarvam-105B – that are designed specifically for Indian languages and voice-first interaction. Herein, 30B and 105B refer to billions of parameters – essentially the model’s brain size. Models in this range broadly fall into the GPT-3 era class of model scale (not the latest frontier tier, but still significant). Training models at this level demands serious compute, high-quality data, and deep research talent.
AI infrastructure is capital intensive. Sustained access to compute and long-term funding are critical. Venture-backed firms operate within funding cycles, which can introduce uncertainty. Desai feels that running mission-critical systems for a country requires consistent uptime, security, and the ability to serve millions simultaneously. It also requires matching or exceeding global benchmarks. That is not impossible for a startup, but it demands partnerships, ecosystem support and institutional backing. “Sovereign AI should be a collaborative architecture, not dependent on a single entity.”
Sarvam AI represents a bold and commendable step towards India’s aspiration for sovereign AI, said Srividya Kannan, founder & CEO, Avaali Solutions. “However, achieving true sovereignty requires more than innovation – it demands scale, resilience and sustained economic muscle.” In her opinion, sovereign AI must be built on a foundation strong enough to withstand capital constraints, rising compute costs, and intensifying global competition. “Sovereignty cannot rest on isolated effort. It requires ‘Make in India’ to be embedded meaningfully into public procurement, with evaluation frameworks that favour domestic players and credibility at scale,” she adds.
Jaspreet Bindra, co-founder & CEO, AI&Beyond, stresses that sovereign AI is less about beating ChatGPT or Claude and more about control of national data and local language capability. Even smaller countries are pursuing this approach; France is backing Mistral AI and UAE is building models through Technology Innovation Institute. These players are tiny compared to US tech giants, yet, they are part of their countries’ sovereign AI strategies. “There is fragility in startups, but I think real innovation and good AI models can sometimes be better built by well-funded startups.”
Moving Beyond the Startup
But, here’s the real strategic angle: India has 900 million+ internet users and produces one of the world’s richest multilingual data streams across dozens of languages. Yet most global AI models still struggle with Indian languages, regional dialects and real Bharat use-cases. Therefore, if companies like Sarvam execute well, India could move from being just an AI market to becoming an AI builder for the world, especially for multilingual intelligence, Bindra emphasises.
“The pursuit of sovereign AI is a matter of national security and strategic autonomy rather than merely technical innovation,” says V Ramgopal Rao, Group vice chancellor, BITS Pilani and former director, IIT Delhi. “Sarvam AI has demonstrated that domestic talent can move beyond application layers to innovate at the foundational level, specifically by tailoring models for Indian languages and localised contexts. This is a critical first step that is required to be done.”
Frugal innovation has historically been, and continues to be, a significant edge for India. “This approach enables the creation of affordable, high-quality, and sustainable solutions that address unique, resource-constrained, and diverse needs, and this is what Sarvam has demonstrated,” says V Kamakoti, director, IIT Madras. “Its models have outperformed major global competitors like Gemini and ChatGPT on key benchmarks for Indian languages, document processing, and audio tasks. They address the unique challenges of the Indian market, such as voice-first interfaces for non-English speakers and multilingual capabilities across 22+ languages.”
Rao believes that true autonomy cannot be contingent on the survival of a few private entities or the volatility of global supply chains. “It requires a mission mode approach analogous to our successes in space and atomic energy, where the state anchors long term risk and infrastructure while the private sector drives agility. Without an institutionalised framework for massive GPU clusters and a non negotiable commitment to data sovereignty, our ambitions remain vulnerable,” he adds.