The IndiaAI Mission enters its second year with the long-awaited promise that its first wave of sovereign models will finally arrive. Twelve firms are now at work on everything from bilingual LLMs to speech and vision systems. Sarvam is slated to release its model this month, with others likely to follow early next year.
On paper, this should be the moment the mission hits its stride. After a year of big announcements, hundreds of millions in funding, and tens of thousands of GPUs being mobilised, the stage seemed perfectly set.
Yet, a persistent question keeps resurfacing: even if these models arrive on schedule, will they truly matter for enterprise use in a market that has already standardised on OpenAI’s GPT, Anthropic’s Claude and Google’s Gemini?
Enterprises in India do not buy software out of patriotism. They buy whatever solves their problems. The first set of sovereign models is aimed at Indic languages and national datasets. Their value lies in cultural nuance and regulatory alignment, not frontier performance.
That is important for sovereignty. But it does not automatically make them competitive from an enterprise standpoint. Ankush Sabharwal, CEO of CoRover.ai, explained to AIM, “Enterprises don’t and will not just choose sovereign models because of patriotism, and currently there is no law which forces them to.”
“Enterprises care about solving their challenges with whatever it takes, being legal, which is the right thing to do first for profit-oriented businesses,” he added.
This is the tension at the heart of IndiaAI. The mission seeks to build sovereign capability. Indian enterprises want world-class tools. Today, the two are not always aligned.
What Happens Next?
Sabharwal has seen how far small, targeted models can go. “I strongly believe, and have witnessed, that with intelligent and grounded GenAI using RAG and enterprise-specific (small) models, many problems or use cases get solved,” he said.
Yet, he also argues that expecting a single sovereign model to match the breadth of global frontier systems is unrealistic today. “Enterprises want all the problems of all their consumers to be solved by one model, which cannot be provided currently by sovereign models including BharatGPT [CoRover’s AI model]. Difficult even to aim for that for now.”
This is why his platform lets users pick whichever model works best. He noted that BharatGen aims for a trillion-parameter model and added, “I wish we have one, eventually.”
The IIT Bombay-incubated AI firm BharatGen is aiming for the biggest model of all, designed as a mix of English and Indian languages. It maintains that sovereignty doesn’t mean shutting the door on global players. The team has also signed an agreement with IBM to collaborate on model technologies, data preparation, and scaling data prep work for complex, governed pipelines. This follows the team’s continued partnership with NVIDIA.
The same applies to IndiaAI as a whole, which has actively partnered with OpenAI for several recent announcements.
Indian founders say the gap is not talent but time, compute and long-term capital. “India is moving in the right direction, but we need to be honest: we are not yet competing at the frontier level of global AI models, but we absolutely can,” Ashutosh Singh, co-founder & CEO of RevRag.AI, said.
He attributes the gap to “compute access, deep research culture, and long-term capital”. The world’s leading labs have enjoyed years of compounding, while India is attempting to compress that timeline into a fraction.
Even so, he argues that sovereign Indian models have a distinct purpose. “The Indian models being built today are solving something unique: ground reality, multilinguality, regulatory alignment, affordability and enterprise-grade deployment.”
Enterprises, he says, care about adaptability and trust, not just size. “Models don’t win on parameter count, they win on adaptability and trust in production.” In his view, India can catch up if it continues to invest in open compute and research.
“We’ll see India produce globally competitive systems within the decade.” In the meantime, he predicts that Indian enterprises will adopt AI faster than most regions because the models emerging from India will be “fit for purpose”, aligned with regulation and cheaper to run.
Sovereign, Not Substandard
Still, this is not the whole story. India’s sharp turn towards Swadeshi tech has raised another concern. Some founders now fear that building Indian AI must not become an excuse for building lower-quality AI.
Apoorv Sood, global GTM head of Smallest.ai, told AIM, “If you’re building Swadeshi Tech, there’s nothing bad about it. But if you’re building Swadeshi Tech from a standpoint that, just because you’re local and have access to the market, you would like to throw away external competition and not build world-class products, I think you’re completely wrong.”
Real Swadeshi, he argues, must be global in ambition. His analogy is simple. If a city builds a metro, people expect the best possible system, not a cheaper local substitute.
“I feel the real Swadeshi tech is… built for the globe, keeping your own country in mind,” he said, citing Zoho as an example. “India should not settle for substandard tools in the name of nationalism.”
Meanwhile, Amartya Jha, co-founder of CodeAnt AI, agrees that mindset is a barrier. “Most Indians are risk-averse,” he said. He explains that, by now, US investors have lived through cycles from the mobile boom to the AI winter and know how to price risk.
Indian founders often must go abroad first for validation before Indian capital even looks their way. This cycle slows down product quality in India and forces founders to prove themselves overseas before being taken seriously at home.
Yet, Jha maintains that, with the right ambition, building from India for the world is well within reach. His company hires top engineers in India at global salaries and sells to Indian banks, power companies and major airlines, including Akasa Air. It now competes directly with established global players like SonarQube.
He believes the real shift will come when Indian media and investors start celebrating Indian founders who succeed globally, not just Indian products built for India alone.
The missing ingredient in India’s AI story has often been genuine frontier experience. Abhishek Upperwal, co-founder and CEO of Soket AI Labs, says India has talent, but very few people who have trained very large models before.
“If you ask me how many people in India have already built a one-trillion or maybe 100-plus-billion perimeter models already and actually have that experience—in India, there are probably not many,” he said, adding that this is unsurprising because the only labs with such experience—Google, Microsoft, Meta and OpenAI—work with small teams largely located outside India.
Now, these companies are eager to set up offices in India and seek out Indian customers.
His point is simple: India will not get frontier capability by waiting for these rare experts to return. It must learn by doing.
He also believes more startups should be supported under the mission. Redundancy is not a waste. It is how deep tech matures.
“When it comes to these hard technologies, you need redundancy. You don’t know what would work and what would not work.”
The Case for Quality
India’s sharp turn towards Swadeshi tech has brought the conversation to a decisive moment. Ganesh Gopalan, co-founder and CEO of Gnani.ai, argues that India’s language landscape is not a disadvantage but a doorway to a new class of AI systems.
“Most frontier AI models today are optimised for global languages like English, and even the newest systems such as DeepSeek are anchored in Mandarin as primary tokens,” he told AIM.
He believes that the next big leap will not come from generic models alone. As he puts it, “Generic LLMs are fantastic general problem solvers, but the real step-change in impact will come from models that are tuned to India’s context and its most critical use cases.”
His team is building foundation models that are grounded in India’s own requirements. They are, he says, “built ground-up for India, with a deep focus on speech, multimodal intelligence, secure on-prem deployments aligned with the IndiaAI mission and digital sovereignty goals.”
Gopalan explains that many enterprises already rely on their smaller, specialised models, and the right path is a blend of foundation models shaped for India and production-grade SLMs shaped for industry.
“This combination of India-first foundation models and production-grade, domain-specialised SLMs is what will allow Indian enterprises to create AI products that are globally competitive while being deeply rooted in the Indian context.” He adds that the goal is global.
The question of sovereignty itself remains delicate. Jaspreet Bindra, co-founder of AI & Beyond, frames it with clarity. “India’s push for AI sovereignty is both timely and strategic, but it must not come at the expense of quality or innovation,” he told AIM.
He warns against lowering the bar in the name of national pride. “Sovereignty should mean that India can build world-class AI systems, not merely local alternatives.” Early Indian models, he says, show promise where culturally relevant datasets matter, adding, “It would be unrealistic to say that we are already at global parity across all domains. And that is perfectly fine.”
What matters, according to him, is ambition. “The real win for India will be when sovereign AI and world-class performance become the same thing.”
Vikas Singh, chief growth officer at Turinton, draws a parallel with IT services. He says the industry did not grow by reinventing global tools. It did so by assembling them smartly and solving real enterprise pain. “Use what’s best available. If it’s open-source, use it. If it’s a global model, use it. If it’s homegrown, great. The question shouldn’t be ‘Is it Indian?’ The question should be ‘Does it solve the business problem better?’”
He believes India should build its advantage in platform architecture, not foundational models. As he says, “Sovereignty and quality innovation don’t have to conflict. But it requires pragmatism, not protectionism.”
This is where the IndiaAI Mission finds itself today. The models built under the programme will matter. They will raise national capability, deepen local research culture, and give India long-overdue compute access. They will also not replace GPT 5.1 or Claude or Gemini inside enterprise stacks in the near term.
These two truths can coexist.
IndiaAI must prove that sovereignty is not a label. It is a capability. The first wave of models only marks the starting line. They are not meant to beat frontier labs today, but to help India build the muscle memory to make better ones tomorrow.
The post Enterprises Won’t Choose Sovereign Models For Patriotism’s Sake appeared first on Analytics India Magazine.
