Can Sarvam justify its $53 mn backing and govt mandate in India’s AI Mission?

Sarvam.AI, the poster child of India’s sovereign AI mission, is facing criticism for its latest release—Sarvam-M. While the company claims strong Indian language capabilities, critics say it lacks originality. With $53 million in funding and government backing, many expected more. The launch has reopened questions around India’s AI roadmap.

Sarvam.AI, one of the most prominent names in India's emerging artificial intelligence ecosystem, has once again grabbed attention, though this time for reasons that raise more questions than acclaim.

The Bengaluru-based startup, which is among the key companies selected under the Centre’s IndiaAI Mission to develop the country’s first sovereign foundational model, recently launched Sarvam-M, a 24-billion-parameter open-weights large language model (LLM) built upon Mistral Small, an open-source LLM developed by French startup Mistral AI. This is the first release in the series where they plan to drop their regular findings.

While Sarvam touts Sarvam-M as a multilingual model tuned for Indian languages with capabilities in mathematical and programming reasoning, the response from the tech community has been mixed. Some have welcomed the initiative, but others have questioned its necessity, pointing out that similar fine-tuned models already exist in abundance.

“Sarvam is a well-funded company, connected with the government, and has access to GPUs. But instead of building a true model, they’ve simply fine-tuned an existing one. That’s not an ambitious move,” says Puneet Pandey, Founder and Chief Innovation Officer of Ojas Softech Pvt. Ltd., which runs AstroSage AI.

He adds, “Fine-tuning is now a commodity. There are hundreds of Indian companies—including ours—that have done it. So, what’s new here? Even the benchmarks they published claim marginal improvements over models like LlaMA and Claude, which no one is really using anymore.”

Sarvam.AI has raised $53 million in Series A funding from marquee investors such as Lightspeed and Peak XV Partners, making it one of the most well-capitalised deep tech startups in the country. Given this, many stakeholders expected a bigger leap from its latest announcement.

Industry insiders suggest the company may be facing pressure to publicly showcase progress, especially under the spotlight of the IndiaAI Mission. But despite its well-resourced backing, Sarvam-M has received underwhelming initial downloads, raising questions around adoption and real-world demand.

Yet, demand for truly Indian AI models is real, say some experts.

“There is a massive gap in existing LLMs when it comes to Indian languages. Models like OpenAI’s GPT or Google’s Gemini allocate just 1-3% of training data to Indian languages,” says Pandey. “These models think in English and translate into regional languages—they don’t ‘understand’ them natively.”

But not everyone is dismissive. Sridhar Vembu, Founder of Zoho and a long-time advocate for indigenous tech innovation, defended Sarvam’s efforts on social media, noting that “instant success is neither necessary nor sufficient for long-term impact.”

Jaspreet Bindra, Founder of AI & Beyond, shares a more nuanced view. “While Sarvam-M’s initial downloads may look low, let’s not ignore what they’ve achieved. Building a 24B parameter multilingual model that supports 10 Indian languages, in a country where deep tech is still maturing, is no small feat.”

He notes that Sarvam-M has performed well on uniquely Indian benchmarks such as answering competitive exam questions in Hindi with high accuracy. “What may be holding back adoption is limited tooling, lack of documentation, or even the absence of urgent India-specific use cases.”

Bindra suggests the path forward lies in developing sharper, more tangible use cases like AI for agriculture, rural governance, or healthcare. “That’s how you drive meaningful engagement and unlock real value,” he adds.

Sarvam.AI may be learning that in India's AI race, fine-tuning alone won't win the game. As the ecosystem matures and expectations from sovereign models grow, the bar for innovation is only getting higher. For Sarvam, the challenge now lies in moving beyond symbolic launches to building solutions that resonate with India’s grassroots needs and proving that it’s not just present in the AI revolution, but leading it.