In a historic step, the Indian Government has chosen Bengaluru start-up Sarvam AI to create the country’s first native Large Language Model (LLM), a vital step towards the goal of technology self-reliance in the rapidly changing field of artificial intelligence. Chosen from among 67 contenders under the Rs 10,371 crore IndiaAI Mission, Sarvam AI represents a new wave of Indian entrepreneurship geared toward strategic digital sovereignty.

Announced by IT Minister Ashwini Vaishnaw, Sarvam AI’s selection marks a historic first for India and underscores a broader ambition to transition from being a passive consumer of AI technologies to an active co-creator.

The Vision: Local Innovation for a Global Future

Founded in July 2023 by Vivek Raghavan and Pratyush Kumar, both former leaders at the AI4Bharat initiative backed by Infosys co-founder Nandan Nilekani, Sarvam AI brings a uniquely Indian approach to artificial intelligence. The start-up focuses on full-stack generative AI offerings, ranging from fundamental research in model training to creating enterprise-grade platforms for authoring and deployment.

The company’s proposed LLM, designed to be fluent in Indian languages and optimized for voice interactions, aims to serve not just private enterprises but public-good applications at a population scale. “Our goal is to build multi-modal, multi-scale foundation models from scratch. When we do, a universe of applications unfolds,” said Raghavan.

Building Blocks: Scale, Scope, and Strategy

Sarvam AI’s flagship model will feature an architecture of 70 billion parameters, placing it in the league of global competitors. It will be built entirely in India, leveraging local infrastructure and homegrown talent. The government has committed critical resources to the effort, including access to 4,000 high-end GPUs over six months for training, sourced from data center players like Jio Platforms, Yotta, and Tata Communications.

The project will produce three distinct model variants:

• Sarvam-Large: Designed for complex reasoning and generation tasks.
• Sarvam-Small: Tailored for real-time interactive applications.
• Sarvam-Edge: Optimized for compact, on-device deployments.

Importantly, Sarvam’s models will not be fully open-sourced; instead, they will be fine-tuned to serve the linguistic and cultural nuances of India’s diverse population.

Competitive Edge: Learning from DeepSeek

The Indian government’s urgency is also concerned with international developments, especially China’s DeepSeek, an open-source LLM that shook international AI markets with high effectiveness at low cost. DeepSeek’s model, constructed with limited resources but producing stunning outcomes, stunned the tech world and reconsidered AI strategies globally.

Drawing lessons from DeepSeek’s success, India is betting that Sarvam’s indigenous LLM developed with strategic innovations in programming and engineering can achieve global competitiveness without the heavy resource demands of Western models.

Funding and Ecosystem Support

The scale of the task is enormous. However, Sarvam AI is well-capitalized to meet the challenge. In December 2023, the start-up raised $41 million in a Series A round led by Lightspeed Ventures, with participation from Peak XV Partners and Khosla Ventures.

Additionally, under the IndiaAI Mission’s broader vision, the government plans to onboard two to three more start-ups to build complementary AI capabilities. This ecosystemic approach ensures that Sarvam AI will not operate in isolation but will be part of a broader national strategy to create critical AI infrastructure.

Challenges Ahead

Despite the optimism, building a foundational model from scratch is fraught with challenges. Training a 70-billion-parameter model demands not just hardware but also deep expertise in model alignment, ethical safeguards, and data stewardship. Moreover, ensuring that the models serve Indian languages which often lack the vast datasets available for English—is a major technical hurdle.

Another critical point is the model’s commercialization strategy. Since Sarvam AI’s LLM will not be entirely open-source, ensuring widespread accessibility while maintaining commercial viability will be a delicate balancing act.

We Can’t Be Just Users

The implications of Sarvam AI’s work extend beyond technology. At a geopolitical level, India’s success in AI is crucial for maintaining digital sovereignty amid increasing global competition. Domestically, indigenous AI models promise to revolutionize sectors like healthcare, education, agriculture, and governance, especially when built with Indian languages and cultural contexts in mind.

As Minister Vaishnaw put it, “We can’t just be users of AI. We have to be its co-creators”. Sarvam AI’s selection marks a critical step toward fulfilling that vision positioning India not just as a participant, but as a leader in the global AI revolution.

Also Read: Sarvam AI Asks: What Good Is AI If It Doesn’t Speak Your Language?