Today is an important milestone for Indian AI. Sarvam is launching the country's first sovereign AI foundation model. It is built on sovereign data, trained from scratch, optimized for India's needs, and served locally at a price-point that could support population-scale usage. I wanted to put down a few thoughts on why Lightspeed - an investor in other foundational AI companies such as Anthropic, xAI, Mistral - invested in Sarvam, and what is Sarvam's strategy for India. It took all of us at Sarvam a few years to both understand and then build the various pieces we needed to get right to service India's very unique enterprise & sovereign use cases. While we were on this quest, the world at large has had some genuine questions, and misunderstandings, about what we are trying to do, why we need to exist, how we are different from other global LLM companies who will, of course, enter India, and more. It is time to answer those questions. 

The most important question at hand is why even attempt to build a sovereign AI company. Our journey with Sarvam began in 2023 when we met Vivek & Pratyush. We were, by then, already seeing the AI supply chain fragment geopolitically, starting from the silicon layer, to LLMs, to data, talent and even applications (e.g. tiktok ban in US/India). As global geopolitical alignments shifted, the United States implemented export controls limiting access to advanced semiconductor technologies for specific countries.  We expected that to extend to models, people, power (especially nuclear) and more. AI was seen not just as a technology but as a natural resource that nations wanted to control. Similar to how India built sovereign capabilities in nuclear, parts of defense, and space, we needed to take the necessary steps to secure access to AI.  Sarvam founders had the unique combination of being technically the top 1% across the world and 100% committed to building for India and dedicating their lives for it. We at Lightspeed led Sarvam’s Seed & Series A rounds. 

Building a sovereign model company isn’t enough to build a large, profitable company. Model layers are hard to build and even harder to defend from commoditization risk. We started with the sovereign layer because it does offer some protection especially when our technical stack is globally competitive. Over time, Sarvam built out India’s “intelligence infrastructure”: models, APIs, agentic platforms, and products to make India AI-capable at population scale for citizens, for enterprises, and for the state. Our tech and product is simple: we will always strive to be SOTA in voice, text, image/OCRs, and anything that has anything to do with India & Indian use-cases. 

We believe we have a unique right to win in the markets and segments we operate in. Besides product & technical superiority for India-centered use-cases, we plan to have the strongest ground-game in town. The Indian government is never going to use a Chinese DeepSeek, nor would they use an American OpenAI. They could use an Indian OSS model but a technically world-class team is needed to build, run and manage end-to-end large-scale AI solutions, hardware up. Similarly, the Indian Enterprise segment cannot be served by throwing code over the border or API access alone; they need hand holding. 

On the state/sovereign side, Sarvam will run a two-pronged strategy. The first leg starts with the Prime Minister’s office down and aligns to our sovereign use-cases across defense, central government, national priorities in health, education, and more. The second leg starts with the state government up and enables each state to independently adopt AI for their citizens and institutions  (“Sarvam x States”). On the Enterprise side, our approach will be to “white-glove” them into the AI era through top-down board-driven transformations. These are large contracts that span 2-5+yrs and run in the $1-10 millions per year. 

Building these playbooks needs deep partnership with large GPU providers, capital providers, and full stack solutioning that needs a ground-up organization most young startups aren’t built for. Sarvam’s commercial strategy is uniquely tuned to building all of this, and for this part of the world.  At scale, Sarvam would look like part OpenAI, part Palantir, and part new-age AI-transformation consulting firm. We know how to go from hardware up, to training LLMs, to building APIs, to full end-to-end vertical solutions. We are building air-gapped datacenters one day, pre- or post-training a model on another, and scaling our voice API infra the third. Our FDEs are deployed at the national border, in largest private banks in Mumbai, in the political hub in Delhi, and conglomerates across India. 

We aren’t building Sarvam just because India needs its own AI stack. India is a massive, open market with tens of billions of dollars in addressable spend across software, compute, services, and productivity. The real bottleneck is access. The demographic dividend only converts if 900 million non‑English-first Indians can be skilled, educated, and made productive, and in India access runs through language. When AI understands every Indian language natively, skilling, healthcare, financial inclusion, and governance can scale from the English-speaking few to 1.4 billion. Jio made data ubiquitous; we intend to make intelligence ubiquitous. Sarvam means “all”. We plan to build it all, build with all, build for all. Our impact needs to be at the GDP-level, and we intend to have an impact at P&L level for every global and local company that serves India at scale.

This is our work. This is our wager. This is our life’s mission. That is what Sarvam is here to build. Thank you all for reading, for your support, and for your feedback.