Skip to content
Best Place To BuildBest Place To Build

Pratyush Kumar, Co-founder, Sarvam AI | "Sarvam means everybody- AI should be for everyone."| Ep. 24

In this conversation with Pratyush Kumar, co-founder and CEO of Sarvam AI, we dive deep into India's AI revolution. As the leader of one of India's hottest AI startups focused on solving Indian language challenges, Pratyush shares insights on building sovereign AI capabilities, the multi-layered approach to AI development, and the vision of making technology accessible to every Indian. Recently selected by the Government of India to build the country's sovereign language model under the India AI mission, Sarvam AI represents the intersection of technological innovation and national strategy. This episode offers a compelling look at how homegrown AI is positioning India at the forefront of the global AI landscape, with valuable lessons for aspiring technologists and entrepreneurs. 00:00:00 - Introduction and Background 00:03:04 - The Birth of AI4Bharat 00:05:07 - How IIT students helped build foundational components 00:08:00 - Birth of Sarvam AI 00:14:51 - The Four Layers of AI Development 00:21:33 - Real-World Applications of AI in India 00:26:04 - Strategic Autonomy in Technology 00:28:43 - Sovereign AI: India's Approach 00:34:40 - AI as a Utility for Everyone 00:38:54 - Technology as an Equalizer 00:41:40 - The Economics of AI Development 00:40:57 - Cost Structure of AI Business 00:42:58 - The Value Loop & Long-term Vision 00:45:05 - Market Dynamics & Competition 00:46:15 - Managing Fast-Paced Growth & Focus 00:47:47 - Indian AI Ecosystem & Academic Integration 00:51:28 - Talent Pipeline & Educational Infrastructure 00:53:22 - National AI Landscape & Government Engagement 00:55:07 - Work-Life Balance & Personal Fulfillment in AI 00:57:20 - AI Integration in Daily Work & Workflows 00:59:06 - Human-AI Relationship & Philosophical Implications 01:03:12 - Sarvam's Roadmap & Closing Thoughts

Pratyush Kumarguest
May 22, 20251h 4mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Building sovereign AI for India: language, scale, autonomy, utility access

  1. Sarvam AI grew out of IIT Madras’ AI4Bharat efforts to build high-quality Indian-language AI using data, compute, and community-driven research.
  2. Kumar frames “sovereign AI” as strategic autonomy: the capability to build and deploy core AI technology domestically without isolation from global collaboration.
  3. He outlines a four-layer full-stack approach—inference, models, orchestration, and applications—arguing that India needs all layers to make AI reliable, low-cost, and scalable.
  4. Sarvam’s work emphasizes India-specific language challenges (low-resource languages, culture tokens, Romanization, and code-mixing) to make AI usable for “everybody.”
  5. He argues AI should become a national-scale utility (like UPI), where per-capita AI usage could become a proxy for national productivity and competitiveness. જોકે

IDEAS WORTH REMEMBERING

5 ideas

Indian-language AI requires more than translation; it needs cultural and usage realism.

Kumar highlights not just language tokens, but “culture tokens” from undigitized material and evolving forms like Romanized Hindi and code-mixed text, which must be represented for models to work for everyday Indians.

Sovereign AI is primarily about capability, not isolation.

He defines sovereignty as the ability to build strategic tech “from scratch” domestically while still collaborating globally, giving India leverage and resilience in critical sectors.

Full-stack execution is essential to make AI affordable and reliable at national scale.

Sarvam splits the stack into inference efficiency, model training, orchestration (systems + workflows), and domain applications—because real deployments require low latency, telemetry, reliability, and scalable operations.

Deployments create a fast compounding “value loop” that should remain local.

Unlike slower hardware iteration cycles, AI can improve in months based on usage feedback; Kumar argues India must keep this loop in-country so learnings, data, and economic upside reinforce domestic capability.

AI in India can follow a UPI-like public–private scaling path.

Instead of only US-style big-tech scale or China-style heavy state control, he suggests India can catalyze compute and standards while enabling private innovation on top—making AI a low-cost utility.

WORDS WORTH SAVING

5 quotes

“Sarvam in Sanskrit means everybody, everyone, because the intention is that it should be used by everyone.”

— Pratyush Kumar

“You should have the ability to build it yourself… happy to collaborate with whoever in the world… but you should have the ability to build it yourself.”

— Pratyush Kumar

“I think AI could start looking like [electricity consumption] soon… your per capita consumption of AI is a decent proxy for how advanced or competitive you are as a country.”

— Pratyush Kumar

“We see ourselves as a full stack company… we see it as four layers.”

— Pratyush Kumar

“In the basement of Aadhaar, we have a set of boxes… GPUs… which contain… models and the orchestration layer… to deal with… calls… when biometric fails.”

— Pratyush Kumar

AI4Bharat origin and evolutionIndian-language data scarcity and “culture tokens”Code-mixing and Romanized Indian-language inputFour-layer AI stack: inference, models, orchestration, appsSovereign AI and strategic autonomyAI as a public utility akin to UPIGPU/compute economics and the rapid “value loop”Real deployments: Aadhaar voice workflows, insurance outreach, courts, NITI Aayog analyticsEcosystem building: academia–startup–VC integrationHuman–AI dependence and philosophical implications

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome