Dr. Pratyush Kumar

By IndianAI.in | Indian AI Founders Series | July 2026

Estimated reading time: 18 minutes

Quick Facts — Dr. Pratyush Kumar

Introduction: The Weight of a Billion Voices

There is a particular kind of silence that falls when a technology, however powerful, cannot hear you. Not because you are quiet, but because it was never trained to understand the way you speak, the language you think in, the script your grandmother wrote in, the accent your district gave you.

For over a billion Indians, that silence has been the default setting of the global artificial intelligence revolution. The world's most powerful AI systems — built largely on English-language data, trained in California, designed for users comfortable with Roman script and Western cultural context — have arrived in India as useful but incomplete tools. They can tell a Mumbai investment banker what the weather will be in New York. They struggle, with embarrassing frequency, to transcribe a sentence spoken in Tamil, translate a land record written in Devanagari, or process a health insurance document filled out in Bengali.

Dr. Pratyush Kumar decided, quite deliberately, that this was unacceptable.

Not in the way a press release might say it — not as a mission statement drafted for investors. He decided it the way a researcher decides things: systematically, with evidence, with a clear-eyed understanding of the technical gap and a plan to close it. Over the course of more than fifteen years of academic and industrial research — at IIT Bombay, ETH Zurich, IBM Research, Microsoft Research, and IIT Madras — he built the technical foundations for what would eventually become Sarvam AI, the Bengaluru-based startup that is now India's newest AI unicorn, valued at $1.5 billion following a $234 million Series B round led by HCLTech in June 2026.

The story of how he got there is the story of a scientist who never stopped thinking like one, even when the world around him started moving at startup speed.

The Making of a Mind: IIT Bombay and the Path to ETH Zurich

Pratyush Kumar grew up in India, and from an early age, he was drawn to the kind of problems that require both mathematical precision and an understanding of how systems — technical, social, linguistic — actually work in the real world. His academic journey began at the Indian Institute of Technology Bombay, one of India's most rigorous undergraduate institutions, where he completed a Bachelor of Technology in Electrical Engineering.

<cite index="60-1">He holds a B.Tech in electrical engineering from the prestigious Indian Institute of Technology (IIT) Bombay and earned his Ph.D. in computer engineering from ETH Zurich in Switzerland.</cite>

ETH Zurich is one of the world's foremost technical universities — the institution that counts Albert Einstein among its alumni, that has produced more than two dozen Nobel laureates, and that consistently ranks among the top five engineering universities globally. For an Indian engineer to complete a doctorate there is not simply an academic credential. It is an indicator of a particular kind of intellectual ambition: the willingness to situate oneself in the most demanding research environment available and to compete at a global level.

<cite index="62-1">Kumar went on to earn a PhD in Computer Engineering from ETH Zurich in 2014.</cite>

His doctoral work in computer engineering at ETH Zurich gave him a foundation in systems thinking — how large, complex computational architectures are designed, how they fail, and how they can be made robust at scale. This is not the same as building a machine learning model. It is about understanding the infrastructure that those models run on: the memory hierarchies, the data pipelines, the distributed systems that make large-scale computation possible in the first place. That depth would prove to be exactly what India's AI moment required.

The Research Years: IBM, Microsoft, and the Slow Build of a Vision

When Pratyush Kumar returned from ETH Zurich, he did not immediately start a company. He did something harder in some ways: he kept doing science.

<cite index="62-1">He joined IBM Research in 2015 as a research scientist, working on advanced AI and computing projects, and later spent two years at Microsoft Research. Between 2018 and 2021, he served as faculty at IIT Madras and continued to remain associated as adjunct faculty.</cite>

IBM Research is one of the world's oldest and most respected industrial research laboratories. It has produced more Nobel laureates than most countries. Working there as a research scientist in AI and systems engineering is not a stepping stone — it is a serious research career in its own right. The same is true of Microsoft Research, which in the late 2010s was among the most important places in the world to work on large-scale AI systems, parallel computing, and the infrastructural challenges of machine learning at enterprise scale.

What Pratyush was building during these years was not a product. He was building a point of view — a deep, technically grounded understanding of what it would take to build AI that actually worked for languages and contexts outside the English-speaking world. He saw, from inside two of the world's most sophisticated AI research environments, how much of the global AI effort was concentrated on problems defined by Western needs and Western data. He also saw, with increasing clarity, what the implications of that concentration would be for India.

The parallel track he was running — as adjunct faculty at IIT Madras — kept him embedded in India's academic ecosystem even as he was working inside global corporate research labs. IIT Madras is not simply a teaching institution. Its research park and incubation ecosystem have produced some of India's most consequential technology ventures, and its faculty connections reach into every major AI research group in the country. Pratyush's dual life as a researcher and an educator gave him a network and a perspective that purely commercial founders rarely develop.

AI4Bharat: The Research Infrastructure India Needed

Before Sarvam AI, before the unicorn valuation and the government mandates, there was AI4Bharat.

AI4Bharat is a research initiative that Pratyush Kumar co-founded at IIT Madras with the specific mission of building open-source AI tools, datasets, and models for Indian languages. It is, in the most literal sense, a public good — a research infrastructure project aimed at the problem that Pratyush had been watching accumulate for years: the near-total absence of high-quality, large-scale datasets for India's twenty-two scheduled languages.

You cannot build an AI model for a language if you do not have the data to train it on. You cannot have the data if no one has systematically collected, annotated, and released it. AI4Bharat was the answer to that foundational problem. It created and released datasets for machine translation, automatic speech recognition, optical character recognition, and natural language understanding across Indian languages — work that would become the bedrock for Sarvam AI's models later.

<cite index="60-1">AI4Bharat, a research lab at IIT Madras focused on open-source AI tools, datasets, and models for Indian languages, secured significant funding including Rs. 36 crore from the Nilekani family in 2022 to establish the Nilekani Centre for AI.</cite>

The Nilekani connection is significant. Nandan Nilekani is not simply a technology philanthropist — he is the architect of Aadhaar, the man who built India's digital identity infrastructure from scratch and who understands, better than almost anyone alive, what it takes to deploy technology at the scale of 1.4 billion people. His family's investment in AI4Bharat was a signal of institutional confidence in the direction Pratyush was pointing: that Indian-language AI was not a niche academic problem but a foundational national priority.

<cite index="66-1">The startup was founded by Vivek Raghavan and Pratyush Kumar, who previously worked at AI4Bharat, an Indian-language AI research initiative at the Indian Institute of Technology Madras backed by technology veteran Nandan Nilekani.</cite>

AI4Bharat also connected Pratyush with his future co-founder, Vivek Raghavan — a complementary figure in almost every relevant respect. Where Pratyush brought deep systems engineering and AI research, Vivek brought a decade of experience building India's digital public infrastructure, including twelve years as a volunteer architect of Aadhaar's biometric systems at UIDAI. The two men had arrived, by different routes, at the same conviction: that India needed AI built from the ground up, for its own data realities, in its own languages, at its own scale.

PadhAI: The Teacher's Instinct

There is another chapter of Pratyush Kumar's story that tends to get less attention than his research credentials or his unicorn valuation, but that says as much about him as either. He co-founded PadhAI, an affordable online learning platform for deep learning and AI, with his IIT Madras colleague Mitesh Khapra.

<cite index="60-1">He co-built PadhAI, an affordable online learning platform that educated 50,000 students in deep learning for as little as Rs. 1,000 per course between 2016 and 2023.</cite>

<cite index="77-1">One Fourth Labs is a technology startup founded by IIT Madras faculty members Mitesh Khapra and Pratyush Kumar. It is incubated at the IIT Madras Research Park and educates and researches on Artificial Intelligence, specifically Deep Learning.</cite>

Fifty thousand students, at Rs. 1,000 per course. That number deserves to sit in silence for a moment. In a country where quality AI education is concentrated in a handful of elite institutions and is effectively inaccessible to the vast majority of engineering graduates, PadhAI made deep learning comprehensible and affordable to people who would otherwise never have encountered it.

This is not a conventional startup move. It is not the kind of thing you do if you are optimising for valuation multiples or press coverage. It is the kind of thing you do if you believe, at a fairly fundamental level, that knowledge should not be rationed by socioeconomic circumstance — and if you are willing to spend your own time and institutional resources to act on that belief.

The teacher's instinct that produced PadhAI is the same instinct that produced Sarvam AI's commitment to open-source model releases. The Sarvam-30B and Sarvam-105B models are not proprietary, locked behind a paywall accessible only to large enterprises. They are released under the Apache 2.0 license, freely available on Hugging Face, designed to be used, extended, and built upon by developers, startups, academic researchers, and governments across India and the world. That is a deliberate choice, made by someone who has spent his career thinking about how to maximise access to technology rather than how to maximise extraction from it.

Founding Sarvam AI: The Generative AI Moment Arrives

By the time Pratyush Kumar and Vivek Raghavan founded Sarvam AI in August 2023, the global AI landscape had been dramatically reshaped by the arrival of ChatGPT in November 2022. OpenAI's product had done something that years of academic AI research had failed to do: it had made the power of large language models legible to ordinary users. Suddenly, AI was not a research problem. It was a product experience, and hundreds of millions of people were having it.

For Pratyush, this was both a confirmation and an urgency. The confirmation was that the bet he had been making for years — that language models at scale would transform how people interact with information — was right. The urgency was that if India did not move quickly to build AI infrastructure tailored to its own linguistic and contextual realities, the default would be a global AI ecosystem in which India was a consumer but not a creator — one in which 1.4 billion people interacted with AI that understood them imperfectly, that processed their documents poorly, that spoke to them in accents and scripts it had been grudgingly retrofitted to handle.

<cite index="63-1">Pratyush Kumar is an Indian computer scientist and entrepreneur renowned for advancing artificial intelligence technologies tailored to Indian languages and democratizing access to generative AI. He co-founded Sarvam AI in July 2023, a startup dedicated to developing a sovereign AI stack that supports voice-based interactions and efficient models for India's linguistic diversity, aiming to serve over a billion users through practical applications like enterprise agents and legal tools.</cite>

The company they built was not designed to be a thin wrapper over an existing model. It was designed, from the first commit, to be a full-stack sovereign AI platform — meaning that Sarvam would build the foundational models, the inference infrastructure, and the enterprise applications, all in-house, all in India, all optimised for Indian languages and Indian use cases. This is an extremely ambitious technical programme. Most AI startups in India — indeed, most AI startups anywhere — do not attempt it. Building a foundational model from scratch requires enormous computing resources, world-class research talent, and a willingness to operate at timescales that are uncomfortable for venture-backed companies.

Sarvam attempted it anyway.

The Funding Journey: From Seed to Unicorn in Three Years

The financing of Sarvam AI is itself a story about the maturation of India's AI ecosystem, and about the growing conviction among global investors that India's AI moment is real.

In December 2023, just months after founding, Sarvam AI announced a combined seed and Series A funding round of approximately $41 million. <cite index="84-1">The round was led by Lightspeed Venture Partners, with participation from Peak XV Partners and Khosla Ventures.</cite>

Khosla Ventures deserves a specific note here: it is the fund founded by Vinod Khosla, the IIT Delhi alumnus who co-founded Sun Microsystems and who is one of the most important AI investors globally. His fund's participation in Sarvam's Series A — alongside Lightspeed and Peak XV, two of India's most respected institutional investors — was a signal that the smart money was taking Sarvam's mission seriously from the very beginning.

Then, in June 2026, came the moment that changed everything.

<cite index="66-1">Sarvam has raised $234 million at a $1.5 billion valuation, the company announced on Monday. The Bengaluru-based company is now India's newest AI unicorn. $150 million of that money comes from HCLTech, the IT subsidiary of Indian conglomerate HCL Group and lead strategic investor in the round. Bessemer Venture Partners also participated alongside existing backers Khosla Ventures and Peak XV Partners. Sarvam hopes to raise a total of $300 million for its Series B round.</cite>

HCLTech's $150 million investment is not simply financial. It is structural. HCLTech is one of India's largest IT services companies, with a global enterprise client base and engineering operations across dozens of countries. By investing strategically in Sarvam, it gains access to Sarvam's AI models and research capabilities. In exchange, Sarvam gains access to HCLTech's distribution network — the enterprise relationships that take a sovereign AI platform from a technical achievement to a commercial reality at scale.

<cite index="67-1">Bessemer Venture Partners' partner Pankaj Mitra said: "Sarvam is building and deploying India's sovereign AI platform — serving 1.4 billion citizens, mission-critical sectors, and large enterprises. Pratyush and Vivek have brought together the rare combination of research depth, engineering talent, and institutional trust to meet India's voice and agentic needs, and we are proud to partner with them."</cite>

In his own statement at the time of the raise, Pratyush was characteristically precise about what Sarvam is actually building: <cite index="67-1">"We are clear that research-led innovation to create AI that works at India's scale is a very large opportunity. That means models that understand our voices, read our documents, and serve intelligence at a cost every enterprise and government can afford. Building on this template, we are innovating on a full-stack offering for enterprises to own and operate their own sovereign AI."</cite>

That statement is worth reading carefully. "AI that works at India's scale" is not hyperbole — it is a technical specification. "Models that understand our voices, read our documents" is a product roadmap. "A cost every enterprise and government can afford" is a pricing philosophy that runs counter to how most AI companies think about monetisation. And "enterprises to own and operate their own sovereign AI" is a political and institutional stance, not just a business model.

The Government Mandate: Building India's Sovereign LLM

In April 2025, the Ministry of Electronics and Information Technology (MeitY) made one of the most consequential decisions in India's AI history: it selected Sarvam AI under the IndiaAI Mission to build India's first sovereign large language model.

<cite index="84-1">In April 2025, the Ministry of Electronics and Information Technology (MeitY) selected Sarvam AI as one of the companies to develop an indigenous foundational model under the IndiaAI Mission. As part of the</cite> programme, Sarvam received access to 4,096 NVIDIA H100 GPUs for model training — a computing resource that would otherwise cost tens of millions of dollars to access commercially.

The selection of Sarvam for this mandate was not a lottery. It reflected the government's assessment that Sarvam had the technical depth, the institutional relationships, and the demonstrated track record — through AI4Bharat, through PadhAI, through the early Sarvam models — to be trusted with a genuinely national infrastructure project. This is the kind of institutional confidence that cannot be manufactured. It is earned, slowly, through years of showing up and doing the work.

The sovereign LLM mandate also clarified the stakes. <cite index="66-1">The debate over AI sovereignty gained fresh urgency when Anthropic disabled access to its latest models after the U.S. government ordered the company to suspend their use by any foreign national, citing national security concerns. The move highlighted how access to cutting-edge AI systems remains concentrated among a small number of overseas providers.</cite>

In other words: the risk that Pratyush and Vivek had been warning about for years — that India's AI future could be constrained by decisions made in Washington or San Francisco, by companies with no accountability to Indian users or the Indian state — became concrete and visible in a single policy move. Sarvam's sovereign model is not an abstract ambition. It is infrastructure for a country that cannot afford to have its AI systems switched off by a foreign government's executive order.

The Products: What Sarvam Has Built

Under Pratyush Kumar's technical leadership, Sarvam AI has built one of the most comprehensive Indian-language AI portfolios in existence. Each product in the portfolio solves a specific, documented problem in the Indian context.

Sarvam-30B and Sarvam-105B

<cite index="84-1">On 18 February 2026, the company announced the release of two foundational models: Sarvam-30B — a 30-billion parameter model based on a mixture-of-experts design — and Sarvam-105B — a 105-billion parameter model activating approximately 9 billion parameters per token, with a 128,000-token context window.</cite> Both models were trained from scratch in India on datasets prioritising Indian languages, and both were released as open-source under the Apache 2.0 licence, hosted on Hugging Face and AIKosh. Open-sourcing a 105-billion parameter model is a statement: Sarvam is not trying to lock India's AI ecosystem behind a proprietary wall.

Sarvam Vision

<cite index="60-1">Sarvam Vision OCR leads benchmarks with 84.3 percent accuracy on olmOCR-Bench, surpassing Google's Gemini 3 Pro, DeepSeek OCR v2, and ChatGPT, while excelling in complex layouts and math formulas.</cite> The significance of this is harder to communicate in a headline than the benchmark number suggests. India's administrative and legal systems run on documents written in dozens of scripts — Devanagari, Tamil, Telugu, Bengali, Kannada, Malayalam, Gujarati, and more. These documents are the evidence of land ownership, insurance coverage, legal judgments, health records, and civic identity. A system that can read and process them accurately is not a productivity tool. It is governance infrastructure.

Bulbul V3

<cite index="60-1">Bulbul V3 text-to-speech model offers natural-sounding voices in 11 Indian languages, competing with global leaders like ElevenLabs at lower costs.</cite> Eleven Indian languages at natural-sounding quality. That sentence represents years of data collection, linguistic annotation, and model training — and it matters most not for enterprise call centres but for the elderly farmer in Rajasthan who receives an agricultural advisory in Hindi that sounds like a human speaking, not a robot reading a script.

Sarvam Samvaad

Samvaad is Sarvam's conversational AI platform for enterprises — designed to build, test, and launch AI agents fluent in 11 Indian languages across telephone, WhatsApp, web, and app channels. The pricing, according to Pratyush's public communications, targets Rs. 1 per minute for enterprise voice interactions. That figure is not accidental. It is a commitment to a price point that makes AI-powered customer service accessible to Indian banks, insurance companies, and government departments at a scale that global platforms cannot match.

Sarvam Kaze

<cite index="84-1">'Sarvam Kaze' is an indigenous AI-powered wearable glass that listens, understands, and captures what users see the world through their eyes in real time. The device supports more than 10 Indian languages, enabling voice-based interaction and potentially real-time translation.</cite> At the India AI Impact Summit 2026 in February, Prime Minister Narendra Modi was photographed wearing the Sarvam Kaze glasses — a moment of institutional endorsement that is difficult to overstate in its symbolic and commercial significance.

Sarvam-Translate

Released as a free API, Sarvam-Translate enables translation between all 22 official Indian languages and English. Pratyush announced its release on X with characteristic restraint: "New model drop — Sarvam-Translate is here. Can translate between 22 Indian languages and English. Significantly better than much larger models."

The Deployments: AI That Is Actually Reaching People

The test of any AI system is not whether it performs well on benchmarks. It is whether it reaches people who need it, in contexts that matter.

<cite index="69-1">Sarvam reported that its multilingual voice agents enabled India's Ministry of Agriculture and Farmers Welfare to gather information from 17 million farmers. Also, a nationwide insurance campaign helped policyholders renew 45 million policies.</cite>

Seventeen million farmers. Forty-five million insurance policies. These are not demo users or beta testers. They are citizens interacting with AI infrastructure in the course of real transactions — crop price inquiries, policy renewals, government service access — that have real consequences in their lives.

<cite index="66-1">Sarvam said a large fintech company is using its agentic AI platform to support a sales force of more than 350,000 people.</cite>

<cite index="72-1">Sarvam Vision is already being used to digitise over 35 million pages of records, including insurance documents, land records and other critical data.</cite>

SBI Life, Life Insurance Corporation, Tata Capital, CRED, IDFC, and UIDAI are among the named customers and partners. These are not small pilots. They are production deployments at institutional scale, by organisations that have no tolerance for systems that fail in the field.

Pratyush Kumar's Philosophy: What He Has Said in His Own Words

One of the most valuable windows into Pratyush Kumar's thinking is his public communications on X (@pratykumar), where he posts with unusual candour and technical specificity for a founder at his stage. A few excerpts that illuminate his worldview:

On India's response to DeepSeek's models, in early 2025: "The deepseek models are a step change and will enable faster progress towards even more powerful intelligent systems. What we in India should do: 1. We should absolutely build our own 'sovereign models' from scratch. 2. Our sovereign models should not compromise between" [quality and sovereignty].

On the government equity arrangement for the IndiaAI Mission compute: "Clarification: This is not a grant. A gov body will take equity in Sarvam for the compute we receive. And we are committed to building public interest use-cases and enabling the ecosystem in various ways such as hyper-optimising the inferencing costs in India."

On what Sarvam is fundamentally trying to do: addressing his audience of potential researchers, he wrote: "Next week I am in the Bay Area looking to meet exceptional researchers in the AI space. Come see what we have been building at Sarvam AI and what you can do to innovate for a billion Indians."

These are not the communications of a founder managing a narrative. They are the communications of someone who has been thinking very hard about a specific set of technical and institutional problems, and who is genuinely interested in building a community of people willing to work on them.

At the India AI Impact Summit 2026, speaking on behalf of Sarvam, <cite index="62-1">Pratyush Kumar emphasised the need for India to move "beyond English" in AI. He argued that user experiences differed widely across regions and cultures and that benchmarks must capture subjective and contextual realities, such as voice, vocabulary and lived experience.</cite>

The Significance: Why Pratyush Kumar's Work Matters Beyond Benchmarks

It is tempting to measure Pratyush Kumar's impact in the metrics that startup ecosystems produce: valuation, funding, headcount, revenue. Those numbers matter, and by any of them, Sarvam AI's trajectory is remarkable for a company less than three years old.

But the deeper significance is harder to quantify.

India is the world's most linguistically diverse large democracy. It has twenty-two scheduled languages and hundreds of dialects, scripts that predate the Roman alphabet by millennia, and a population of which the majority does not speak or read English with fluency. The global AI revolution, if it follows its current trajectory, will deliver most of its benefits to people who do: who can prompt a model in English, receive an English response, and navigate an English-language interface. For everyone else, the revolution will arrive late, partial, or not at all.

What Pratyush is building is the infrastructure for a different possibility: an AI future in which a woman in rural Maharashtra can ask a question in Marathi and receive a useful, accurate, affordable answer; in which a government clerk in Tamil Nadu can process a document in Tamil script without scanning it manually; in which the intelligence embedded in India's enormous diaspora of bureaucratic, legal, agricultural, and commercial knowledge can be made accessible to the people that knowledge is supposed to serve.

That is not a product pitch. It is a vision of what technology is for.

<cite index="71-1">Sarvam's rise to unicorn status is more than a startup success story. It represents a broader shift in India's technology ambitions. For years, India has been known primarily as a global IT services hub. Now, companies like Sarvam are attempting to position the country as a creator of foundational technologies rather than just a consumer or implementer of them.</cite>

The Road Ahead: What Comes Next for Sarvam AI

With $275 million raised and a $1.5 billion valuation established, the immediate future for Sarvam AI involves several clear priorities.

<cite index="66-1">The company will fund research into its next-generation AI models focused on agentic, coding, and cybersecurity applications, while also expanding access to computing infrastructure as it scales deployments across India.</cite>

The agentic AI direction is particularly significant. Agentic AI — systems that can execute multi-step tasks across business processes with minimal human intervention — is the fastest-growing category in enterprise AI globally. Sarvam is positioning itself to build agentic systems optimised for Indian languages and Indian institutional contexts: for banking workflows that run in Hindi, for insurance claims processes that require reading documents in Gujarati, for government service delivery that needs to operate in twelve different languages across twelve different states.

<cite index="84-1">In March 2026, Sarvam AI launched the Sarvam Startup Program, providing selected early-stage companies with 6–12 months of API credits scaled to their needs, priority engineering support, and access to production infrastructure for developing multilingual AI applications.</cite> This is ecosystem building — Pratyush is not just building Sarvam AI; he is building the infrastructure for a generation of Indian AI startups to build on top of it.

The target for the Series B is $300 million. With $234 million raised in the first close and the strategic partnership with HCLTech providing distribution reach, the second close seems well within reach.

Professional Profiles and Links

For those who want to follow Dr. Pratyush Kumar's work directly, learn from his thinking, or explore what Sarvam AI is building:

Dr. Pratyush Kumar — Direct Profiles

• X (Twitter): https://x.com/pratykumar — Most active; where he announces model releases, shares technical thinking, and responds to the community.
• LinkedIn: Search "Pratyush Kumar Sarvam AI" on LinkedIn or visit Sarvam AI's page at https://www.linkedin.com/company/sarvamai

Sarvam AI — Official Channels

• Website: https://www.sarvam.ai
• Developer Dashboard: https://dashboard.sarvam.ai
• X (Twitter): https://x.com/SarvamAI
• LinkedIn: https://www.linkedin.com/company/sarvamai
• Hugging Face (models): https://huggingface.co/sarvamai
• GitHub: https://github.com/sarvamai

Related Organisations

• AI4Bharat: https://ai4bharat.iitm.ac.in
• One Fourth Labs / PadhAI: https://www.onefourthlabs.com
• Sarvam Startup Program: https://www.sarvam.ai/startup-program

Timeline: Dr. Pratyush Kumar's Journey

Frequently Asked Questions

Who is Dr. Pratyush Kumar? Dr. Pratyush Kumar is an Indian computer scientist, AI researcher, and entrepreneur. He is the co-founder and CEO of Sarvam AI, a Bengaluru-based AI company building India's sovereign AI platform. He holds a B.Tech from IIT Bombay and a Ph.D. in Computer Engineering from ETH Zurich.

What is Sarvam AI? Sarvam AI is an Indian AI company founded in August 2023 by Pratyush Kumar and Vivek Raghavan. It builds foundational AI models, inference infrastructure, and enterprise AI products specifically designed for Indian languages and contexts. In June 2026, it became India's newest AI unicorn after raising $234 million at a $1.5 billion valuation.

What companies did Pratyush Kumar found before Sarvam? Before Sarvam AI, Pratyush co-founded AI4Bharat (open-source Indian-language AI research at IIT Madras), One Fourth Labs (the parent company of PadhAI, an affordable deep learning education platform), and OpenHathi (an Indian-language LLM initiative).

Where did Pratyush Kumar work before founding Sarvam AI? He held research positions at IBM Research (from 2015) and Microsoft Research (from 2017), and served as faculty and adjunct faculty at IIT Madras from 2018 onwards.

What is India's sovereign AI model? India's sovereign large language model is a foundational AI model being built entirely in India, on Indian data, for Indian languages and use cases. In April 2025, the Indian government's MeitY selected Sarvam AI to build this model under the IndiaAI Mission, providing access to 4,096 NVIDIA H100 GPUs for training.

Has Sarvam AI beaten ChatGPT and Google Gemini? Sarvam Vision, the company's OCR system, achieved 84.3% accuracy on the olmOCR-Bench, outperforming Google Gemini 3 Pro (80.2%) and ChatGPT (69.8%) on multilingual document understanding, particularly for Indian-language scripts.

What is Pratyush Kumar's X (Twitter) handle? His X handle is @pratykumar — https://x.com/pratykumar.

Who are Sarvam AI's investors? Sarvam AI's investors include HCLTech ($150M strategic lead in Series B), Bessemer Venture Partners, Lightspeed Venture Partners, Peak XV Partners, and Khosla Ventures.

This article is part of IndianAI.in's Indian AI Founders Series. We profile the builders, researchers, and entrepreneurs shaping India's AI ecosystem.

Last updated: July 2026.

Disclaimer

This article has been researched and written using publicly available sources including TechCrunch, Business Standard, Forbes India, Free Press Journal, Digit.in, Storyboard18, BusinessOutreach.in, the official Sarvam AI website, Wikipedia, Tracxn, and Dr. Pratyush Kumar's public posts on X (@pratykumar). Direct quotes attributed to Pratyush Kumar are sourced from his verified public communications and official company statements. All financial figures, valuations, and funding amounts are sourced from credible reporting at the time of writing.

IndianAI.in is an editorial platform. We have not independently verified every detail, and some information may have changed between the time of research and publication. Dr. Pratyush Kumar has not reviewed or authorised this article. Any factual errors are unintentional. If you are Dr. Pratyush Kumar or a representative of Sarvam AI and wish to correct or update any information, please contact us and we will update promptly.

External links are provided for reader convenience and informativeness. IndianAI.in does not have a commercial relationship with Sarvam AI or any of the companies or investors mentioned in this article.