5 Key Takeaways
- Devendra Singh Chaplot, an IIT-Bombay alum with experience at Mistral AI, Thinking Machines Lab, and xAI, joined Sarvam AI as a part-time advisor to bridge India's AI ambitions with Silicon Valley frontier research.
- Sarvam AI raised $234 million in Series B funding at a $1.5 billion valuation and is expanding into the U.S. with a San Francisco office and Bay Area research lab.
- Chaplot helped build major Mistral models like Mistral 7B, Mixtral 8x7B, Mistral Large, and Pixtral Large, and led Mistral's U.S. office, giving him firsthand experience scaling frontier AI startups.
- Chaplot's motivation for joining Sarvam is the need to build locally customized AI models for India's linguistic and cultural diversity, since global models largely under-serve Indian languages.
- His appointment reflects a reverse brain drain of Indian AI talent returning home, giving Sarvam credibility in building a trillion-parameter model despite intense competition and heavy compute challenges.
Sarvam AI's Quiet Coup: Why IIT-Bombay Alum Devendra Chaplot Could Be India's Secret Weapon in the Global AI Race
When Sarvam AI hosted its Epoch 2026 developer conference in Bengaluru this week, the spotlight fell unexpectedly on a soft-spoken researcher who wasn't a founder or a venture capitalist. Devendra Singh Chaplot, a name that resonates deeply within the corridors of the world's most elite artificial intelligence labs, had just signed on as a part-time advisor to the Indian startup. His arrival signals more than a high-profile hire; it marks a strategic move by Sarvam to bridge the gap between India's AI ambitions and the frontier research happening in Silicon Valley. For a company openly declaring plans to build a trillion-parameter AI model from scratch, Chaplot's presence is a credible bet on homegrown talent returning to shape the country's technological destiny.
The stakes couldn't be higher. Sarvam has raised $234 million in the first close of its Series B funding round at a valuation of $1.5 billion. It is expanding aggressively into the United States with a new office in San Francisco and a dedicated research lab in the Bay Area. At the center of this transcontinental push is Chaplot, an IIT-Bombay graduate and one of the very few Indian-origin researchers to have held key roles at three of the world's most talked-about AI companies: Mistral AI, Thinking Machines Lab, and Elon Musk's xAI.
Who Is Devendra Chaplot?
To understand why Chaplot's appointment matters, one must trace his journey from the classrooms of Mumbai to the bleeding edge of machine learning. Chaplot earned his Bachelor of Technology in Computer Science and Engineering from the Indian Institute of Technology Bombay, an institution famous for producing engineers who go on to lead global tech giants. He then moved to the United States, where he completed a Master's in Language Technologies at Carnegie Mellon University, followed by a PhD in Machine Learning from the same institution. His doctoral research focused on intelligent autonomous navigation, a field that sits at the intersection of robotics, computer vision, and artificial intelligence—teaching machines to perceive, move, and make decisions in physical spaces.
Even before finishing his PhD, Chaplot was already gaining hands-on experience at some of the biggest names in consumer technology. He interned at Samsung Electronics and Apple, cutting his teeth on real-world AI challenges. But it was his five-year stint at Facebook AI Research (FAIR) that truly honed his expertise. There, he worked on embodied AI, robotics, and computer vision—building systems that help machines understand and navigate the physical world. Embodied AI, for the uninitiated, refers to algorithms that can control a physical body or avatar, a stepping stone toward robots that can interact naturally with their environment.
In 2023, Chaplot took a career-defining leap. He became one of the founding researchers at Mistral AI, a Paris-based startup that quickly emerged as Europe's boldest challenger to OpenAI. At Mistral, Chaplot was instrumental in building some of the company's most celebrated models—software systems trained on vast datasets to generate text, analyze images, or perform complex reasoning. These included Mistral 7B, Mixtral 8x7B, Mistral Large, Pixtral 12B, and Pixtral Large. A model like Mistral 7B has 7 billion parameters, which are the internal knobs and weights the system tunes during training to learn patterns from data. The larger the number of parameters, the more nuanced the model can be, though it also requires more computational power. Chaplot didn't just create these models; he also set up and led Mistral's U.S. office in Palo Alto, California, giving him firsthand experience in scaling a frontier AI startup across continents.
In early 2025, Chaplot moved to Thinking Machines Lab, the venture founded by former OpenAI Chief Technology Officer Mira Murati. As Tech Lead for Data and Pre-training, he worked on Tinker—the company's training API (Application Programming Interface, a set of tools that allows developers to build on top of existing technology)—and helped shape its large language model infrastructure. His tenure there was brief but influential. Earlier this year, he joined Elon Musk's xAI (now operating as SpaceXAI), where he worked on superintelligence research before leaving after a short, roughly two-month stint. That whirlwind month brought him into the orbit of Sarvam, a company whose mission dovetailed perfectly with his own growing conviction: that India needed to build its own frontier AI.
Why Sarvam? Why Now?
Chaplot's decision to advise Sarvam is rooted in a clear philosophy he articulated at the Epoch 2026 conference.
"I have been part of founding teams at frontier labs. You need a handful of experts and experienced people, but we need more motivated talent who can catch up very quickly and run the company."
The word "frontier" in AI refers to the cutting edge of research and development, where companies push the boundaries of what models can do. Chaplot's statement underscores a truth often overlooked: sheer expertise matters, but so does the hunger and agility of a team that feels a deep connection to the problem it is solving.
For Chaplot, that problem is India's linguistic and cultural diversity.
"For India, we have so many languages and cultures. We want to have models that can be customised. This is the reason we absolutely need to build models here."
Most global AI models are predominantly trained on English and a handful of other widely spoken languages. Indian languages—with their distinct scripts, grammar, and cultural contexts—are often underserved. Building a model that understands and generates text in Assamese or Malayalam, that can parse the nuances of Indian-English code-switching, or that respects regional sensitivities is not a luxury; it is a necessity if AI is to serve over a billion people. Chaplot's insight is that off-the-shelf solutions from Silicon Valley will never fully capture this richness. Indigenous models, built by teams who live and breathe the culture, are the only path forward.
According to his LinkedIn profile, Chaplot officially joined Sarvam in July as a part-time advisor. He will be based in Palo Alto, California, working from the company's U.S. research base. This arrangement lets him stay plugged into the global AI ecosystem while directly shaping Sarvam's strategy. It is a unique hybrid role: not a full-time employee, but a guiding force who can bridge worlds. Given his track record, he brings not just technical chops but also the scar tissue of having helped scale a startup from its earliest days.
Sarvam's Grand Ambitions
Sarvam is not playing small. The startup is building a foundational AI model with more than one trillion parameters. To put that in perspective, earlier generations of models like GPT-3 had 175 billion parameters. A trillion-parameter model is an order of magnitude larger, requiring staggering amounts of data, computational power, and engineering finesse. Foundation model is the term for a large, general-purpose AI system trained on diverse data that can then be adapted to many downstream tasks, from writing code to translating languages to assisting scientific research.
Sarvam's research scope spans coding, speech, vision, cybersecurity, and scientific AI. This broad mandate reflects a belief that frontier AI is not just about text generation but about multiple forms of intelligence working in concert. The company's expansion into the Bay Area signals a determination to attract world-class talent and collaborate with the best minds in the global ecosystem, rather than attempting to innovate in isolation.
The Epoch 2026 conference in Bengaluru served as a showcase for these ambitions. While the event featured product demos and technical talks, Chaplot's quiet presence sent the loudest message: Sarvam has the credibility to woo a researcher who could easily have stayed at any of the most prestigious labs in the world. His compensation, while undisclosed, is reported to be substantial—an acknowledgment that world-class advisors command world-class paychecks.
The Broader Implications
Chaplot's move is part of a larger narrative: the reverse brain drain of Indian AI talent. For decades, top engineering graduates left India to pursue opportunities in the U.S. and Europe. Now, with a maturing startup ecosystem and growing capital availability, some are choosing to contribute directly to India's technological rise. Sarvam's unicorn valuation and its bold technical roadmap make it a magnet for such talent.
There are challenges ahead. Building a trillion-parameter model is a compute-heavy endeavor. It requires access to tens of thousands of specialized chips, robust data pipelines, and a team that can orchestrate training runs that can last months. Competition from OpenAI, Anthropic, Google DeepMind, and others is fierce. These incumbents have deep pockets and massive head starts. Yet, Chaplot's presence alone won't guarantee success, but it dramatically improves Sarvam's odds. He has navigated the exact path the startup now traverses: from assembling a founding research team to shipping state-of-the-art models that the world actually uses.
Moreover, Sarvam's strategy to open a U.S. lab while remaining anchored in India mirrors a blueprint that Chaplot executed at Mistral. It allows the company to tap into the Bay Area's talent pool and investment networks while keeping the mission firmly tied to India's unique needs. The outcome could be a virtuous cycle: models developed for Indian languages and contexts find global applications, attracting more talent and investment back home.
For observers of the AI industry, the appointment of Devendra Singh Chaplot at Sarvam is a bellwether. It shows that the next chapter of AI development may not be written solely in San Francisco or Beijing. It could well be penned in Bengaluru, by researchers who once left and are now returning to build something truly from the ground up. As Chaplot himself implied, the frontier belongs to those who show up with both expertise and deep motivation. Sarvam has just found its expert in residence. The coming months will reveal how quickly the rest of the team catches up.
Sarvam AI's Epoch 2026 developer conference took place in Bengaluru. Devendra Singh Chaplot serves as a part-time advisor based in Palo Alto, California.
No comments:
Post a Comment