India’s generative AI revolution has a new vanguard—Sarvam AI. Founded by veteran technologists Pratyush Kumar and Vivek Raghavan, the Bengaluru-based startup is on a mission to build an AI infrastructure that reflects India’s diversity, scale, and needs. Their vision? A sovereign AI stack that serves every Indian—across languages, professions, and devices.
The Case for India’s Own AI Foundation
At the heart of Sarvam AI lies a bold belief: India must not merely consume AI; it must build it. Unlike many AI startups that build applications on top of pre-trained foreign models, Sarvam is developing foundational technology from the ground up—algorithms, training techniques, tokenizers, datasets, and architectures tailored to Indian needs.
With this approach, Sarvam is setting out to do for GenAI what UPI did for digital payments—create scalable, sovereign, and inclusive infrastructure with global export potential.
Backed by Top Investors for Deep Tech Innovation
Despite being just a year old, Sarvam AI—formally Axonwise Pvt. Ltd.—has already raised $53 million in funding from Lightspeed, Peak XV Partners, Khosla Ventures, and Venture Highway. The substantial backing is a testament to investor confidence in India’s potential to create indigenous GenAI platforms, not just derivative products.
Voice, Vernacular, and Efficiency: The Sarvam Stack
India’s linguistic diversity and digital infrastructure constraints demand a unique AI approach. Sarvam’s strategy focuses on
- Small, efficient models for Indian languages
- Voice-first interfaces for accessibility across literacy levels
- Multilingual token optimization for accurate Indian language outputs
Their flagship open-source model, Sarvam 2B, is trained on 2 billion parameters and 2 trillion tokens, with 40% of tokens dedicated to Indian languages—a staggering leap compared to global datasets like Common Crawl, where Indian languages make up less than 0.1%.
Sarvam’s models are also hardware-efficient, trained on 1,000 Nvidia H100 GPUs, versus the 16,000 GPUs used by Meta for Llama 3, making them cost-effective and energy-efficient for enterprise adoption.
Sarvam Agents: India’s Voice-First AI Platform
A cornerstone of Sarvam’s offerings is Sarvam Agents, a suite of voice-enabled GenAI tools that go beyond chat. These are task-oriented AI agents capable of booking appointments, managing transactions, and interacting directly with enterprise systems—across 10 Indian languages and channels like WhatsApp, apps, and the web.
By allowing businesses to integrate AI agents into their workflows at just ₹1 per minute, Sarvam is making voice-based automation affordable and scalable.
AI That Works Where It Matters
From customer service to healthcare and e-commerce to legal tech, Sarvam is building AI for high-frequency, narrow-use tasks that affect millions. Its legal tool, Sarvam A1 Legal, streamlines research, document drafting, and redaction while being continuously updated with real-time Indian regulatory data.
By offering APIs and a modular “orchestration fabric,” Sarvam enables enterprises to pick and choose from its AI stack—or deploy it entirely on-premise for data-sovereignty-compliant use cases.
Sovereign AI, Trusted AI
Recognizing the need for privacy, compliance, and control, Sarvam also offers its models in appliance form, allowing enterprises to run AI on their own servers. The company is working closely with Nvidia to develop a Virtual Unified Environment (VUE) appliance tailored for high-performance GenAI workloads in Indian enterprises.
This approach addresses one of the biggest barriers to AI adoption in sectors like BFSI, healthcare, and government—data security and model transparency.
The People Behind the Mission
Sarvam’s founding team brings unmatched credibility to the table. Vivek Raghavan, a Carnegie Mellon PhD, played a key role in building Aadhaar and IndiaStack. Pratyush Kumar, an IIT-Bombay and ETH Zurich alumnus, previously co-led AI4Bharat at IIT Madras, building open-source models and tools for Indian languages.
Their mission is not only technical but also civic: to ensure AI serves every Indian—not just the tech-savvy elite. Their team of 40 young engineers and researchers is building India’s AI full stack, from tokenizer to voice bot, from inference engine to interface.
A Vision That Goes Beyond India
Though built for India, Sarvam’s models are finding early takers in Africa and Southeast Asia, with enterprises in Singapore already piloting solutions. The parallels to UPI’s international expansion are evident—and intentional.
Sarvam’s founders believe India can lead the Global South in deploying inclusive, low-resource AI. Their models are designed to run on constrained devices, understand code-mixed language, and perform reliably in noisy, real-world conditions.
What’s Next: Building an AI Bharat
From deploying QR-code-driven learning assistants for schools to building custom LLMs for enterprises, Sarvam is entering scale mode. Future workbenches are being planned for fields like healthcare, agriculture, and education.
The startup is also eyeing the “LLM-as-a-service” space, offering to build custom models for organizations with proprietary data—turning its foundational tech into a revenue-generating service layer.
Final Word
As India navigates the GenAI era, Sarvam AI represents more than a promising startup—it’s a blueprint for how developing countries can shape the AI future on their own terms.
By prioritizing voice, vernacular, sovereignty, and scale, Sarvam is building a GenAI ecosystem where AI doesn’t just speak your language—it understands your context.
And that’s how AI, made in India, can serve not just a billion Indians—but billions more around the world.
