India’s DeepSeek Moment: 🇮🇳 Why Sarvam AI is the Talk of Silicon Valley!

sarvam ai Feb 22, 2026

Sarvam AI for all from India

Built on sovereign compute. Powered by frontier-class models.
Delivering population-scale impact.

It’s a story that sounds almost too good to be true: a 2.5-year-old Indian startup, with a lean budget of $41 million, has beaten tech titans ChatGPT and Google Gemini on India-specific AI benchmarks. And the kicker? Their model, Sarvam AI, uses a mere 3 billion parameters compared to Gemini’s estimated 2 trillion – that’s a mind-boggling 667 times smaller! This isn’t just an impressive feat; it’s a watershed moment for Indian AI, a "DeepSeek" moment, if you will. It signals a seismic shift in how we think about and build AI for a country as diverse and complex as India.

This week, we delve deep into how Sarvam achieved this David-vs-Goliath victory, exploring their unique approach, their strategic partnerships with the Indian government, and what this means for the future of AI in the country.

Dominating the Benchmarks That Matter

Sarvam’s success isn’t based on some obscure, cherry-picked test. Their document-reading AI, Sarvam Vision, scored an impressive 84.3% accuracy on the OM OCR benchmark, which evaluates how well AI can read real-world Indian documents. We’re talking handwritten government forms, old financial records, regional language textbooks, and poorly scanned newspapers – the kind of messy, complex documents that global models often struggle with. In comparison, Gemini 3 Pro scored 80.2%, and ChatGPT lagged at 69.8%.

But it’s not just about reading text. Sarvam’s text-to-speech model, Bulbul V3, also excelled in blind studies, proving to be the top-rated model for telephone-grade audio (8 kHz), the standard for call centers and voice agents in India. While global players like ElevenLabs shine in studio-quality audio, Bulbul V3 outperformed them in the real-world, low-bandwidth conditions that define much of India’s digital landscape.

The Secret Sauce: Built for India, by India

So, how did they do it? The answer lies in their fundamental approach. Sarvam didn’t try to out-compete global models on a broad spectrum of general knowledge. Instead, they asked a crucial question: what does India actually need that global models aren’t delivering? The answer was clear: the ability to read Indian scripts, speak Indian languages with natural accents, and understand the intricate nuances of Indian context.

Global models are typically trained on data from sources like Reddit, Wikipedia, and GitHub, which are predominantly in English (around 90%). This makes them ill-equipped to handle the linguistic diversity and messy, real-world data found in India. Sarvam, on the other hand, trained their models exclusively on Indian data: millions of government forms, banking documents, textbooks in 22 regional languages, historical newspaper archives dating back to the 1800s, and voice recordings with diverse regional accents. They even built custom tokenizers specifically optimized for Indian scripts like Devanagari and Tamil, which are notoriously inefficient for standard tokenizers.

A Strategic Partnership for Sovereign AI

Sarvam’s success hasn't gone unnoticed by the Indian government. In April 2025, the startup was selected as the first to develop India’s sovereign foundational large language model under the India AI Mission. This partnership provides them with access to 4,000 GPUs for six months, along with government equity – a clear indication that homegrown AI is now considered critical infrastructure for national security and digital sovereignty.

This was followed by the announcement of "Digital Saum" in January 2026, India’s first sovereign AI research park, a joint initiative between Sarvam, IIT Madras, and the Tamil Nadu government. With a massive 10,000 crore rupee investment over five years, this park will house research labs, startup incubation facilities, an AI governance institute, and a 20-megawatt AI-optimized data center. Sarvam has also partnered with the Odisha government for a 50-megawatt AI compute facility and is working with UIDAI (the organization behind Aadhaar) to power AI voice interactions for 1.38 billion Indians.

This strategic alignment between Sarvam and the government underscores the growing recognition that relying solely on foreign AI for critical services is a strategic vulnerability. Sarvam is positioning itself as the cornerstone of India’s AI self-reliance.

The Power of Focused AI

It’s important to note that Sarvam isn’t trying to be a jack-of-all-trades. You won’t find their AI writing poetry or explaining quantum mechanics. Their focus is laser-sharp: building essential AI infrastructure for Indian enterprises and government entities. This includes document processing for banks, voice AI for rural call centers, and translation services for government outreach.

Their founders, Pratyush Kumar and Vivek Raghavan, bring a wealth of technical expertise and a deep passion for building India-specific products. Raghavan’s background, including a PhD from Carnegie Mellon and nearly 12 years as a volunteer architect for Aadhaar, is particularly relevant to Sarvam's mission.

Their strategy is clear: while global giants fight for the consumer market with general-purpose chatbots, Sarvam is quietly building the foundational AI infrastructure that powers the Indian economy. Their pricing model is also tailored for the Indian market, with their voice AI agent costing just one rupee per minute – significantly cheaper than global alternatives.

In Other News...

New Rules for AI-Generated Content: The Indian government has amended the IT Rules 2021 to regulate AI-generated content, requiring clear labeling, traceable metadata, and faster takedowns (within 3 hours) for illegal content. This move aims to combat deepfakes and misinformation.
Space Tech Funding: Three Indian space startups – Dhruva Space, and Astrome Technologies – have received grants to build indigenous small satellite bus platforms, boosting India’s position in the global small satellite market.
Funding Roundup: This week saw $247 million raised by Indian startups, with notable investments in Radiance Renewables (green energy), Pandorum Technologies (biotech), LightMed (pain relief device), Satlab Space Systems (satellite data relay), and SteamPro (smart steam bath systems).

This week’s news highlights the dynamic and rapidly evolving nature of the Indian startup ecosystem, with significant developments in AI, space tech, and other key sectors. Sarvam AI’s success, in particular, is a powerful testament to the potential of focused, homegrown innovation to solve complex challenges and compete on the global stage.