India Unleashes AI Breakthrough, Defies US-China Tech Domination

Paul Riverbank, 2/9/2026Sarvam AI's homegrown OCR model outperforms global giants, marking a leap for India’s digital sovereignty. With visionary leadership, the firm signals a shift: AI innovation tailored to India’s linguistic diversity, empowering local control in an era where technology equals influence.

Now, to the uninitiated, “optical character recognition” (OCR) may sound drab or technical. It’s software that converts text in images—scans, handwritten notes, signs—into words a computer can handle. But here’s the thing: India, with its patchwork of dozens of scripts, poses a problem that has tripped up the best in the business. While foreign models fumble over Devanagari or Kannada, Sarvam AI’s homegrown effort, Sarvam Vision, just beat out some of the world’s most advanced systems—including ChatGPT, Google’s Gemini, even Anthropic’s vaunted Claude—according to recent independent evaluations.

At the heart of the operation are Pratyush Kumar and Vivek Raghavan. Both have resumes impressive enough to quiet a room. Kumar took his doctorate at ETH Zurich, after cutting his teeth at IIT Bombay and Microsoft Research. He’s got a knack for marrying abstract academic research with challenges on the ground. (His earlier stint with AI4Bharat pushed for language tools that finally gave Telugu or Bengali their due.) Raghavan, for his part, is the sort of veteran technologist who’s helped shape India’s foundational digital infrastructure—think biometric identity, mobile payments, or the secure exchange of sensitive legal documents.

But it isn’t about one successful model. Kumar is frank on this point: It’s about building digital tools that genuinely speak India’s languages, fit India’s paperwork, and meet local realities. “We set out to prove that world-class AI could be built here, not just imported and patched together,” he told me. That distinction—a shift from tinkering with foreign code to constructing foundational models end-to-end, right at home—taps into something deeper. It’s what policymakers now call “sovereign AI.”

Look at Raghavan’s earlier work and you see the pattern too. He played a central role in launching SUVAS, a platform now used by the Indian judiciary to translate court orders and judgments into the languages spoken in every district. In another push, he’s helped develop systems that spot anomalies in GST records—think digital sleuthing to catch tax fraud before it spirals. It’s meticulous, sometimes slow-going work, but it has quietly transformed how official India moves and organizes information.

These local initiatives—driven by both technical prowess and a sensitivity to Indian complexity—echo a larger, international drift. China’s been explicit about building up its own chips and data forts. Europe, perhaps more subtly, insists on keeping its digital assets close, hedging against over-dependence on Silicon Valley or Shenzhen. The thread running through these strategies is familiar: control over your own digital destiny.

Sarvam AI’s victory in OCR, then, is more than a technical coup. Imagine the difference it makes for a government office in rural Odisha, where important land deeds arrive smudged and handwritten in Odia script, or for thousands of visually impaired citizens who depend on accurate text-to-speech. Suddenly, the promise of digital inclusion becomes more than an ideal. It becomes practical—available.

Of course, real challenges remain. Beating global benchmarks is one thing; embedding advanced models in schools, banks, and clinics from Kerala to Kashmir is another. “A great lab result doesn’t always translate to social impact,” Kumar readily admits. For Sarvam AI, the next chapter isn’t about topping charts—it’s about weaving their tools into India’s bewildering diversity.

There’s a broader message here, almost a lesson in economic history: those who build and hold their own infrastructure—be it gold reserves, chip factories, or, now, AI models—claim greater say over their future. Sarvam AI’s story is a modest, but potent, example of that logic in motion. Even as headlines swirl with geopolitical competition, sometimes the quiet revolutions matter most.

If there’s been a prevailing narrative in the world of artificial intelligence, it’s that big American and Chinese firms—think OpenAI, Google, Anthropic—steer the global conversation. For a long time, others were left scrutinizing the rearview mirror, hoping to catch up. Yet, unexpected as it might seem, a team in urban Bengaluru is starting to turn that story inside-out. Their name? Sarvam AI.Now, to the uninitiated, “optical character recognition” (OCR) may sound drab or technical. It’s software that converts text in images—scans, handwritten notes, signs—into words a computer can handle. But here’s the thing: India, with its patchwork of dozens of scripts, poses a problem that has tripped up the best in the business. While foreign models fumble over Devanagari or Kannada, Sarvam AI’s homegrown effort, Sarvam Vision, just beat out some of the world’s most advanced systems—including ChatGPT, Google’s Gemini, even Anthropic’s vaunted Claude—according to recent independent evaluations.At the heart of the operation are Pratyush Kumar and Vivek Raghavan. Both have resumes impressive enough to quiet a room. Kumar took his doctorate at ETH Zurich, after cutting his teeth at IIT Bombay and Microsoft Research. He’s got a knack for marrying abstract academic research with challenges on the ground. (His earlier stint with AI4Bharat pushed for language tools that finally gave Telugu or Bengali their due.) Raghavan, for his part, is the sort of veteran technologist who’s helped shape India’s foundational digital infrastructure—think biometric identity, mobile payments, or the secure exchange of sensitive legal documents.But it isn’t about one successful model. Kumar is frank on this point: It’s about building digital tools that genuinely speak India’s languages, fit India’s paperwork, and meet local realities. “We set out to prove that world-class AI could be built here, not just imported and patched together,” he told me. That distinction—a shift from tinkering with foreign code to constructing foundational models end-to-end, right at home—taps into something deeper. It’s what policymakers now call “sovereign AI.”Look at Raghavan’s earlier work and you see the pattern too. He played a central role in launching SUVAS, a platform now used by the Indian judiciary to translate court orders and judgments into the languages spoken in every district. In another push, he’s helped develop systems that spot anomalies in GST records—think digital sleuthing to catch tax fraud before it spirals. It’s meticulous, sometimes slow-going work, but it has quietly transformed how official India moves and organizes information.These local initiatives—driven by both technical prowess and a sensitivity to Indian complexity—echo a larger, international drift. China’s been explicit about building up its own chips and data forts. Europe, perhaps more subtly, insists on keeping its digital assets close, hedging against over-dependence on Silicon Valley or Shenzhen. The thread running through these strategies is familiar: control over your own digital destiny.Sarvam AI’s victory in OCR, then, is more than a technical coup. Imagine the difference it makes for a government office in rural Odisha, where important land deeds arrive smudged and handwritten in Odia script, or for thousands of visually impaired citizens who depend on accurate text-to-speech. Suddenly, the promise of digital inclusion becomes more than an ideal. It becomes practical—available.Of course, real challenges remain. Beating global benchmarks is one thing; embedding advanced models in schools, banks, and clinics from Kerala to Kashmir is another. “A great lab result doesn’t always translate to social impact,” Kumar readily admits. For Sarvam AI, the next chapter isn’t about topping charts—it’s about weaving their tools into India’s bewildering diversity.There’s a broader message here, almost a lesson in economic history: those who build and hold their own infrastructure—be it gold reserves, chip factories, or, now, AI models—claim greater say over their future. Sarvam AI’s story is a modest, but potent, example of that logic in motion. Even as headlines swirl with geopolitical competition, sometimes the quiet revolutions matter most.