In a significant development in the global AI race, the balance has tilted in favour of a Bengaluru-based startup. HCLTech is investing $150 million in a $300 million funding round led by Bessemer Venture Partners, with strategic investment from NVIDIA and participation from Prosperity7, Activate, Glade Brook Capital, and existing investors.
The deal values the three-year-old company at approximately $1.5 billion, nearly seven times its 2023 valuation, and is more than just another large venture capital infusion. It signals a serious challenge to the long-held notion that Silicon Valley knows best when it comes to cutting-edge AI.
While OpenAI and Google continue to dominate English-first applications for global markets, Sarvam AI is building foundational infrastructure for India’s 1.4 billion people, many of whom prefer speaking in any of the country’s 22 official languages over typing.
With HCLTech committing around $150 million and NVIDIA providing critical technical backing, the message is unmistakable – India is no longer just a consumer market for artificial intelligence. It is fast emerging as a major innovation laboratory for it.
From Aadhaar to AI: The emergence of Sarvam
Sarvam AI didn’t just appear out of nowhere. It was forged in the research labs of IIT Madras and the bureaucratic fire of India’s digital public infrastructure. Founders Vivek Raghavan and Pratyush Kumar were eager to set it up.
Raghavan spent over a decade as a key volunteer architect behind Aadhaar, the world’s largest biometric ID system, and the India Stack. He understands “India Scale” – the kind of scale where a 1% error rate means millions of people are left behind.
Kumar, a former researcher at Microsoft Research and an associate professor at IIT Madras, led the groundbreaking work at AI4Bharat, a research lab dedicated to Indian language technology.
When they founded Sarvam in August 2023, the goal wasn’t to build a better poet or a snarkier chatbot. They were asking a fundamental question: How do we make AI work for a kirana store owner in Bihar who only uses voice commands?
Their early journey was defined by a rejection of the ‘translation layer’ approach, where Western models translate English logic into Hindi. Instead, they spent their first years scouring the subcontinent for trillions of tokens of indigenous data, capturing everything from legal Sanskrit to street-level Hinglish.
Why Sarvam isn’t just another GPT clone
To understand why NVIDIA and HCLTech are writing nine-figure checks, you have to look at the ‘Tokenisation Tax.’
Most global LLMs are trained primarily in English. When you ask them a question in an Indian language like Telugu or Marathi, the model has to break those words into tokens. Since these models aren’t optimised for Indian scripts, a single word in Hindi might take four times as many tokens as its English equivalent. This makes running AI in India four times more expensive and significantly slower.
Sarvam’s differentiation lies in three core technical pillars:
The voice-first architecture: Unlike Western models designed for text-heavy interfaces, Sarvam’s Saaras (speech-to-text) and Bulbul (text-to-speech) models are built for the low-latency, noisy environments of real-world India. They handle ‘code-switching’ — the natural way Indians hop between English and regional languages, with a fluidity that makes Siri sound like a relic.
The MoE efficiency: Their latest models, the Sarvam-30B and Sarvam-105B, use a Mixture-of-Experts (MoE) architecture. Instead of firing every “neuron” in the model for every query, the system only activates the experts needed for that specific task. This drastically reduces the compute cost, making it viable for a mid-sized Indian bank or a government agency to run the model without a Silicon Valley budget.
Edge intelligence: With the debut of Sarvam Kaze – AI-powered wearable glasses – the company is moving AI off the cloud. These devices perform real-time translation and document reading locally, a necessity in areas where 5G is still a dream.
The corporate logic: Why NVIDIA and HCLTech?
The entry of NVIDIA and HCLTech as lead investors isn’t just about the money. Rather, it’s about a symbiotic survival strategy.
NVIDIA wants to build the “Sovereign AI” moat
For NVIDIA, Sarvam is the ultimate proof of concept for their “Sovereign AI” narrative. CEO Jensen Huang has been vocal about the idea that every nation should own its own “intelligence production.” By backing Sarvam, NVIDIA ensures that India’s national AI mission is built on NVIDIA H100 and B200 GPUs, thus ensuring a new business stream.
“India has the data and the talent; what it needs is the sovereign capacity to build,” an NVIDIA spokesperson hinted at the recent IndiaAI Impact Summit.
Sarvam is also a key member of NVIDIA’s Nemotron Coalition, an elite group of global firms developing open-source foundation models. For NVIDIA, Sarvam is the gateway to ensuring that the next billion AI users are running on green-and-black silicon.
HCLTech’s survival pivot
For HCLTech, the $150 million lead investment is a defensive masterstroke. The traditional IT outsourcing model – “bodies in seats” doing manual coding and back-office tasks, is under existential threat from AI automation. By integrating Sarvam’s foundational models into their “AI Factory,” HCLTech can offer something no global consultancy can – a deeply localised AI that can automate a hospital’s patient records in Tamil or a logistics firm’s fleet management in Gujarati.
It seems that HCLTech isn’t just investing in a startup but buying an insurance policy. They are transforming from a service provider into an AI orchestrator.
Sarvam Indus to bridge the digital divide
The culmination of this journey is the Indus app—Sarvam’s answer to the consumer AI race. But unlike ChatGPT, which feels like a blank cursor, Indus feels like a digital assistant that actually knows where you are from.
As Sarvam moves into its next phase, the challenge will be scaling these models for the “bottom of the pyramid.” While Big AI firms in the US are obsessed with reaching AGI (Artificial General Intelligence), Sarvam is focused on ‘General Utility.’
“We aren’t trying to build a god in a box,” Vivek Raghavan recently stated in a public statement. “We are trying to build a tool that speaks your language.”
With $300 million in the bank and the world’s most powerful chipmaker in its corner, Sarvam AI has officially ended the era of India being an AI consumer and begun the chapter of an AI player.