India’s AI growth will come from smaller, cost-effective models, says Sridhar Vembu

 At the recently concluded India AI Impact Summit 2026 in New Delhi, a clear narrative emerged that could define the subcontinent's technological trajectory for the next decade. While global tech leaders like OpenAI’s Sam Altman and Google’s Sundar Pichai walked the halls of Bharat Mandalay, it was a distinctively homegrown philosophy that captured the spotlight. Zoho founder and CEO Sridhar Vembu articulated a vision that diverges sharply from the Silicon Valley playbook: India’s AI growth will not come from competing in the "parameter. The "Unglamorous" Workhorses of the Economy Speaking on the sidelines of the summit, Vembu observed that the current investment trend in India is firmly tilted toward models that prioritize efficiency over sheer size. Those should be unglamorous right now, but they get the job done," Vembu stated. This pragmatic approach acknowledges a fundamental reality: for the vast majority of India's economic and social needs, trillion-parameter models are overkill. Vembu pointed to homegrown startups like Sarvam AI as exemplars of this trend, noting that they are already launching models with a smaller computational footprint. He predicted that this would become the norm, envisioning a landscape filled with "smaller resource efficient, energy efficient models" that can scale up over time as infrastructure costs inevitably decline.  Union Minister for Electronics and IT, Ashwini Vaishnaw, used the summit platform to declare a major strategic pivot. In a moment that surprised many, he announced that India would not be chasing gargantuan models, stating that "95% of AI work" can be accomplished with smaller architectures. The government is now betting heavily on "Sovereign Small Models" tailored for Indian languages and contexts. Why Small is a Strategic Bet? The logic behind this pivot is multilayered, driven by both necessity and opportunity. The Economic Survey of 2025-26 highlighted that developing frontier models is a capital-intensive endeavor concentrated among a handful of global players with resources India currently lacks.

 The compute constraint: India faces a significant hardware gap. Vembu himself has previously pointed out that while companies like OpenAI have access to hundreds of thousands of GPUs, India receives a quota of only about 50,000 high-end GPU chips annually. Trying to out-compute the incumbents is a losing battle. Small models, however, can be trained and run on modest infrastructure, making them accessible to a wider range of startups and institutions. Democratization and Scale: Small models are cheaper to deploy, which means they can be integrated into everyday applications without requiring users to own the latest hardware.  "This is going to be the norm in India," Vembu reiterated, emphasizing the scalability of this approach as costs continue to fall. With a growing focus on sustainability, the lower energy consumption of smaller models aligns perfectly with India's commitment to green technology. These models can run on edge devices and legacy hardware, drastically reducing the carbon footprint of AI deployment. The "Jugged" Innovation: Solving Real Problems. The summit served as a showcase for how this "small is beautiful" philosophy translates into tangible impact. These are not just theoretical models; they are solutions "with a smell of the soil," as one observer put it. Gnani.ai's "Inca" Voices: A 5-billion parameter model that can clone a local accent in seconds. It is being used in rural Andhra Pradesh to enable uneducated farmers to execute financial transactions using simple voice commands in Telugu. Bharat Gen's "Parameter 2": A 17-billion parameter model that can translate between 22 official Indian languages, breaking down linguistic barriers in governance and education. Amul's "Saracen" App: Born from a direct challenge by Prime Minister Modi on whether AI could help dairy farmers, this model monitors cow health by analyzing their calls and eye images. Project Stand: A specialized model focused on material science, specifically designed to help India reduce its reliance on imported lithium by accelerating research into sodium-ion batteries.

The Demographic Dividend and The Future of Work Beyond the technology itself, Vembu highlighted India's greatest asset: its people. "With our vast youth population, we have the most AI-enthusiastic population in the world," he said.. This rapid adoption rate is a crucial competitive advantage, creating a massive user base that can provide feedback and data to refine these small models further. Regarding the perennial fear of job losses, Vembu remains optimistic. He draws on historical precedent, noting that past technological transitions—from the PC to the internet to e-commerce—ultimately created more jobs than they destroyed. "The nature of jobs may change and roles may evolve," he noted, but the key lies in adaptability. For software engineers specifically, he advises a shift in focus: moving from pure coding toward getting "closer to the customer, [to] solve customer problems He advocates for a focus on skill over degree, encouraging students and professionals to engage in hands-on projects and continuous learning. Endorsements and Challenges The strategy is gaining international recognition.

 OpenAI CEO Sam Altman, who once famously suggested it would be hopeless to compete with OpenAI on training foundational models, praised the "incredible small language AI models" emerging in India. He remarked that the building energy in India is "quite remarkable" and unlike anything he has seen elsewhere. However, the path forward is not without its skeptics and challenges. An opinion piece in the Economic Times warns that the term "small models" is an umbrella that covers very different strategies—from fine-tuning open-weight models like Llama to building purpose-built models from scratch. It cautions that if India relies too heavily on fine-tuning foreign open-source models, its AI capability could remain "downstream" of decisions made in the U.S. Furthermore, a debate exists within the tech community about whether this focus on narrow applications might cause India to miss out on foundational research. Some worry that the country might become a "user" of AI rather than a "creator" of the core algorithms that power it. The recent public exchange between Verb and former Infosys CFO Mohandas Pai highlights the tension between those who believe in aggressive, service-led deployment and those who advocate for a more cautious, foundational approach. Conclusion Sridhar Vembu’s vision for Indian AI is a masterclass in playing to one's strengths. By sidestepping a futile battle over who can build the biggest model, India is instead focusing on a war of attrition in utility and reach. The goal is to saturate the economy—from the dairy farm to the district court—with specialized AI tools that enhance human productivity rather than replace it. As Vembu succinctly put it, the path forward involves a lot of experimentation.  . If the innovations on display at the summit are any indication, that direction is leading India toward a future where AI is less about sci-fi wonder and more about the quiet, efficient, and "unglamorous" work of nation-building.