The Global Obsession with Size
For years, the artificial intelligence landscape has been dominated by a technological arms race. Giants like Google, OpenAI, and Meta have been locked in a battle to build the largest, most powerful Large Language Models (LLMs). The prevailing wisdom
was that more parameters—billions, or even trillions—equated to more capability, leading to jaw-dropping feats of text generation, reasoning, and creativity. However, this pursuit of scale comes at a staggering price. Training these frontier models costs millions of dollars in computing power, requiring vast farms of specialized GPUs that are both expensive and in short supply. For most of the world, including a majority of Indian enterprises and startups, building or even running these behemoth models is simply out of reach.
The High Cost of Being Left Out
The challenges posed by massive, global AI models go beyond just the initial cost. They present significant structural hurdles for India. Firstly, there's the 'language tax'. Most global LLMs are trained predominantly on English-language data, as it makes up the vast majority of content on the internet. When processing queries in Indian languages, these models often have to translate to English and back, a clunky process that increases costs, slows down response times, and often misses cultural and linguistic nuance. This makes them inefficient for a country with 22 official languages and thousands of dialects. Secondly, reliance on foreign-built models raises concerns about data privacy and digital sovereignty, prompting a government-backed push for homegrown AI capabilities under initiatives like the IndiaAI Mission.
Enter the Small Language Model
In response to these challenges, a powerful new trend is taking hold: the rise of Small Language Models (SLMs). These are leaner, more focused AI systems with parameters in the millions or low billions, rather than hundreds of billions. Instead of trying to be a jack-of-all-trades, SLMs are designed to be masters of one. They can be trained faster, more efficiently, and at a fraction of the cost. Because they are less resource-intensive, they can run on private servers or even on-device, which is a major advantage for sectors like banking and healthcare that require strict data privacy and compliance. This shift from generic scale to hyper-localized performance and affordability is proving to be India's strategic advantage.
Built for Bharat: AI with an Indian Context
The true power of smaller models is unlocked when they are trained on India-specific data. Startups like Sarvam AI and Krutrim, along with academic consortiums like BharatGPT, are leading this charge. Sarvam AI's OpenHathi series, for instance, is a Hindi-first model built on top of open-source architecture that has been fine-tuned for the nuances of the language. It dramatically reduces the 'token fertility' issue, where a single Hindi word is broken into multiple tokens by Western models, making it far more efficient. Similarly, Ola's Krutrim is developing a full-stack AI ecosystem, including models that understand over 10 Indian languages and can perform real-world tasks. The IndiaAI Mission is actively supporting these efforts, providing subsidised compute power to help startups build foundational models for everything from agriculture and education to video generation.
A Hybrid Future and the Road Ahead
The case for smaller models isn't about completely replacing their larger counterparts. Instead, India's AI future is likely to be a hybrid one. Massive LLMs can still serve as powerful reasoning engines for complex, broad problem-solving, while nimble SLMs act as the operational backbone for high-volume, localized, and regulated applications. This approach allows Indian businesses to innovate practically, creating chatbots for local-language customer service, AI-powered advisories for farmers, and accessible healthcare information for rural communities. While challenges like the scarcity of high-quality digital data in many Indian languages remain, the path is clear. The focus on cost-effective, culturally-aligned, and efficient AI is not just a workaround; it's a strategic choice that positions India to become a leader in building practical AI solutions for its own population and for other emerging markets facing similar constraints.
