The Global Obsession with Size
For the past few years, the artificial intelligence landscape has been dominated by a race to build ever-larger models. Tech giants have poured billions into creating Large Language Models (LLMs) with hundreds of billions, or even trillions, of parameters.
The underlying logic was simple: a bigger model, trained on more data, would be more knowledgeable, capable, and intelligent. For many general-purpose tasks—like writing an email, summarising a document, or generating code—these massive models have proven incredibly powerful and versatile. They represent a monumental leap in computing, capable of understanding and generating human-like text on a vast array of topics, making them seem like the default choice for any business looking to adopt AI.
The Cracks in a One-Size-Fits-All Model
However, as Indian enterprises move from experimentation to real-world application, the limitations of these global behemoths are becoming apparent. India's digital landscape is unique and presents structural challenges for models primarily trained on English-language data. With 22 official languages and thousands of dialects, the linguistic diversity is immense. Global LLMs often struggle with the nuances, idioms, and cultural context embedded in these languages. A significant portion of online data is in English, meaning Indian languages are underrepresented, leading to poorer performance or costly and inefficient translation processes. Beyond language, there is the challenge of 'code-switching,' where users mix multiple languages in a single conversation—a common practice in India that can confuse foreign models.
The Rise of Local Champions
This gap has created a fertile ground for a new breed of India-focused AI startups. Companies like Krutrim and Sarvam AI are building models from the ground up with a specific focus on the Indian context. For example, Krutrim, which became India's first AI unicorn, has developed a model trained on over two trillion tokens, with a heavy emphasis on Indic languages. Sarvam AI's OpenHathi series is another example, specifically designed to be more efficient and effective for Indian languages and use cases. These are not just scaled-down versions of global models; they are architected differently to prioritise linguistic depth and cultural fluency. This focus allows them to handle tasks in languages like Hindi, Tamil, and Bengali with greater accuracy and relevance than their much larger global counterparts.
The Small Model Advantage: Cost, Speed, and Privacy
The case for smaller, specialised models—often called Small Language Models (SLMs)—goes beyond just better language support. A key driver is practicality. Running massive LLMs requires immense computational power, which translates to high costs and reliance on cloud infrastructure. SLMs are cheaper, faster, and more efficient. They can be deployed on private servers or even on-device, which is a critical advantage for sectors like finance, healthcare, and legaltech, where data privacy and compliance with regulations like the Digital Personal Data Protection (DPDP) law are non-negotiable. For businesses dealing with real-time voice AI or customer service, the lower latency of SLMs provides a much better user experience. This shift proves that for many enterprise applications, the most powerful model isn't the best one; the most practical and efficient one is.
Context is the New King
Ultimately, the trend reveals that in AI, as in business, context is king. A model's value isn't just determined by the size of its dataset, but by the relevance of that data to the task at hand. While a global LLM can write a poem about the monsoon, a local model understands its cultural significance, its economic impact on farmers, and the specific regional vocabulary associated with it. This deep contextual understanding is what separates a generic tool from a true solution. The rise of India-focused AI tools shows that the market is maturing. It's moving beyond the initial hype of giant, do-everything models toward a more nuanced, hybrid approach where specialised tools are used for specific jobs. This strategy allows businesses to balance performance, cost, and compliance effectively.
