From ‘hopeless’ to 'hopeful': Indian AI startups that can challenge global AI giants

In 2023, during a visit to India, OpenAI CEO Sam Altman asserted that it was “totally hopeless to compete with us on training foundation models,” suggesting that developing large-scale AI models required resources beyond the reach of most startups.

However, the recent emergence of DeepSeek, a Chinese AI startup, has challenged this notion. DeepSeek’s open-source AI model, R1, developed at a fraction of the cost of its Western counterparts, has demonstrated capabilities comparable to leading models like OpenAI’s ChatGPT. This development has prompted a reevaluation of the competitive landscape in AI model training.

In India, several companies are making significant strides in AI development, focusing on creating Large Language Models (LLMs) tailored to the nation’s diverse linguistic landscape:

1. Sarvam AI
• Sarvam 2B: An open-source language model trained on 4 trillion tokens, supporting 10 Indian languages, including Hindi, Tamil, and Telugu. It focuses on tasks like translation and summarization.

2. Tech Mahindra
• Project Indus: An open-source AI model designed to enhance enterprise solutions such as customer support and automation, with a focus on Indian languages.

3. Gyan AI
• Paramanu: A series of lightweight AI models optimized for Indian languages like Assamese, Bangla, Hindi, and Tamil. These models are efficient, requiring fewer computational resources while delivering high performance.

4. Yellow.ai
• YellowG: A proprietary AI model aimed at automating customer service through chatbots and voice assistants, providing human-like conversational experiences across various industries.

5. Uniphore
• Conversational AI: Specializes in AI-driven speech recognition and automation to improve call center efficiency, reduce customer wait times, and enhance response quality, supporting multiple Indian and global languages.

6. Ola
 • Krutrim: A homegrown large language model trained on extensive datasets of Indian languages and contextual data, enabling accurate text generation and conversational AI. It’s integrated into Ola’s services, including mobility and customer support.

7. Hanooman AI
• Hanooman: A multimodal AI model capable of handling text, speech, and vision-based tasks in multiple Indian languages, designed for applications like chatbots and video analysis.

8. CoRover.ai
• BharatGPT: A multilingual virtual assistant designed for sectors like government services, banking, and e-commerce, supporting regional Indian languages to facilitate communication for non-English speakers.

9. AI4Bharat
• IndicTrans2: Focuses on developing high-quality machine translation models for all 22 scheduled Indian languages, creating large training datasets and multilingual models to promote accessibility.

10. Abhinand Balachandran
• Tamil-LLAMA: A language model specifically designed for the Tamil language, enhancing vocabulary and fine-tuned for tasks relevant to Tamil language processing.

11. KissanAI
• Dhenu 1.0: An AI solution tailored for the agricultural sector, assisting farmers in identifying diseases in crops like rice, maize, and wheat through a conversational interface.

12. Telugu LLM Labs
• Navarasa 2.0: An advanced language model supporting 16 languages, including Hindi, Telugu, Tamil, Kannada, and Malayalam, suitable for applications like content generation and educational resources.

These developments indicate that Altman’s earlier assessment may have underestimated the potential for innovation in regions like India. The success of companies like DeepSeek underscores that significant advancements in AI are achievable without the extensive resources traditionally deemed necessary.