Can India Build AI Disruptors Like DeepSeek?

In January 2025, the AI landscape witnessed a seismic shift with the emergence of DeepSeek R1, a large language model (LLM) developed in China. Unlike previous AI breakthroughs, DeepSeek R1 did not stun the world with superior capabilities but rather with its resource efficiency. While leading models from OpenAI, Google, and Meta relied on tens of thousands of GPUs for training, DeepSeek R1 reportedly required only a few thousand, challenging the long-standing belief that compute power is the ultimate determinant of AI success.

DeepSeek's success raises an important question: Can India develop a similar AI disruptor? To answer this, we must examine the four key factors contributing to AI success: innovation, compute power, data quality, and serving capacity.

Innovation: Can India Match DeepSeek’s Breakthroughs?

DeepSeek’s remarkable efficiency stems from its innovative approach, leveraging Mixture of Experts (MoE) models and Multi-head Latent Attention. This breakthrough highlights the importance of cutting-edge AI research rather than brute-force computing power.

India, with its vast pool of STEM talent, is well-positioned to drive AI innovation. The country has a strong tradition of excellence in mathematics and computer science, and its IT sector is one of the most advanced in the world. AI breakthroughs do not require an army of engineers—just a handful of brilliant minds making novel contributions. Given India's intellectual capital, there is no reason why Indian researchers cannot achieve a similar breakthrough in AI efficiency.

Compute Power: Does India Have the Necessary Resources?

DeepSeek R1’s parent company, High-Flyer, a well-funded Chinese hedge fund, reportedly owns tens of thousands of GPUs. While DeepSeek R1 was trained using a few thousand GPUs, the actual number used for training its base model remains unclear. If it truly required minimal compute resources, this would signal a major shift in AI development.

India’s access to GPUs has traditionally been a challenge, with most advanced chips controlled by US.-based companies like NVIDIA. However, India has been making strides in AI infrastructure, with government initiatives aiming to boost domestic semiconductor production. If AI efficiency breakthroughs continue, India's relative lack of GPU resources may become less of a bottleneck.

Data Quality: Can India Curate High-Quality Datasets?

Data is a critical component of any AI model. Companies like OpenAI, Meta, and DeepSeek invest heavily in curating high-quality datasets to train their models. India, with its diverse linguistic and demographic landscape, has the potential to build expansive and high-quality datasets for AI training.

While data access and processing capabilities remain a challenge, the country’s growing AI ecosystem, backed by government and private sector initiatives, is well-positioned to address these gaps. The key will be ensuring that Indian AI models are trained on clean, diverse, and unbiased data to remain competitive.

Serving Capacity: The Biggest Roadblock?

Even if India successfully builds an efficient AI model, the challenge of deploying it at scale remains. Models like GPT-4 and Gemini serve millions of users simultaneously, requiring vast data centers powered by thousands of GPUs and consuming gigawatts of electricity. While DeepSeek may have achieved efficiency in training, its widespread adoption still demands significant compute resources for inference and deployment.

India’s data center infrastructure is improving, but large-scale AI deployment remains a challenge. Unlike training, where efficiency can reduce GPU dependence, inference and serving require robust, energy-intensive infrastructure. Without significant investment in AI infrastructure, Indian models may struggle to achieve widespread adoption.

Regulatory and Ethical Considerations

Major AI models undergo rigorous safety evaluations and comply with strict regulations concerning content moderation, copyright compliance, and ethical AI use. DeepSeek, operating under fewer governance constraints, has an advantage in rapid deployment.

India’s AI regulations are evolving, and any homegrown AI disruptor will likely face scrutiny before public deployment. While necessary for responsible AI development, this could slow down progress compared to models operating in more permissive environments.

Can India Create the Next DeepSeek?

India has the talent, innovation potential, and data resources to build efficient AI models. However, compute power constraints and the need for large-scale deployment infrastructure present significant challenges. While models like DeepSeek prove that breakthroughs are possible without massive compute power, serving AI at scale remains a major hurdle.

Ultimately, India can achieve a breakthrough in AI, but taking the world by storm will require more than just a lean and efficient model. Investment in AI infrastructure, regulatory clarity, and a focus on deployment capabilities will be key to ensuring that India not only builds AI disruptors but also successfully deploys them on a global scale.

Pawan Prabhat

---

Pawan Prabhat is Co-Founder of Shorthills AI.