India, July 14 -- India stands at an inflection point in Artificial Intelligence (AI). The debate is often framed as a binary choice: Should India invest in building its own large language models (LLMs), or should it simply build AI applications on top of models developed elsewhere? This is the wrong question. The right strategy is not "build versus use." It is "build enough to remain technologically sovereign while winning commercially at the platform and application layers." There is little doubt that the greatest commercial opportunity over the next five years lies above the foundation model layer. Intelligent model routing, enterprise AI infrastructure, trusted data platforms, multimodal AI, edge deployment, and domain-specific systems for health care, agriculture, education, manufacturing, finance, etc, will create enormous value. These are areas where India enjoys significant structural advantages: world-class engineering talent, Digital Public Infrastructure, cost-efficient innovation, and access to large, real-world datasets. However, it would be a profound strategic mistake to conclude that India can dominate these higher layers without investing seriously in foundational AI research. Every major breakthrough in the next generation of AI - reasoning models, autonomous agents, neuro-symbolic systems, and architectures that may eventually succeed today's transformers - will emerge from deep understanding of current foundation models. No nation can leapfrog technologies it has never mastered. The organisations that define tomorrow's AI will be those that understand today's AI. I know this from 50 years of my own experience in semiconductor technologies; Today's semiconductor miracles could not exist without 50 years of foundational learning. Every innovation at the frontier of semiconductors today is not a departure from the past - it is the past, compressed, recombined, and pushed to its physical limits. China's DeepSeek offers an important lesson. Its success came after it mastered large-scale model training, distributed computing, optimisation, reinforcement learning, and model distillation. The breakthroughs that attracted global attention were possible only because the underlying scientific and engineering capabilities had already been built. The lesson is clear: Innovation follows mastery - not avoidance. India, therefore, does not need to outspend the US or China in a race to build the world's largest frontier model. But it can't afford to abandon foundation model development altogether. The objective should instead be to build strategic capability. India should establish one flagship national frontier-model initiative, supported by a small number of complementary research centres at leading institutions such as IISc, the IITs, and a capable private-sector partner. Rather than dispersing scarce talent and compute across multiple disconnected efforts, the country should concentrate its resources on building one world-class programme, with affiliated centres contributing specialised research in multilingual AI, Indic language tokenisation, evaluation frameworks, reinforcement learning, safety, and domain adaptation. Such a programme is not about surpassing the latest models from the AI leaders. It is about ensuring that India possesses the talent, engineering expertise, and institutional knowledge required to understand, audit, improve, and independently operate advanced AI systems. France's experience with MistralAI is instructive. Mistral was never intended to replace OpenAI globally.Its value lies in giving Europestrategic autonomy - the ability to inspect, customise, and deploy frontier-classAI without depending entirely on foreign providers. That carries long-term strategic value. India should pursue a similar objective. In the near term, however, the country's most realistic and commercially valuable goal is developing world-class expertise in adapting and extending leading open-weight models such as Llama, Mistral, and DeepSeek. Sovereign capability in post-training, reinforcement learning, safety, evaluation, domain adaptation, and multilingual fine-tuning is likely to generate far greater practical value over the next three to five years than attempting to compete head-on in trillion-dollar pretraining races. This approach still requires deep research capabilities. Fine-tuning frontier models for India's languages, health care, agriculture, legal frameworks, education, and governance demands sophisticated AI research and advanced engineering talent. The skills developed through this work will ultimately position India to undertake increasingly ambitious foundation-model efforts in the future. None of this, however, ispossible without sovereign AI infrastructure. High-performance GPUclusters are strategic infrastructure on par with power grids and telecommunications networks. Without domestic compute, India cannot securely train, fine-tune, or deploy AI systems using sensitive national datasets spanning health care, agriculture, financialservices, legal records, scientific research, and government applications. Over the past two years, I have had the opportunity to work closely on developing sovereign GPU technology for India as a founding advisor to the emerging startup, Agrani. This experience has reinforced my conviction that indigenous compute capability is a strategic necessity. AI sovereignty begins with compute sovereignty. Equally important is India's ability to run and customise leading open-weight models on infrastructure located within its own borders. The GPU capacity now being created through the India Semiconductor Mission and related national initiatives is, therefore, far more than an infrastructure project. It is the foundation upon which India's long-term AI independence will rest. Yet compute alone will not be sufficient. India's greatest challenge may well be talent. Many of the world'sleading AI researchers of Indian origin currently work at the world's premier AI laboratories. Building a sovereign AI capability requires attracting at least some of this extraordinary talentback into the national ecosystem through globally competitive research environments, ambitious missions, entrepreneurial opportunities, and meaningful long-term support. India has repeatedly demonstrated in the past that it can createglobally competitive technologyecosystems when policy, talent, and entrepreneurship align. AI presents the next great opportunity. India should not attempt to win the global AI race by building the biggest language model. Nor should it resign itself to becoming merely an application developer dependent on technologies built elsewhere. Instead, India should invest enough in frontier AI research to preserve technological sovereignty, build world-class expertise in open-model adaptation and post-training, develop sovereign compute infrastructure, attract exceptional research talent, and focus commercial energy where India has the greatest comparative advantage - the platform and application layers that will define AI's real-world impact. The choice before India is whether to remain a creator of the technologies that will shape the future - or become permanently dependent on those created by others....