As global tech giants pour billions into India’s AI landscape, a critical question emerges: can the nation achieve true technological sovereignty, or will it remain dependent on foreign infrastructure and governance? A S Rajgopal, CEO & MD of NxtGen Cloud Technologies, cuts through the hype surrounding India’s AI ambitions with a pragmatic assessment of what sovereignty actually means—and what it takes to build it.
CEO & MD
NxtGen Cloud Technologies
In this conversation, Rajgopal examines the gap between investment announcements and indigenous capability, the strategic value of GPU-dense training clusters, and why India needn’t replicate Silicon Valley’s playbook to succeed. He argues that operational control matters more than physical ownership, and that India’s path to AI leadership lies not in chasing technological firsts, but in demonstrating transformative impact at scale across healthcare, education, and governance.
CIO&Leader: India wants sovereign AI—but the gap is real. What’s the hardest barrier to closing it today?
A S Rajgopal: Several global technology companies have announced investments exceeding USD 50 billion in India by 2030. However, most of these investments are driven by market expansion and talent availability, not by the creation of sovereign AI capabilities. These companies operate under U.S. jurisdiction, which means India does not have full control over the infrastructure, data, or governance frameworks that power their AI systems.
Globally, nations are reassessing their digital dependence on a small group of technology providers. The concerns span national security, data sovereignty, economic competitiveness, cultural and linguistic relevance, and regulatory independence. For India, the challenge is to strike a balance between accessing advanced technologies and retaining strategic control over them. Encouragingly, efforts are underway to build local capabilities across AI models, platforms, and the high-performance compute infrastructure required to run them.
CIO&Leader: Can India win in AI without copying the OpenAI or DeepSeek model? What should it do differently?
A S Rajgopal: Neither OpenAI nor DeepSeek introduced entirely new foundational concepts in AI; both are built on long-standing research approaches to machine intelligence. Many leading models today use the Mixture of Experts (MoE) architecture, a concept first introduced in 1991 and progressively refined over the years. By around 2013, researchers began integrating MoE layers into deep neural networks at scale.
This highlights that there are multiple established pathways for innovation in AI. Developing Indian models using MoE or similar techniques should not be viewed as imitation, but as part of the natural evolution of research-driven progress. Globally, this approach is already in practice, for instance, G42 has announced a Hindi–English model built by pre-training open-source architectures such as Meta’s Llama and Qwen 2.5.
For India, leveraging open-source models can accelerate AI development while preserving cultural and linguistic relevance, alongside targeted efforts to build models from scratch where strategic differentiation is required.
CIO&Leader: How does NxtGen’s GPU-dense training cluster change India’s ability to build models at scale?
A S Rajgopal: Training large AI models requires highly stable, high-performance infrastructure capable of processing massive datasets over extended periods. India has historically lacked such large-scale training environments.
The objective of a GPU-dense training cluster is to update model parameters at speed while maintaining numerical stability at scale. Even minor inconsistencies can force weeks of training to restart. Equally critical is high-performance storage, ensuring GPUs are continuously fed with data and do not sit idle.
NxtGen has drawn on learnings from large global deployments to architect a training environment optimized for low latency, high availability, and massive parallelism. This kind of infrastructure is foundational for India to train advanced models reliably and at scale.
CIO&Leader: “Sovereign AI” is widely used—but what does it actually mean in infrastructure and control terms?
A S Rajgopal: Sovereign AI goes beyond owning physical infrastructure. True sovereignty means full control across the stack, including data, operations, and the ability to run and manage systems independently without reliance on foreign entities.
Operational sovereignty is especially critical. If a system cannot be operated, updated, or secured without external dependencies, it is not truly sovereign. Sovereign AI must ensure autonomy not just in ownership, but also in execution.
CIO&Leader: As AI adoption accelerates, how do you balance speed, regulation, and trust by design?
A S Rajgopal : India has access to advanced AI technologies, but the challenge lies in sustaining investment in infrastructure that becomes obsolete very quickly. AI systems today often have a lifecycle shorter than three years.
Global demand has already created shortages in critical components such as memory, impacting not only AI infrastructure but consumer devices as well. This reality will likely slow deployment in India, but it also provides an opportunity to align infrastructure growth with evolving regulatory frameworks.
Trust in AI systems must be built through compliance, transparency, and responsible deployment. Speed without trust will not deliver sustainable adoption.
CIO&Leader: Five years from now, what must India get right to become a serious global AI contender?
A S Rajgopal : Success should not be measured by technological leadership alone, but by impact. In five years, the key question should be how AI has improved quality of life, productivity, and global competitiveness for Indian businesses and citizens.
India has the opportunity to lead by demonstrating how AI can be applied at scale across healthcare, education, governance, and citizen services. Leadership in responsible and inclusive use of AI will matter far more than being first to market with a particular model.