When we conceived Shakti Cloud, we anchored it to a single, non-negotiable truth: Computing only finds its true purpose within a secure, sovereign environment. But let’s be clear: sovereignty isn’t about isolation or building walls; it’s about having the agency to innovate on your own terms.
I've spent years building infrastructure in India. And if there's one thing I’ve learned, it's that the real contest is never about what's on the surface. It's about what runs underneath.
AI is no different. Everyone is watching the models, the applications, the launches. I'm watching the stack. Because whoever controls the foundational layer controls where the value ultimately lands - and for how long.
India has never had a shortage of demand or talent. What we've historically lacked is ownership at the bottom of the stack. Consider this: India generates nearly 20% of the world's data; yet holds just about 3% of global data centre capacity. We have built for the world, brilliantly, but rarely focused on the infrastructure we control.
That gap was manageable in a services-led or SaaS-driven model. AI changed that. As the stack deepens, value increasingly shifts to those who control its foundations. If the infrastructure layer is not built domestically, the risk is not just economic leakage, but a steady erosion of control over how and where innovation scales.
Where AI Runs Now Matters as Much as What It Does
The next five years will be defined not by which country builds the best models, but by which country builds the infrastructure those models run on.
When AI was experimental, it didn't matter much where it was trained or deployed. But AI is no longer experimental. It is embedded in financial services, healthcare, manufacturing, and governance. At that point, the question of where it runs becomes as consequential as what it does.
Milliseconds matter. A delay in fraud detection, a lag in a clinical support system, latency in an industrial operation - these are not technical inconveniences. They are business risks. And data is not freely mobile. Data residency, compliance, and sovereign requirements are reshaping how systems are architected from day one.
Then there’s the shift in workloads. The early AI era was about training -- large, intermittent bursts of compute. The next era is about inference -- running models continuously, at scale, in live environments. Inference is already overtaking training and is expected to account for nearly two-thirds of all AI compute. Unlike training, inference is persistent. It is directly tied to usage, revenue, and user experience. Inference is where value is realised. And that makes infrastructure performance inseparable from business performance.
The Real Contest Is Over Infrastructure
When we conceived Shakti Cloud, we anchored it to a single, non-negotiable truth: Computing only finds its true purpose within a secure, sovereign environment. But let’s be clear: Sovereignty isn’t about isolation or building walls; it’s about having the agency to innovate on your own terms. We didn’t build this to turn inward, but to project outward. Shakti Cloud is made in India for the world -- a high-performance engine designed for global interconnectivity and built to carry the weight of the world’s most ambitious workloads. Beyond providing raw power, we offer a comprehensive ecosystem of accelerated platforms, automated AI workflows, and enterprise-grade software services. By bridging local self-reliance with a global service architecture, we aren’t just building a cloud; we are creating the definitive global hub for the next era of international AI innovation.
What sovereignty also means is: No single country or company should be able to dictate your digital future. Use global technologies, but use them within your own environment, under your own control. Data processing, storage, model development -- none of it should be vulnerable to external disruption or someone’s unilateral decision.
That conviction shapes every decision we make at Yotta. We function as the physical realisation of the entire AI stack. At the foundational infrastructure level, we provide specialised high-density power and liquid cooling essential for heavy AI workloads. Moving up to the compute and networking layers, Yotta offers scalable access to massive clusters of specialised accelerators (GPUs) and high-speed interconnects, essentially turning raw hardware into a utility. Finally, we bridge the gap to platform services by providing integrated software environments and developer tools via our Shakti Cloud platform that allows organisations to train, fine-tune, and deploy AI models without managing the underlying complexity, making it a true end-to-end provider for sovereign AI development.
Shakti Cloud today runs more than 10,000 Nvidia GPUs in live production, supporting training, fine-tuning, and inference workloads. Our roadmap extends beyond 80,000 NVIDIA GPUs by FY 2026-27, backed by over $2 billion committed to Nvidia's latest chips. What these numbers represent is end-to-end control over compute, data, and execution within a single governed framework - not fragmented across providers or jurisdictions.
When critical infrastructure sits outside your control, options narrow over time - on pricing, on access, on the ability to scale when it counts. Owning infrastructure is about ensuring that decisions on cost, security, and innovation remain yours to make.
India is already beginning to reflect this shift. We are seeing high-priority cloud and AI workloads being rerouted from geopolitically sensitive regions into India. Enterprises are placing greater weight on resilience for mission-critical systems, not just cost efficiency. Improving subsea connectivity, access to high-performance compute, and a maturing policy environment around data governance are all moving in the right direction. India is being evaluated seriously as a part of the underlying infrastructure layer for global AI systems. Not just as a consumption market. At the India AI Impact Summit 2026, companies pledged $240 billion toward India's AI ecosystem, a calculated conviction about where the next phase of global AI infrastructure will be built.
The Question That Actually Matters
Will we remain a large, sophisticated consumer of AI built and run elsewhere? Or will we emerge as a location where AI is trained, deployed, and monetised on infrastructure we control?
I know where I'm placing my bet. And I know what it will take. Not just investment and policy, but the conviction to build foundational infrastructure even when the returns will take time to fructify and the risks are real.
The shift happening below the surface is quieter than a model launch. Less visible than a funding round. But in my experience, the quieter the shift, the more consequential it tends to be.
The author is Co-founder, Managing Director and CEO at Yotta. Views expressed are personal.