India’s AI conversation has centered on compute clusters and cloud infrastructure. Rajesh Subramaniam, CEO and founder of embedUR, argues this misses the bigger opportunity. In this exclusive interview with Analytics Insight, he makes the case for edge AI as the missing layer of India’s sovereignty strategy—one that keeps intelligence on factory floors, in clinics, and on farms rather than routed through servers abroad. He also explains how initiatives like IndiaAI Mission, Digital India, and Make in India could converge at the device level, and what enterprises need to do to turn AI pilots into measurable business results.
Q: Can India truly achieve sovereign AI if intelligence remains concentrated in the cloud while edge deployment is overlooked?
A: Sovereignty involves more than the location of model training or data centers. If every inference in an Indian factory, hospital, or utility relies on a cloud, often managed abroad, the intelligence layer of the economy remains dependent on external infrastructure, pricing, and policies. True sovereignty requires intelligence to operate where India’s data is generated: on devices, factory floors, clinics, vehicles, and other endpoints. Edge AI keeps data local by default, operates reliably with limited connectivity, and reduces foreign exchange costs associated with large-scale cloud inference. Cloud AI and edge AI are complementary. However, a sovereignty strategy limited to data centers perpetuates structural dependency. India should control both layers.
Q: Why will sovereign edge infrastructure be critical to India’s long-term AI ambitions?
A: When we look at India’s AI journey, three factors stand out: scale, resilience, and economics. Let’s start with scale. Our opportunity is massive—AI in India is not just about a few thousand enterprise deployments, but about reaching over a billion endpoints across agriculture, manufacturing, healthcare, energy, and mobility. Relying solely on the cloud for real-time inference at this scale is simply not practical or sustainable. The way forward is clear: embed intelligence directly into our devices, and use the cloud for training, coordination, and data aggregation. This is how we unlock true impact. Resilience comes next. Edge intelligence keeps working even when connectivity drops. For critical infrastructure like our power grids, water supply, and transportation, this is not just a feature—it is essential for national strength and reliability. The third factor is economics. Every time we use the cloud for inference, we pay ongoing costs—often to companies outside India. On-device intelligence, on the other hand, means we invest upfront and update as needed, but the value stays with us. By aligning our semiconductor ambitions with edge AI, we create a powerful ecosystem: chips made in India, models improved locally, and devices built by our own teams. This is how we achieve real independence and drive long-term growth.
Q: How can Edge AI democratize AI adoption by bringing intelligence closer to where data is generated?
A: Cloud-based AI demands high-speed internet, ongoing subscriptions, and complex data management. This leaves out a huge part of India’s economy: small manufacturers, rural healthcare providers, farmers, and businesses in Tier-2 and Tier-3 cities. Edge AI changes the game. Affordable, microcontroller-based solutions work offline, cut out per-inference fees, and keep sensitive data local. This is how we bring AI to everyone. Small foundries can now run visual inspections without cloud contracts. Primary health centers can deliver diagnostics on-site. Farmers can control pumps without relying on connectivity. Over the last two decades in embedded software, I have seen firsthand that technology becomes truly accessible when it is built into affordable hardware. As AI models get smaller and fit into MCUs and NPUs, embedding AI in the devices people already use is the most practical and inclusive way forward for India.
Q: What is the role of embedded systems, edge computing, and intelligent devices in enabling scalable, real-world AI deployment?
A: Real value comes when intelligence is built into products that ship, survive in the field, and run for years on limited power and memory. Models that stay in demos or data centers never make an impact. The real work is delivering solutions that last. This is the heart of embedded systems. Moving from research to real products means optimizing for the actual chip, integrating with sensors and connectivity, and making sure it works in the real world, not just the lab. Many AI projects get stuck here. The challenge is not building the model, but getting it deployed and keeping it running. This is exactly why we launched ModelNova as an independent company from embedUR. The industry is moving toward production-ready models that are proven on real silicon, so device makers are not left to bridge the gap alone. India has the talent and the drive in embedded engineering and electronics manufacturing. We have a real opportunity to lead the world in this deployment layer.
Q: How can national initiatives such as the IndiaAI Mission, Digital India, and Make in India converge to strengthen India’s AI ecosystem?
A: Each of these initiatives tackles a critical layer. IndiaAI Mission is driving our compute capacity, datasets, and model development. Digital India has laid the foundation with robust connectivity and digital public infrastructure. Make in India, alongside the India Semiconductor Mission, is powering up our hardware manufacturing. Where these efforts come together is at the intelligent device—the next frontier for India’s leadership in technology. Building an AI-enabled device entirely in India is within our reach: silicon designed and produced here, models trained on Indian data using our own compute, and seamless deployment and updates over our digital infrastructure. We have already proven our ability to deliver this kind of end-to-end innovation—UPI is a world-class example of how public infrastructure and private ingenuity can leapfrog outdated systems. Now, we have the opportunity to do the same for edge AI. The key is focused policy that brings it all together: incentives for manufacturing AI-capable devices, standards and certification tailored for Indian conditions and languages, and procurement that demands on-device intelligence for public projects in health, agriculture, and infrastructure. When these missions align at the device level, India moves from being a consumer of AI to becoming a global exporter of AI innovation.
Q: What do Indian enterprises across manufacturing, healthcare, and energy need to do to translate AI innovation into measurable business impact?
A: Start with the business outcome that truly matters. In my experience, technology earns its place only when it solves a real problem and drives results for the business. Too often, I have seen AI projects lose focus chasing the latest trends. The projects that succeed are the ones that zero in on the numbers that matter: reducing defects, cutting downtime, increasing patient throughput, or minimizing transmission losses. Identify the key metric, then work backward to the simplest AI solution that delivers measurable results. In manufacturing, I always advise starting with a single line. Focus on visual inspection or predictive maintenance for your most critical asset, and track progress against a clear baseline. Once you prove value, scale it across the plant. In healthcare, true impact comes from point-of-care intelligence that integrates directly into clinical workflows, operates on-device, protects patient privacy, and keeps running even when connectivity drops. In energy, drive intelligence to the grid edge, where losses occur and where waiting for the cloud is simply not an option. There are three disciplines that turn pilots into real business impact. First, deploy intelligence where the data is generated. Latency, connectivity, and data privacy challenges often derail cloud-only projects, especially in the realities of the Indian market. Second, treat AI as a product-engineering initiative, not just a data science experiment. Invest in integration, validation, and lifecycle maintenance, because that is where the real value is created. Third, measure relentlessly against your original baseline. If AI cannot deliver results you can show in a quarterly review, it is just a science project. If it does, it becomes a line item your CFO will champion.
