India’s AI leap: from surveillance to foresight

Artificial Intelligence (AI) in India is breaking out of its surveillance shell. No longer just a passive observer of events, it is evolving into a real-time decision-maker: detecting threats, guiding autonomous vehicles, and even decoding shopping behavior. The transformation is bold, fast, and deeply intertwined with edge computing, national constraints, and practical use cases that go beyond hype.

Leading this shift is Manmeet Singh, Senior Director and India Business Head for Automotive, IoT, Connectivity & Broadband, Qualcomm India. From AI-powered surveillance to autonomous cars and voice-first vehicle interactions, the future he outlines is no longer speculative. It is arriving, and it is being built on Indian roads, retail aisles, and industrial shop floors.

AI is shifting from cloud to edge

At the heart of this evolution lies one architectural change: the move to edge inference. Traditional AI depended heavily on cloud systems. But latency, privacy, and bandwidth limitations in India are forcing the AI engine to run closer to where the data is generated—on-device or on-premise.

This architectural shift has immediate implications across product categories. In smartphones, integrated Neural Processing Units (NPUs) are enabling large language models (LLMs) to run directly on the device. The same is happening in PCs, XR devices, and, more importantly, in AI surveillance systems that must act, not just record.

Edge-based inference eliminates latency. In applications like extended reality (XR), where delays can ruin immersion, this is non-negotiable. It also safeguards privacy, particularly important in corporate and public sector environments that do not want sensitive data hitting the cloud.

From perception to prediction in surveillance

Surveillance in India is transforming from post-event analysis to real-time threat detection. This is not just a software story; it is about building entire platforms that blend edge processing, predictive AI, and contextual decision-making.

Insight, Qualcomm’s end-to-end AI platform, embodies this transition. Designed to analyze video feeds on the edge, it flags anomalies as they happen, not hours later. Whether it is detecting a weapon, tracking a suspicious person across multiple cameras, or triggering alarms when a fire is detected, the emphasis is on action, not observation.

For retail, AI is decoding footfall and customer movement in physical stores. It analyzes which aisles convert walk-ins into sales and tracks user behavior just as efficiently as e-commerce sites do. AI cameras, paired with facial recognition and people-counting algorithms, are becoming silent sales strategists.

Accuracy: the real challenge

AI’s promise hinges on precision. And in safety-critical domains, “almost right” is unacceptable. A 60 to 65 percent detection rate is not a success story—it is a failure.

India’s complex environments make this accuracy even harder to achieve. AI models trained on Western datasets often fail when deployed here. What counts as an “unattended suitcase” in a U.S. airport may look completely different in a crowded Indian terminal. The model must be retrained, contextually localized, and tested at scale. Only then can it reach the 95 to 99 percent reliability needed for real-world impact.

Hardware matters too. Low-quality deployment dilutes AI’s effectiveness. Qualcomm’s industrial AI boxes—scalable from 4 to 16 video streams and powered by NPUs with up to 100 TOPS—offer options that can match the scale and complexity of each deployment. No more one-size-fits-all.

India’s AI constraints are its design drivers

AI systems in India cannot be designed like those in bandwidth-rich geographies. Data privacy concerns, patchy cloud connectivity, and uneven IT infrastructure mean edge computing is not just preferred; it is required.

Enter on-premise AI acceleration. With compute boxes that replicate cloud functionality locally, organizations can train and run models without exposing sensitive data. This is especially vital for government setups and enterprises wary of uploading surveillance footage or biometric information.

Then there is cost. Edge-based inference lowers power bills, reduces latency, and removes the dependency on high-bandwidth cloud links. It is efficient by design because it has to be.

Safety at scale: mobility, public spaces, and more

The next frontier is AI that understands context and adapts on the fly. In public spaces like Kumbh Melas or cricket stadiums, human monitoring simply does not scale. AI can step in: detecting threats, tracking movement, and alerting authorities instantly.

By 2026, edge AI will handle biometric workloads, facial detection, and voice-controlled interfaces—all offline. This is more than convenience; it is about trust. Users prefer local processing of personal data, even if cloud providers promise encryption.

In vehicles, AI is turning dashboards into voice-first interfaces. Instead of navigating confusing menus, users can say, “It’s a sunny day,” and the car adapts the lighting and music. The manual becomes a conversational assistant trained on the vehicle’s data and preferences.

The road to Level 3 and beyond

India’s roads may be chaotic, but they are also the ultimate training ground for AI in automotive. Level 2 and Level 2+ Advanced Driver Assistance Systems (ADAS) are already here. Highway autonomy is next, followed by urban self-driving, possibly within a couple of years.

But India poses unique challenges: irregular lane markings, inconsistent driver behavior, and low rule compliance. That is where self-learning models come in. Instead of relying on a rigid rulebook, the car watches how humans behave and mirrors it, adapting to the chaos instead of fighting it.

Voice assistants, once a novelty, are maturing fast. Small language models that support regional Indian languages will make voice control not just possible but natural for users across demographics. English fluency is no longer a prerequisite for interacting with high-end car features.

Future-ready retail and industrial intelligence

Security will remain the most immediate use case for AI in India. But the long-term value lies in analytics, especially in retail and manufacturing.

In physical retail, AI can now track not just how many people entered a store, but what they looked at, where they spent time, and what they ended up buying. Over time, this will influence store layouts, product placements, and marketing strategies, making offline shopping as measurable as online.

In manufacturing, AI will spot defects, monitor operations, and optimize workflows. The real impact will not be just about efficiency; it will be about learning—systems that improve themselves by observing how humans work, then suggesting smarter ways to do the same task.

What matters most now

India’s AI future will not be defined by flashy demos. It will be judged by systems that actually work: reliably, repeatedly, and in real time. Whether it is a fire alarm triggered early, a car that understands Tamil, or a surveillance model that detects danger before disaster, the bar is high.

But as edge computing becomes the norm and accuracy crosses the 95 percent threshold, India’s AI ecosystem will not just be ready; it will be leading. The nation’s complexity, once seen as a limitation, is now the proving ground for global-grade AI.

Because if it works here, it can work anywhere.