How AI in Indian Defence Went From the Lab to the Battlefield

For nearly a decade, artificial intelligence in the Indian armed forces was a story told in the future tense: pilot projects, proofs of concept, demonstrations tucked into the sidelines of military exercises. That story has changed. AI in Indian defence has now moved through a genuine inflection point, with systems once confined to trial ranges being fielded in live conflict conditions. But how far has that shift gone, and does India's progress in Indian armed forces AI adoption hold up against the world's leading military powers? Two industry leaders building AI systems for India's forces offer strikingly different vantage points: one cautious and systems-driven, the other emboldened by a recent operational milestone. Together, their perspectives paint a fuller picture than either view alone.

Two Readings of the Same Curve

The first point of divergence is over how much of India's AI push has actually left the laboratory. One assessment holds that adoption is "real, but uneven," noting that the Ministry of Defence reported over 140 AI deployments in forward areas as early as December 2021, with a 2025 review describing further experiments on autonomous weapons and an induction roadmap. This expert cautions against a single deployment percentage, arguing that a demonstration, a procurement contract, and sustained operational capability are separate milestones often conflated in public discourse. Adoption, in this reading, should be judged by what happens after installation, whether personnel use a system routinely and whether it survives real operational constraints.

The second perspective frames the shift more emphatically, pointing to a specific inflection: the last 18 months, and Operation Sindoor in particular. AI-enabled air defence and counter-drone systems, in this account, were not merely trialled but operationally deployed during live conflict, and performed well. That is a categorically different maturity signal than a lab result. Even here, the picture is not uniform: counter-drone work and air-defence cueing moved fastest because the threat forced the pace, while autonomous decision support and deep AI integration into command and control remain mid-transition.

Read together, both experts agree on the underlying shape of the problem even while disagreeing on how far along the curve India sits. Adoption is real, it is patchy, and the frontline threat environment, not policy alone, has been the biggest accelerant.

Where the Priorities Diverge

Asked which applications matter most, the two responses split along an interesting axis: breadth versus threat-specificity. One view prioritises applications that protect personnel and widen situational awareness: ISR, cyber defence, counter-drone detection, and equally, "unglamorous" applications like predictive maintenance, logistics and training simulations that prevent supply shortfalls or downtime. The long-term bet is on autonomous systems and human-machine teaming that expand coverage while reducing personnel exposure, provided the full lifecycle economics, not just unit cost, are accounted for.

The other view puts nearly all its weight on the threat already materialised: cheap, swarming, GPS-denied drones, described as the most democratised weapon in modern warfare. Detecting, classifying and responding to swarms within seconds is framed as a machine-speed problem no human operator can handle alone. Four areas are flagged as decisive for the next decade: offensive and defensive autonomous swarms, ISR and target recognition, electronic warfare (particularly signal takeover of hostile drone links), and predictive maintenance. Of these, swarm-on-swarm engagement is singled out as the biggest shift, moving warfare from platform-versus-platform to system-versus-system.

Measuring Up to the US and China

On India's standing relative to the US and China, both experts resist easy triumphalism, but for different reasons. One separates the maturity of a country's broader technology ecosystem from the readiness of individual military systems, warning that the US's commercial-AI integration or China's military-civil fusion should not be mistaken for proof that every advertised capability is operational. This expert does not claim India is ahead overall, prioritising gaps in representative defence data, dependable compute infrastructure, platform integration, and, notably, intellectual property ownership, arguing sovereignty should mean the practical ability to operate, inspect and sustain systems even when external access is disrupted.

Also Read:

AI in Healthcare: The Shift From Reactive Treatment to Predictive, Patient-Centric Care

The other account is more willing to call out a real gap in compute, data volume and semiconductor depth versus the US and China, while arguing it is closing faster than expected. The evidence offered is Operation Sindoor itself, characterised as a live comparison of Indian systems against Chinese-origin platforms fielded by Pakistan, in which India's layered air defence and counter-drone response is said to have held up well. The residual gap, in this telling, is less about individual point solutions, where India competes well, and more about ecosystem depth: compute infrastructure, large-scale defence-specific datasets, and the industrial base to produce processors and sensors at volume, some of which are still imported.

Also Read:

IIT Madras Pravartak Marks Five Years of Driving AI, Deep-Tech Innovation and National Technology Capabilities

Do the Institutions Work?

Both experts credit India's institutional scaffolding, namely the Defence AI Council, the Defence AI Project Agency, DRDO's research programmes and iDEX, with meaningfully changing conditions for innovators, while flagging the same underlying friction: speed. One frames iDEX's progress in concrete terms, noting that by February 2026 the government had reported 45 procurement contracts worth roughly ₹2,326 crore, while cautioning that such figures describe iDEX technologies broadly rather than confirm how much AI is operationally deployed. The real test, in this view, is whether these programmes produce durable suppliers, tracked by the time from successful trials to repeat orders and sustained use.

"Trust has to be earned through observable performance. Operators should understand what an output is based on, where uncertainty remains and when to intervene. I would make that usability and accountability part of the engineering requirements from the beginning"

Nandagopal P, CEO, Asymmetri

The other expert points to a structural shift: DAP 2026's move from a "buyer" posture to a "builder" posture, requiring earlier proof of technology and manufacturing readiness, as evidence that accountability is being pulled forward. The push here is for the cycle time between a validated trial and a signed contract to keep compressing, since in AI-enabled warfare "six months can be the difference between relevant and obsolete."

The Shared Bottlenecks

It is on the biggest obstacles to scaling that the two views converge most closely. Both flag data as a fundamental constraint: representative, defence-specific data that can only be built through real field deployment, not generic commercial datasets. Both name compute and hardware sovereignty, particularly imported processors, as a genuine vulnerability even in otherwise indigenous systems. Both stress that testing must go beyond controlled trials to cover terrain, weather, GPS denial and adversarial conditions. And both converge on the same closing point: trust. A system highly accurate in a lab still has to earn the confidence of a commander deciding under fire, built only through transparent performance and repeated proof under real conditions.

A Doctrine Being Rewritten in Real Time

Looking five to ten years out, both experts expect military AI in India to reshape doctrine, not just support it. That means deeper embedding in logistics, intelligence, planning and coordination between crewed and uncrewed platforms, and a shift toward networked, software-defined forces built on many affordable autonomous systems rather than a few expensive platforms.

"Swarms will become a standard formation, not a novelty... Human machine teaming will become the operating model, not an experiment... That balance, more than the raw technology, is what will define whether India's AI enabled doctrine is seen as credible and responsible on the world stage"

Kiran Raju, CEO, Indrajaal

Neither expert frames this as a march toward full machine autonomy over lethal decisions. Both insist meaningful human control, meaning understandable information, clear authority, and a real opportunity to intervene, must be engineered in from the start, not bolted on afterward. For a country pursuing autonomous defence systems India can trust and sustain on its own terms, that combination of ambition and restraint may be the truer measure of progress than any single deployment count.