The AI Military Complex: Companies, India and the Future of Defence Technology

Defence AI is the application of machine learning, computer vision, natural-language processing and related techniques to military functions — intelligence analysis, logistics, cybersecurity, maintenance, training and decision support. It is a broad, mostly unglamorous category: the great majority of publicly documented defence AI work looks less like a science-fiction weapon and more like an inventory-management upgrade, a radar-image classifier, or a chatbot that helps a logistics officer draft a supply request faster.

The single most important distinction in this field is between AI-assisted and autonomous systems. An AI-assisted system processes information and presents recommendations, but a human makes the consequential decision — this describes the overwhelming majority of fielded military AI, from predictive-maintenance dashboards to intelligence-fusion tools. An autonomous system, by contrast, can take a specific class of action — most sensitively, selecting and engaging a target — without a human approving that specific instance in real time. The U.S. Department of Defense’s governing policy on this, Directive 3000.09, does not ban autonomous weapon systems, but it requires that “autonomous and semi-autonomous weapon systems…be designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force” — a requirement for governed autonomy, not a requirement that a human person approve every individual engagement.

Military decision support is the category most defence AI investment actually targets: tools that fuse sensor feeds, flag anomalies, model logistics scenarios or summarise intelligence reporting for a human analyst or commander who retains decision authority. This is the least visually dramatic and most heavily funded category of defence AI, and it is where India’s public programme — from DRDO research to iDEX-funded startups — has concentrated its documented, non-classified output.

Dual-use technology describes tools with both civilian and military applications — computer vision, natural-language processing, autonomous navigation and predictive analytics are dual-use almost by definition, which is part of why a drone-delivery startup and a defence contractor can draw on overlapping underlying technology, and why India’s National Strategy for Artificial Intelligence, though not defence-focused in its five headline priority sectors, still shapes the broader talent and research base that defence AI draws on.

India’s AI ambitions in defence sit inside a larger self-reliance (“Atmanirbharta”) push: reducing dependence on imported defence hardware by building indigenous design, manufacturing and software capacity, with AI treated as one enabling layer among several — alongside indigenous engines, semiconductors and precision manufacturing — rather than as a standalone national objective. Official statements consistently frame AI as augmenting existing modernisation goals, not as a separate strategic pillar with its own budget line comparable to, say, aircraft carrier procurement.

Why AI is reshaping modern defence planning comes down to data volume: satellite constellations, distributed sensors, cyber-network telemetry and open-source intelligence now generate more material than human analysts can review unaided. The documented, unclassified case for military AI is fundamentally a case about triage and speed — helping a smaller number of trained people manage a much larger volume of information — rather than a case about replacing trained judgment, which every major published policy framework covered in this guide explicitly preserves as a human function.

The “global AI race” framing that dominates popular coverage of this topic is real in the sense that multiple governments — the United States, China, India, Israel and NATO member states among them — are visibly and simultaneously investing in defence AI, each publishing its own strategy documents and standing up its own institutions, documented throughout this guide’s timeline. It is misleading, though, when it implies a single finish line or a single dominant capability that one country will “win.” The public record instead shows several parallel, differently structured national programmes — India’s DPSU-anchored, startup-widening model; the US’s venture-backed contractor model; China’s documented civil-military fusion model — each optimising for its own institutional strengths and constraints, rather than converging on one template.

DRDO (the Defence Research and Development Organisation) remains India’s primary in-house defence research body, operating a network of specialised laboratories, including work on AI and robotics concentrated in and around Bengaluru. DRDO’s role is fundamentally different from iDEX’s: DRDO conducts government-funded research and develops systems directly, while iDEX funds external startups and MSMEs to build solutions the services then evaluate and, if successful, procure.

Public-sector undertakings — principally Bharat Electronics Limited (BEL), Hindustan Aeronautics Limited (HAL), and Bharat Dynamics Limited (BDL) — form the manufacturing backbone that converts DRDO and private-sector AI research into fielded hardware and software at scale. BEL and HAL are also, notably, iDEX’s founding institutions through the Defence Innovation Organisation, giving public-sector manufacturers a direct institutional stake in the startup ecosystem rather than treating it as a purely external supplier base.

Private-sector majors have expanded their defence AI investment substantially since the late 2010s. Tata Advanced Systems Limited (TASL) and Larsen & Toubro (L&T) Defence are the most established, working across autonomous platforms, sensors and command-and-control software; Adani Defence & Aerospace and Bharat Forge’s Kalyani Strategic Systems have both expanded into unmanned systems and precision manufacturing relevant to AI-enabled platforms. These companies typically compete for and fulfil large-value defence contracts rather than operate on the startup-funding model iDEX uses.

Startups and listed defence-tech firms are the fastest-growing and most numerous part of India’s defence AI ecosystem, tracked closely by industry analysts. Zen Technologies (listed 2015) has built a specialisation in counter-drone systems combining radar, electro-optics and electronic countermeasures. ideaForge (listed 2023) supplies surveillance drones to Indian security forces. NewSpace Research and Technologies develops swarm-combat and drone-coordination systems and raised a reported $33 million Series B round. Sagar Defence Engineering focuses on maritime and naval uncrewed systems, with a reported $25.4 million Series A round. Tonbo Imaging (optics and night-vision) and Paras Defence (optics and electronics) round out the sensor-hardware side, while Raphe mPhibr’s reported $100 million Series B was, per Tracxn’s 2025 tracking, the largest single funding round in the sector that year. Data Patterns and Astra Microwave Products (listed 2004) supply defence electronics that increasingly incorporate AI-assisted signal processing.

Per Tracxn’s September 2025 military-tech report, India’s defence-tech startups raised approximately $192 million year-to-date in 2025, with Bengaluru the second-largest funding hub after the Delhi-NCR region, reflecting the city’s concentration of both DRDO facilities and private AI talent.

Industry bodies also shape this ecosystem without directly building or fielding systems themselves. NASSCOM, India’s national technology-industry association, has published policy commentary and convened industry discussion on defence and dual-use AI as part of its broader national AI advocacy work, positioning itself as a bridge between the commercial software industry and defence procurement rather than as a defence contractor in its own right. The Indian Army, Indian Navy and Indian Air Force each engage this ecosystem through their own service-specific requirements and trials — the naval and maritime-domain-awareness applications discussed later in this guide are a direct example of the Indian Navy’s role as an end-user setting requirements that DRDO, DPSUs and startups like Sagar Defence Engineering compete to meet.

The talent pipeline behind this ecosystem is worth naming explicitly, because it explains why Bengaluru and the Delhi-NCR region dominate the funding map above. India’s IITs and other engineering institutes feed both DRDO’s research cadre and the private and startup sides of this ecosystem through direct recruitment, faculty consulting arrangements, and, increasingly, iDEX’s own academia-facing funding stream, which extends the programme’s reach beyond commercial startups to university labs working on early-stage defence-relevant AI research.

Behind the companies and government agencies named throughout this guide sits a smaller, less publicised layer of research institutions that trains the engineers and produces the early-stage research those organisations later build on. DRDO’s own Centre for Artificial Intelligence and Robotics (CAIR), based in Bengaluru, is the laboratory most consistently named in official statements as the organisation’s dedicated AI and robotics research unit, working across areas including robotics, AI and information systems for defence applications. The Defence Institute of Advanced Technology (DIAT) in Pune, a DRDO-affiliated deemed university, trains postgraduate engineers specifically for defence research roles, including in AI-adjacent disciplines. India’s IITs contribute through a mix of direct DRDO collaboration, iDEX’s academia-facing funding stream, and the ordinary flow of graduates into both DPSU and private-sector defence engineering roles documented in the companies table above.

This guide treats this research layer distinctly from the companies and government bodies covered elsewhere: research institutions publish papers and train people; they do not, on their own, field operational systems. That handoff — from a DRDO lab or a university research group to a fielded capability — is exactly what the procurement lifecycle described later in this guide is designed to formalise.

Naval systems use AI primarily for sonar-signal classification, hull and engine predictive maintenance, and maritime-domain awareness — fusing radar, satellite and automatic-identification-system data to flag unusual vessel behaviour. Sagar Defence Engineering’s uncrewed maritime systems and DRDO’s naval research programmes are India’s most publicly documented efforts in this domain.

Air defence applications concentrate on radar-signal processing, threat classification and fire-control decision support — helping human operators prioritise which of many simultaneous tracks warrant attention, rather than autonomously deciding to engage. Zen Technologies’ counter-drone systems, which fuse radar, electro-optic and electronic-countermeasure data, are a clear, documented Indian example of this AI-assisted triage model.

Land systems AI work spans uncrewed ground vehicles, robotic logistics carriers and soldier-worn sensor systems, an area DRDO has researched since well before the 2018 policy formalisation, building on decades of interest in automation and robotics for reducing soldier exposure in high-risk roles.

Cybersecurity is one of the most mature and least controversial applications of defence AI: machine-learning-based anomaly detection to flag unusual network activity faster than manual monitoring allows, protecting both military networks and, increasingly, critical civilian infrastructure with military-relevant dependencies.

Logistics and predictive maintenance represent, by most industry accounts, the single largest and least visible category of fielded military AI: digital-twin models of aircraft engines, ships and vehicle fleets that predict component failures before they occur, and route-optimisation tools that manage the enormous logistics burden of moving supplies to forward units.

Satellite intelligence and border surveillance apply computer vision to satellite and sensor imagery to detect changes — new construction, vehicle movements, unusual activity patterns — along borders and in areas of strategic interest, an application area that draws directly on the same ImageNet-era computer-vision techniques discussed earlier in this guide’s timeline.

Command and control is the domain where “decision support” as a category is most concentrated: software that fuses intelligence, logistics and operational data into a single picture for commanders, explicitly designed — per every major published framework this guide cites — to inform rather than replace human command judgment.

Across all seven of these domains, the pattern this guide’s timeline established repeats: the technology that eventually reaches a fielded military system usually began as unclassified, often civilian, research — computer vision from academic labs, predictive-maintenance techniques from commercial aviation and manufacturing, anomaly detection from civilian cybersecurity — adapted and hardened for military reliability requirements rather than invented from scratch inside a classified programme. That adaptation process, not a single dramatic invention, is what most of the procurement lifecycle described later in this guide is actually built to manage.

Public discussion of defence AI regularly collapses several distinct spectrums into one — treating “has AI” and “is autonomous” as synonyms, or assuming a country’s published strategy document describes what it has actually fielded. The four comparisons below are drawn directly from the definitions and sourced facts established earlier in this guide, laid out side by side specifically to make those collapses harder to make by accident.

Human oversight is the consistent, stated principle across every major published military AI framework this guide has reviewed — not a single country’s talking point. The US Department of Defense’s five principles (2020), NATO’s six Principles of Responsible Use (2021, revised 2024), and India’s institutional design through DAIC all frame AI as augmenting, not replacing, human command judgment, even as they differ in how formally that commitment is codified.

The UN’s regulatory process runs through the Convention on Certain Conventional Weapons (CCW), where a Group of Governmental Experts on lethal autonomous weapons systems has met periodically since the mid-2010s. Its current mandate, adopted in 2023, tasks the group with formulating, by consensus, elements of an instrument on LAWS, reporting to the CCW’s Seventh Review Conference in 2026. As of late 2025, negotiations remained deadlocked despite a joint ICRC-UN Secretary-General call for a legally binding instrument by the end of 2026; further sessions are scheduled for March and August-September 2026.

The UN General Assembly has separately weighed in through non-binding resolutions — most notably Resolution 78/241, adopted December 22, 2023, by a vote of 152 in favour, 4 against and 11 abstaining. India was among the four states voting against it, a position its government has explained not as opposition to regulation itself but as a preference for pursuing a non-binding political declaration through the existing CCW process, built on guiding principles the CCW adopted by consensus in 2019, rather than negotiating a new legally binding treaty.

What this means in practice is that, as of this update, no country is bound by an international treaty specifically restricting autonomous weapons; governance instead rests on national policy (like the US’s DoD Directive 3000.09), alliance-level frameworks (like NATO’s principles), and non-binding UN guidance — a patchwork this guide describes precisely rather than characterising as either adequate or inadequate, since that judgment is contested among the states and institutions involved.

Accountability and bias are the two concerns that recur most consistently across every published framework this guide has reviewed, independent of country. The US DoD’s “Equitable” principle commits to minimising unintended bias in AI capabilities; NATO’s framework names “Bias Mitigation” as one of six core principles rather than a peripheral concern. In practice, this means testing AI systems against a range of operating conditions before fielding them, and maintaining traceable records of how a system was trained and validated — documentation requirements that, per published policy, apply regardless of whether the system in question is autonomous or purely decision-support.

Industry self-regulation has also played a role distinct from government policy. Several AI companies with defence contracts, including firms named earlier in this guide, have published their own internal principles governing which military applications they will and will not pursue — a development independent analysts trace directly back to the 2018 Project Maven controversy, when Google’s employee backlash forced the broader tech industry to publicly articulate where it drew its own lines, well before most governments had done the same.

Set against the sweep of this guide’s timeline — from a room-sized air-defence computer in 1958 to iDEX-funded startups pitching swarm-coordination software in 2026 — the throughline is not a story of machines seizing control of warfare. It is a story of a specific, traceable, heavily documented set of institutions — DRDO, NITI Aayog, iDEX, the Defence AI Council, the Pentagon’s CDAO, NATO’s policy office, the UN’s CCW forum — each publishing, negotiating and revising a narrower and narrower set of rules for how AI enters military decision-making, at almost exactly the pace the underlying technology has matured.

The engineer annotating radar data in Bengaluru and the planner reviewing a logistics tool at the opening of this guide are not exceptions to how defence AI works; they are the norm, replicated across DRDO labs, DPSU engineering floors, private contractors and startup offices in India and allied countries alike. Most publicly documented military AI programmes — India’s included — focus on augmenting human decision-making, improving logistics efficiency and strengthening cyber resilience, not on removing human judgment from the use of force.

Readers evaluating any specific claim about military AI — a company’s product demo, a think-tank’s forecast, a headline about “autonomous armies” — are best served by returning to the official policy documents, defence research and peer-reviewed or institutional analysis this guide has drawn on throughout: NITI Aayog’s published strategy, the Ministry of Defence’s task-force report and iDEX documentation, DoD Directive 3000.09, NATO’s AI Strategy, and the ongoing, still-unresolved UN CCW process. None of those sources describes an AI-run military. All of them describe, in granular and often bureaucratic detail, an effort to keep responsible governance, transparency and human oversight at the center of how defence AI is built — and that, more than any single system or company, is the actual state of the field as of this update.