The Hook
Big news out of Lucknow: a boot-strapped defence start-up just demoed an AI drone that costs less than a mid-range sedan. The pitch? A 1,000-unit autonomous swarm that can loiter, hunt, and strike—and still leave change from the capital budget.
News Breakdown
Divyastra, built by NextGen Aeronautics Pvt. Ltd. (incubated at IIT-Kanpur), is a 3.2 kg fixed-wing quadcopter hybrid powered by a Qualcomm RB5 AI board and a custom Edge-TPU that ingests 4K EO + LWIR feeds at 60 fps. The drone ships with two operational modes:
- Loiter-on-Edge: 42-min endurance at 65 km/h; neural net tracks 256 objects simultaneously.
- Kamikaze: 1.2 kg cumulative warhead, dive speed 170 km/h, <2 m CEP.
Unit price: ₹18.3 lakh (EXW), 38 % cheaper than the imported switch-blade equivalent, according to the MoD’s Price Benchmarking Unit.
Key Specifications
- AI chip: 8 TOPS INT8, 5 nm process
- Range: 12 km LOS, mesh-relay to 40 km
- Swarm logic: gossip-protocol sync; 1 pilot supervises 50 drones
- Weather rating: IP54, 30 °C to 55 °C
Expert Call-out
“The real breakthrough isn’t the airframe—it’s the ₹1.7 lakh per-unit AI stack that can be hot-swapped into any UAV,” says Air Vice Marshal (retd.) A. K. Sahni, now an advisor to DRDO’s Aeronautics cluster. “If the Army accepts Divyastra at scale, India becomes the second country after Israel to field a certified, low-cost, AI-driven loitering munition.”
The NextCore Edge
Our internal analysis at NextCore suggests the MoD will green-light a rapid-induction order of 2,500 units before FY27 ends—double the publicly floated number. What the mainstream media is missing is Hindustan Aeronautics Ltd.’s quiet work on a twin-bay ALH pylon that can air-launch 32 Divyastras in under 90 seconds, effectively turning a utility chopper into a stand-off swarm carrier. The programme is codenamed Project Kookaburra and could flip the cost-curve for border patrolling along the LAC.
Tech Analysis
The drone’s Edge-TPU runs a pruned Yolo-v9 network at 8-bit precision, delivering 12 fps/W—a 3× efficiency gain over Google Coral. More critically, the Qualcomm board’s Hexagon DSP hosts a federated-learning routine: after each mission, flight logs are encrypted, split into shards, and uploaded to a DEFCON-isolated server in Hyderabad. The aggregated model is then pushed back to the fleet, giving every drone the collective experience of the swarm without ever exposing raw tactical data.
This architecture mirrors the shift we saw in AI scheduling systems where privacy-preserving learning trumps raw centralization.
Risk Check
Yet three red flags remain:
- RF Spectrum Congestion: Divyastra hops across ISM 2.4 GHz + 5.8 GHz, bands already crowded by consumer Wi-Fi along the Punjab border.
- Thermal Signature: The RB5 board idles at 42 °C; in Rajasthan summers, the drone’s IR visibility spikes 18 %.
- Explosive Payload: The shaped charge uses CL-20 imported from Germany—supply security is not guaranteed under an ITAR squeeze.
Pro Tip
If you’re a component supplier, target the ₹9,000 optical gimbal and the ₹4,500 carbon-fiber wing spar. Both are single-sourced right now, and the MoD’s Atmanirbhar auditors are pushing for dual-vendor lock-ins within 18 months.
External Validation
Industry Insights: #IndustrialTech #HardwareEngineering #NextCore #SmartManufacturing #TechAnalysis
Bringing you the latest in technology and innovation.