The US currently leads the world in two critical military technologies — satellite-based low-latency internet and AI. The former gives its weapons global reach, while the latter provides unprecedented accuracy. Together, they could enable the US to maintain its military dominance across the world for decades.
In a low-key manner, the US is already flexing its Starlink-based global reach and AI-powered accuracy through the drones it is supplying to Ukraine.
Perennial Autonomy, a company founded by Eric Schmidt, former CEO of Google, has developed two drones that have put Russian forces on the back foot — the Merops Surveyor interceptor drone and the Hornet kamikaze drone. The Surveyor interceptor uses AI to bring down Russian drones, while the Hornet drone uses both Starlink and AI to wreak havoc on Russia's ability to supply its troops along the front line.
Merops AS-3 Surveyor
The Merops AS-3 Surveyor is a mobile, truck-portable counter-drone system comprising:
- Radar and electro-optical sensors for detection and tracking
- A ground control/command station
- Pneumatic or mobile launch platforms
- A fleet of Surveyor interceptor drones
The fixed-wing Surveyor interceptor was first combat-tested in Ukraine around June 2024. By late 2025, it had reportedly achieved over 1,900 intercepts. In some sectors, it is claimed to have brought down roughly 40% of Russian Geran drones. Recent reports claim 4,000 successful Russian drone interceptions.
The Surveyor is an effective interceptor on account of its greater AI-based autonomy, speed, and jam resistance. Following launch, the drone is cued and initially guided using the sensors of the Merops system. For terminal guidance, it uses onboard IR and RF sensors, as well as AI-based machine vision. It can home in on targets even when SATNAV and communication signals are jammed.
AI-based machine vision and the ability to fuse inputs from IR and RF sensors are key to the success of the Surveyor.
With its maximum speed of 280 km/h, the drone outpaces Russian Gerans.
Hornet Strike Drone
The Hornet drone can be credited with bringing the Russian offensive in Donbas to a crawl along the line of contact, and even to a complete halt in some sectors. Ukraine is also using the drone to strangulate Russia's ability to supply Crimea.
As with the Surveyor, the Hornet's success can largely be attributed to its AI-powered ability to operate effectively in the absence of SATNAV and communications.
We covered the capabilities of the Merops Surveyor and Hornet drones in an earlier post.
AI-Based Machine Vision
So far, SATNAV has been the gold standard in the precision guidance of drones, missiles, and rockets. AI-powered machine-vision-based navigation outperforms SATNAV in accuracy. More importantly, it is completely immune to electronic warfare.
However, machine vision can be spoofed — for example, by using paint schemes that make optical recognition challenging.
With increased onboard processing power, it will become difficult, perhaps impossible, to spoof AI-powered machine vision.
Global Reach
The ability to control drones and missiles capable of precision guidance globally requires a Starlink-like network. Currently, there is no alternative to Starlink.
Outplaying Emerging Powers
In the days ahead, many nations, including India, will build weapons with AI-powered machine vision. However, doing so without acquiring matching semiconductor fabrication and design capability would not allow them to exercise sovereignty over their own weapons.
Semiconductor fabrication and design technologies are likely to be tightly controlled in order to prevent challenges to US military dominance.
As an analogy, a nation with nuclear weapons technology does not share it with a nation that lacks the technology. Indeed, nations that possess nuclear weapons do their best to prevent the "have-nots" from acquiring weapons-grade fissile material.
The ability to manufacture fissile material, a key enabling technology for nuclear weapons, is tightly controlled.
Similarly, robust space-launch capability, as well as the semiconductor fabrication and design capability needed to deploy a Starlink analogue or facilitate advanced machine vision, will be tightly controlled.
The technological barriers to acquiring these capabilities are formidable. The hurdles span multiple years — perhaps multiple decades — and are rooted in physics, engineering complexity, supply chains, and capital intensity.
A low-latency global or regional broadband constellation requires thousands of satellites (Starlink has over 10,000) in low Earth orbit (LEO, ~550 km altitude), inter-satellite laser links, high-volume satellite manufacturing, and millions of user terminals with electronically steered phased-array antennas.
The number of satellites required can vary based on the architecture of the network and its intended extent of coverage. However, a true Starlink analogue would require the development of a reusable launcher.
China, the EU, and Russia have all embarked on deploying Starlink analogues, but all three have so far made limited progress. Countries like India are unlikely to be in a position to acquire such a capability over the next decade.
User Terminals
Low-latency networks use terminals featuring custom ASICs and advanced RF front-end modules (e.g., BiCMOS technology) for phased-array antennas that track fast-moving LEO satellites.
Starlink has already deployed millions of such terminals. STMicroelectronics has shipped over 5 billion RF chips for the terminals, with daily rates exceeding 5 million.
Network satellites use radiation-hardened electronics, onboard processors, and laser comms chips that require specialized semiconductor fabs, which in turn require decades of ecosystem investment.
AI-Powered Machine Vision
Effective (high-accuracy, low-latency) machine vision in drones and cruise missiles requires real-time object detection, tracking, terrain classification, sensor fusion, and autonomous navigation capability under severe size, weight, power, and cost constraints, harsh operating environments, and contested electromagnetic conditions.
Hardware-wise, machine vision relies on high-performance AI accelerators (NPUs, custom ASICs, or optimized GPUs/FPGAs) that must deliver tens to hundreds of TOPS (trillions of operations per second) for neural networks such as CNNs or lightweight transformers (e.g., YOLO variants).
Such hardware would require leading-edge nodes (7 nm, 5 nm, 3 nm, or below) for the density, speed, and energy efficiency needed to run complex models onboard without excessive power draw.
Only a handful of fabs worldwide — primarily TSMC in Taiwan, with limited capacity from Samsung and Intel — can produce these at scale and yield.
The US itself faces geopolitical and supply-chain vulnerabilities. However, it is likely working on a plan to eventually eliminate them.
Conclusion
As things stand, the US appears uniquely positioned in combining reusable launch capability with access to semiconductor fabrication and design ecosystems that facilitate global reach and high-precision strikes by drones and cruise missiles.
In the discussion above, we confined ourselves largely to drones and cruise missiles. AI and secure global communication have applications in other weapon systems as well — space-based weapons, for example.
It is time for India to take a hard look at its quest for self-reliance in weapon manufacturing. Hopefully, we are not focusing on acquiring sunset technologies, and our efforts to acquire semiconductor fabrication and design technologies will be pursued vigorously.
Copyright © Vijainder K Thakur. First published on Thumkar.
