If "AI" got into the Replicator program, someone is gold-plating it. It's supposed to be about US$500 to US$1000 drones. What's coming out are US$20,000 drones.
[1] https://www.economist.com/united-states/2025/01/05/the-us-ar...
With the right combination of hardware and software, drones could become an incredible general purpose weapons including anti-tank, loitering and mine-like anti-personnel, RF missile, and so on.
$500-1000 drones might be useful in saturation attacks but most of that cost will be eaten up by the warhead. The drones won’t be very useful unless they’re hunter seeker style with small charges and a lot of intelligence (just like the $20k drone) for target selection. I don’t think US doctrine can support them otherwise (the vehicle to deploy them will cost 1000-10000x anyway). These classes of drones work much better for a country like Ukraine in a total war on home soil.
with deployment of cheap point defense systems - like cheap radar (Starlink phased array is $500 or so) plus shotgun (Russians currently do "soldier with shotgun") - on every tank, IFV, truck, etc. only stealth (not an option in those price ranges really) or saturation will be the way to go. Single $20K drone would be just wasted by such point defense. To the point defense add automated predator/interceptor drones patrolling against the attack drones. Thus saturation with large autonomous hives with drones of various types - coordinators, predators, and actual attack drones.
The whole problem with cheap drone swarms is that their range is shit and whatever you use to deploy them will just inherit the existing weaknesses - whether that’s cost, logistics, or payload.
I think saturation attacks are only going to be useful against static targets with large numbers of squishy meatbag targets as a cheaper way of taking out soldiers as measured in $/life, where a (maybe very expensive) plane drops a payload of small hunter seeker drones that target every human-looking IR source in the vicinity. I think that could be a lot cheaper per kill than dropping hundred pound bombs.
that is slow. Well under RPG speeds that current active defense systems (like Trophy) easily catch. And evasion on such a slow speed isn't a real evasion - at best 1m shift in 0.01 seconds - even cheap radar would easily track and would take it out with shotgun or machine gun.
> proper missile defense system or CIWS
that resolves, tracks and targets 30 cm diameter missile moving at transsonic speed at 2-4km. The point defense i'm talking about would need to resolve and target a 30 cm target at 10-20m distance (the tank or APC can take a drone warhead explosion at 5m distance) which is orders or magnitude cheaper/easier (you can even put a sub-$100 car radar or LIDAR onto a rotating plate similar like on Waymo car and whenever it picks something moving toward you at a distance less than 20m - just shoot in that general direction. Yes, infantry hates being around tanks with active defense in combat as shrapnel from the ADS hits the infantry, frequent story with USSR army back then in Afghanistan for example)
>Add a small rocket motor to the $20k drone for final approach and it’d probably top out at 400-500mph and it can use propellers to “jump around” randomly in its flight path so targeting it is nigh impossible.
Now you're trying to invent a jumping around RPG or Javelin missile. I agree that such development will happen too (though "jumping" at 200m/s is petty hard), yet it will not be able to always jump out of the shotgun grape shot cloud at that 10m distance. So, yes, there will be race between attack and defense, and my point is that single drone approach will be left behind pretty much from the start.
> cheap drone swarms is that their range is shit
50-80km communication and flying range is what we have already today. Communication - while staying cheap - can be extended further by retranslator drones (and for autonomous/AI drones the communication is much less of an issue - for example no need for high quality video, if at all), and electric drone-plane (not quad) can do up to 200km even today.
>I think saturation attacks are only going to be useful against static targets
I see quite opposite. Modern air-defense shoots down all single targets with high probability (except for stealth, which like $4M for one Storm Shadow stealth cruise missile). On the other hand - Pantsir-S1 can target simultaneously only 4 targets. A battery of S-300/400 can shoot only 12 missiles before reloading. Trophy and the likes easily shoots down one incoming targets and takes half-a-second or similar to get ready for the next target. And improving the simultaneous capabilities is costly. Where is saturation (horizontal scaling :) is an easy and cheap way to overwhelm/defeat such systems.
Defense companies won't like it, because commodity is synonymous with "low margin", so it's not surprising you end up with something that's two orders of mag higher price point and a buzzworld salad ("AI enhanced" because we have added some ML-based damping to a kalman filter to charge more money) to go with it
[1] https://en.wikipedia.org/wiki/Sinking_of_the_Moskva
[2] https://www.forbes.com/sites/vikrammittal/2025/01/24/russia-...
Who's to say that's not what Replicator is aiming for? Sans the "foreign enemy" part, perhaps
The "traditional" defense companies typically hire rest-and-vest talent, rewarding mediocrity over actual innovation. I have had many bright peers leave the defense industry for big tech and HFT, because the defense industry's backwards culture is very much about hammering down the nail sticking out and "sticking to what works" (outdated technologies, scamming the government for 20x the production cost.)
It would almost be too easy to hire some SV talent and make way better products for a tenth of the price. I guess my only qualm is that I don't want to make literal weapons.
It’s specifically because there is no domestic robot production capacity that can meet the requirements. I’ve been buying mostly Chinese robots cause nothing domestic can compete.
Neither BD nor pickle nor cobot etc can actually do what unitree and DJI have done at scale - it’s just a fact.
The root of this is that PE and Venture aren’t investing in US manufacturing - full stop. So they wait around for the govt to invest in infrastructure and then follow on sheepishly.
I’m not seeing the same kind of “private” investment in massive robotic projects so until that happens we’re going to lag on SOTA with US robots