Are we about to create real life Terminators? Humanoid robots built for war?
In this episode of TechFirst I talk with Sankaet Pathak, founder and CEO of Foundation, a California-based humanoid robot company that is not afraid of the defense market. We dig into why he is building humanoid robots that can work three shifts a day, how they plan to scale from dozens of robots to tens of thousands, and why he believes humanoid robots will one day build bases in Antarctica and cities on the moon.
We also dive deep into military use cases. From logistics and infrastructure to “first body in” building breach operations, we explore how humanoid robots could change asymmetric warfare, deterrence, and who wins future conflicts.
- Get deep insights on the TechFirst Substack
- Subscribe to the TechFirst YouTube channel
Check out our conversation here:
Transcript: humanoid robots built for war
Note: this is an AI-generated transcript. It may not be 100% correct.
Sankaet Pathak: If we ship 10,000 robots next year, that’s about a billion dollars in recurring revenue. If we get to 40,000 robots the following year, now it’s like $5 billion in recurring revenue.
John Koetsier: Are we about to create real-life Terminators — humanoid robots built for war? Hello and welcome to TechFirst. My name is John Koetsier. There are hundreds of companies now, tons of companies, building humanoid robots. The reason is simple: those who get them win. Companies will win financially. Countries will win.
Economically, alliances and geopolitical groupings will win. Influence and power. And that’s just worker drones in factories and homes and other places. What about military applications? Can we build humanoid robot soldiers that fight in armies — enlist? It would certainly change asymmetric warfare situations like the Russia-Ukraine conflict, where one country’s a tenth the size of the other.
We have the founder of Foundation. It’s a California-based humanoid robot company. They’re not afraid of the defense market, and his name is Sankaet Pathak. Welcome.
Sankaet Pathak: Thanks, John. Thanks for having me.
John Koetsier: Super pumped to have you. Super pumped to have this conversation. It’s US Thanksgiving and you’re working. I guess it’s the startup world.
Hey, this is where we are. Before we get into everything, we’re going to talk about the military piece. We’re going to talk about what humanoid robots can do in war, and just assisting the army with lots of stuff. We see a lot of that happening in Ukraine as well — like supply, exfiltrating wounded soldiers.
We’ll talk about all that stuff, but I want to talk about your company. I want to talk about your robot that you’ve built. I want to talk about how it works and your master plan.
Before we get there, let’s start here. Your company was founded like 18 months ago. You already have a significant production robot. How on earth?
Sankaet Pathak: It’s been a very busy year and a half. There’s no doubt about it. Pretty much the team has been working nonstop — close to seven-day weeks now.
I think it’s kind of like a formula of three different things. When I started Foundation, I got pretty lucky on a bunch of different fronts. I got pretty lucky on the AI front — we were able to acquire an AI research lab based in Munich that specialized in something that I thought was critical: being able to do both robots that can do a lot of tasks, and then a lot of robots that can do tasks — which are, I think, two different very challenging problems in robotics.
Second, we acquired a hardware company as well, which gave us really good actuators. Actuators are pretty much like the muscles of the robot — they enable the robot to move. So combining both of them, we were able to recruit really good, very talented people from companies like Tesla, 1X, Boston Dynamics, SpaceX — a bunch of these different companies.
So it’s kind of like we were able to start off — no pun intended — on a foundation of pretty mature technologies. We had a really good team, we have a really good team, and then everyone’s just intensely working hard, iterating, which enabled us to have an upper torso of the robot working by January, and then from there till now having a walking robot and actually deploying robots in real factories.
So we essentially have two robots deployed doing actual real work on a production cell at a car manufacturing plant. And now we’re preparing to ship about 10 to 15 more soon. So yeah, it’s been a busy year and a half.
John Koetsier: I can see. I mean, you’re working on US Thanksgiving. It’s a holiday weekend, almost. Nobody does. So I appreciate that. That’s super impressive.
You’ve leapfrogged a lot of people by kind of hacks — getting some teams and getting some technology that helped you go. Your scaling plan is insanely ambitious as well. Your plan is for 100 units in 2025. Sounds like you’ve got like 75 to go, maybe 50 to go — I’m not sure. Your plan is then 10,000 in 2026 and 40,000 just a year after that. All US-manufactured. Mind-blowing. Is it possible?
Sankaet Pathak: There’s a non-zero chance that this happens. So it’s possible. But to your point, it’s exceptionally hard.
This year, we would definitely end up over 40 units. Would we get to 100? I don’t know, but it’s still a pretty good outcome.
Next year, I think we have a decent shot of getting to 10,000. For that we have to manufacture a couple hundred robots in Q1, and then maybe a thousand in Q2, and ramp up to about 5,000 in Q3 and Q4.
So we have a shot of doing it. The good thing is I think we’re starting to understand the manufacturing process quite well. So we built our first manufacturing cell right next to our design studio in San Francisco, and that has really helped us learn what we need to do.
And the irony is, the majority of our manufacturing and assembly is manual.
John Koetsier: Yeah.
Sankaet Pathak: Even though we’re building robots, we’re building them through a lot of people. And the reason primarily for that is you can automate too fast. And when you automate too fast, you make all the wrong assumptions, and then you can miss on a bunch of those.
Tesla is a really good case. That’s the lesson of Tesla right there. And again, our head of manufacturing is an ex-Tesla manufacturing director. He worked on the Model X and Y ramp. So the lessons are just baked in: don’t try to automate too quickly.
So the majority of our manufacturing, even next year, would be pretty manual, which means the manual processes we’ve built right now would scale. The good thing about humanoids is you don’t have very large machines you have to buy — you’re kind of building something sized for a human. You also don’t need as many precision machines as you would for an iPhone.
So it’s right in the middle, in the sweet spot where people can do it effectively. So I think we have a good chance of being able to ramp up to that number next year.
And I feel like now the technology is almost getting to a place where it makes sense to do as well. Like, why are we going to go closer to 40,000? Because we also wanted the technology to mature more.
John Koetsier: Yes.
Sankaet Pathak: And now I feel like we’re really ready to put a lot of gas behind it.
So I think the last 18 months have been exponential. I think the next year and a half or two will be crazy.
John Koetsier: Yes, they will. They will. And it’s not just manufacturing, right? You’ve got manufacturing, but you’re going to have to bulk up sales and sales and marketing as well, right? Because if you’re going to have 10,000 in a couple years, maybe 40,000 in two or three years, that’s a lot to sell.
Will you sell them individually? Will you sell them as a SaaS — like a RaaS, robots-as-a-service?
Sankaet Pathak: Yeah, we currently sell it as robots-as-a-service. I think we’ll stick with that. I don’t think we’re really going to change that. The cost savings are there for our customers, so I think that’s a happy model that works for all of us.
Actually, I don’t think we’re going to have to scale sales and marketing. There is so much demand for this. We would have to scale kind of a support infrastructure — technicians and maintenance and safety and things like that. So I think there’s definitely some scale needed there. Today it’s pretty much just my co-founder, Mike, that does all of it.
And we don’t need 50 deals. We need like five really high-quality, large deals. And yes, they can scope out to be hundreds of millions of dollars in annual recurring revenue purchase orders.
I see a path to getting to a billion dollars in recurring revenue pretty much next year or two. The market is that big. So we probably would not spend almost any money in sales and marketing, but definitely a lot on product, a lot on manufacturing, and then a lot on support and servicing infrastructure.
John Koetsier: Cool. So I want to talk about Phantom, your robot, and its capabilities — what it can do right now, and some of your development plan.
Before we jump to that, you’ve got a pretty interesting master plan that includes a base in Antarctica and thoughts about the Moon and other things like that. In broad terms, what’s your master plan?
Sankaet Pathak: Yeah. When I was starting Foundation, I was at a phase in my life where I was craving working on something that had a very profound impact on humanity and forced me to think at very large timescales.
That’s one of the reasons I named the company Foundation. One of the things that stuck with me from the books — I thought there was an obvious connection.
From the Asimov books, as a child I would be fascinated by how the whole strategy and story unfolded over hundreds of years. And I felt like we always just think next month or next year. So I really wanted that to be a reminder: what we’re building, we’re building for something that can fundamentally change the arc of humanity. And that’s what I strive to do.
And what got me really excited about robots generally in that context: one, it wasn’t too early and it wasn’t too late. I felt like almost all the technologies that needed to exist for this to happen were finally available.
But then the second big piece was: it got me pretty excited thinking about if we can create a world where we pretty much have on-demand infinite labor. And historically we’ve had periods where we’ve had that, except they’ve been not good for a lot of humans because that was through slavery.
So it’s like, how do you — and I would actually argue we haven’t really ended slavery even now. We’ve just offshored it to other countries. You have people working in China and India at such a cheap price that’s not even a living wage, building a bunch of our products and services. It’s kind of like corporate slavery.
If we can create a scenario of infinite labor, I think we can end slavery for good. Which means we can end incentive structures that we’ve built in society to force people to do jobs they don’t want to do.
John Koetsier: Mm-hmm. Mm-hmm.
Sankaet Pathak: And it also enables us to do things that are quite hard for us to do today. For instance, most of the Earth is still unpopulated, and that’s primarily because infrastructure doesn’t exist. Setting up the same infrastructure I have in San Francisco in some remote part on Earth is quite tough.
But if you can change that equation fundamentally — by and large, you have resource abundance, infrastructure abundance across the world, anywhere — I think you’ll have more people move out versus more people move to these coasts.
And then around the same thought process, the idea of going to the Moon and Mars is very exciting. It’s one thing I’d like to see happen in my lifetime. And I don’t see how we do construction and maintenance and defense and all these different things in space, which is totally impossible for humans to do.
So pretty much we drafted the master plan around that, which is: our North Star is we want to chart a path to being able to build a base on the Moon that is like a city. And then you work backwards: step one, you need a product that works. Step two, you need a company that has the capital to invest in something that ambitious. So obviously build Phantom, scale Phantom. And then the third piece is: demonstrate you can do these very complex citywide tasks.
Thus the base in Antarctica — it could be pretty much any remote part on Earth — where the goal would be you send thousands of robots and they’re just building infrastructure and preparing it for humans.
And we need to be able to invest on the order of billions of dollars to do that, which means we need to have profit margins of billions of dollars ourselves. So that’s how the plan’s been formulated.
If we ship 10,000 robots next year, that’s about a billion dollars in recurring revenue. If we get to 40,000 robots the following year, now it’s like $5 billion in recurring revenue, and now it’s starting to make sense that, okay, now we can do something as ambitious as we want to.
But all of that is the goal of doing that. That’s why I was very transparent: I put the master plan, I wrote it down, I put it on the website. I want every single person who joins the company to read it. If it sounds too crazy, you shouldn’t join. By definition, people believe in it. That’s what people want to do. It energizes us, but it’s a lot of work.
John Koetsier: Yep. Yeah, absolutely.
Okay, so in the year and a half that you’ve been around, you come out with a production robot, which looks pretty amazing, is actually working already, and you have plans to scale that significantly.
So it’s Phantom. It’s pretty interesting. It’s about five feet, nine inches — 1.75 meters tall. It’s about 176 pounds, or 80 kilograms. Can carry about 44 pounds, 20 kilograms. Tell me about Phantom. What’s cool? What works? What’s awesome about it?
Sankaet Pathak: Yeah. We kind of sandbag on the payload capacity. You can technically do over 40 kg pretty easily.
It’s a very strong robot that is also safer to be around compared to other robots. Again, anything with heavy weight is dangerous because electronics fail and they fall — it can hurt you. So by no means is it safe enough that I’d have kids around it. But it’s shockingly safer than most of the equipment that exists in factories and warehouses.
John Koetsier: I saw it. You’ve got a video and you’ve got people right around it, and they’re pushing its arm and stuff like that, and it’s not whacking back or anything. It’s quite compliant, in a sense.
Sankaet Pathak: Yeah. This goes back to the actuators we built for the robot — the motors and gearbox things that make the robot move. Essentially, we have a fully back-drivable actuator. And what I mean by that is you can push against the motion of the actuator and it’s not going to break the gearbox. It’s going to go against it if you put force into it.
You have industrial actuators that can do that, but they can do that only once you add something called a torque cell. By physics, they’re unable to go against the motion of the motor. But if you have a torque cell, then it identifies an obstruction and reverses the motion. So technically you have this other sensor trying to detect an obstruction, versus in our case, it’s a very smooth gearbox. So if it feels an obstruction, you can push against it and it’s going to go backwards, which isn’t really possible with others. So it does make it more compliant and safe to be around for that reason.
The other benefit is the gearbox is very efficient. Usually industrial actuators have close to 40 to 45% efficiency. Ours is over 95%. Which again has some downsides too, which I’ll cover in a second.
But we really prioritized very low-friction gearboxes because that means low heat dissipation, which means low wear and tear. It’s kind of like having arthritis in your joint or not. Your longevity becomes much, much higher if you don’t have arthritis, and it’s much less if you do.
So we wanted to build these robots and have them work for 6, 7, 8 years and pretty much not break.
So we picked these low-friction gearboxes. They dissipate very little heat. They’re fully compliant so you can push against them because the gearbox is so smooth. If you have a same-size motor, almost all of the torque in that motor translates down to the robot — you lose very little torque. Very little power in the gearbox.
Unlike industrial actuators, you lose a lot, which means you can build more powerful actuators for the same size as you could with an industrial robot. And most other humanoid robots today are also using industrial actuators. So Phantom ends up being more powerful, dissipates less heat, is more compliant, has a longer runtime.
For instance, we’re the only company that I’m aware of in the US — a humanoid company — that is doing full three shifts every day. So we do three shifts every day with a robot.
John Koetsier: Wow. So is it recharging, or is it swapping out a battery?
Sankaet Pathak: Currently it’s connected to power. Because it’s working in a pretty small cell, so it has to move back and forth, but it’s still a small area, so you can just connect it to power. It doesn’t have to travel long distances, so it ends up working out.
But the other reason why it can do that is because the motors do not overheat and don’t need to be turned off, which some of our competitors have as a challenge today. So overall I’m really happy with the design choice.
Now, obviously when you have a gearbox that has close to no friction, it pretty much has high inertia during motion. So when something starts moving, it’s hard to control it with precision.
Industrial actuators are really good at precision movement because historically industrial actuators had to be blind — they didn’t have cameras. They still don’t. Most of them still don’t have cameras. So all they’re doing is going from point A to point B pretty much constantly, and they cannot have any kind of drift. They cannot go to point A minus one or point A plus one — they’ll be in the wrong place.
But in case of humanoids, where you have cameras, you have vision-based feedback. What that’s called in technical engineering terms is backlash. The gearbox has almost no wiggle — it’s pretty precise. But you do get a little bit of backlash on very low-friction gearboxes.
But even humans have high backlash. We don’t always go to the same place again and again — we use vision to correct it. So we think that’s such a small trade-off to make for what I described: you get a much more compliant actuator, much more powerful actuator.
So that piece is pretty unique about the hardware that we’ve built. And I think that’s the right design choice because it’s enabling us to build robots that can pretty much run for a long period of time.
John Koetsier: Yeah. What about the hands? Hands are generally one of the most challenging parts of a robot and obviously take a lot of wear and tear as well. And trying to replicate human hands is probably insane at this point in our technology, and you probably don’t need that as well.
How do you see your hands — your current tech there? Are you happy with it? Is it pretty versatile?
Sankaet Pathak: I am happy with it. For the use cases we’re going after today, they pretty much have about 12 degrees of freedom, so they can open and close.
What they cannot do is abduction. What they cannot do is independent — every single carpal being able to move — you cannot move them independently. So they always go like this. They open like this, they go like this. Which is good for a lot of pick-and-place: you have a large part, you can pick it up, you can move it.
But I do think for you to build a robot that can interact with the world the way humans do — which I think is the precursor to building a robot that is generally useful, that can do pretty much anything a human can that we don’t want to do — you do have to emulate human degrees of freedom, because the world is built around our degrees of freedom and our assumptions.
So the second-generation hand we’re building right now is a tendon-actuated hand. It sends tendons down the hand. Most of the motors go to the forearm. Some motors still remain up here, and then you can pretty much move it like a human hand.
So that’s the one we’re pretty committed to. We’re building it now, and probably will unveil something next year. And I think that is the last remaining part of the hardware that needs to be perfected, after which it’s a pure AI and data problem.
And these robots will become increasingly useful at a very fast clip.
John Koetsier: Very, very cool.
Okay, so you’ve talked — we’re going to get to military in like two minutes — but you’ve talked mostly about manufacturing. Some robot companies are showing humanoids in houses and doing stuff — cleaning, making food, that sort of thing. Is that any part of your plan, or is that not where your sweet spot is?
Sankaet Pathak: No, I think we’ll do that eventually, but I would really encourage people to not think of home robots as robots.
The question you should be asking is: what’s the next computing platform that, if I could put it in a house, fundamentally changes and improves people’s lives? And I’m not convinced that it’s a humanoid.
I think it’s definitely something that can interact with your environment, for sure. I think that’s important. But I think the design principle for home robots — which I would call the next computing device — is you need a genie in a bottle. Someone who you can just say anything to, and they grant your wish and it just happens.
And I think we’re a few clicks away from being able to build a technology like that. I don’t think you can just clean the dishes and warrant having a robot at the house. I think you have to be generally useful — pretty much like a genie in a bottle.
John Koetsier: Yeah.
Sankaet Pathak: Not at all intimidating.
John Koetsier: Yes. Safe for kids and pets.
Sankaet Pathak: Not just even safe — you have to feel safe. It has to feel safe to be around. And it’s kind of invisible. It’s in sight and then it’s out of sight.
And I think we’re a few clicks away from being able to do something that is truly special there. It’s not a pendant on your neck. It’s not smart devices. It’s not a humanoid. I know for a fact it’s not those things.
John Koetsier: It’s something else.
Sankaet Pathak: Yeah.
John Koetsier: Okay. Cool. Well, I wanted to ask that question because your whole strategy — you talk, you need five big deals, not 100,000 small deals — so that makes perfect sense for your strategy.
Okay. Let’s talk about military. Why do we need humanoid robots for the military?
Sankaet Pathak: My philosophical reason for being interested in the military from the very beginning was: I wanted to build a city on Mars, and I cannot build a city on Mars without learning how to maintain and build infrastructure and defend it.
That was even before Mike and I, and Arjun and I, came together as co-founders. At the beginning, even then, when I was starting to think about Foundation, I was very interested in building a very useful, intelligent tool for the military.
And now my co-founder is an ex-military person. Arjun is an ex–Pentagon person, so they are much more passionate about it at a different level. Their opinion is you should really work hard to give the US military smarter tools so that they can be more effective.
And what that really leads us to is: it’s very important to teach these robots to do tasks that are either too boring or mundane for humans, or not possible for humans, or dangerous for humans.
There are a lot of things in the Department of Defense. For instance, something like drone inspection — you have to hire a workforce around it, and it’s very difficult to hire. So you can have a very understaffed military logistics organization for that reason, and that’s quite dangerous. It’s not safe for soldiers.
And then second, there are so many scenarios where soldiers put themselves in harm’s way because they’re the first body into a situation and they don’t really know what that situation is.
John Koetsier: Exactly. Charge over the hill, don’t know what’s there.
Sankaet Pathak: Yeah. So I think there are tons of things like that — either back-office work and/or first body into an unknown situation — which I think robots are much better suited for and should do.
And we already are seeing robots show up in military. It’s just, I feel like most of that autonomy has been air-based. And it’s a very crude instrument because if you have a point of interest in a building, your only recourse, if you don’t want to risk the lives of your soldiers, is to take down the entire building. Versus with a humanoid, for the first time, you can change that and be very precise about who in that building you’re going after — not everyone in that building.
So I do think this technology will also reduce loss of innocent life because you can be more precise with your mission.
So yeah, for all of those reasons, I think it’s very important to build tools that are smart and can help with a lot of the mandate of defense.
And other countries — China, Russia — are pursuing it. And we’ve already seen this play out a little bit in Ukraine where there’s this big cost disparity: we are spending $100,000 for every $1,000 they’re spending. And some of that is economically feasible, and then eventually economic feasibility impacts lives of people as well — either through inflation and/or through loss of life.
So I think it’s much better to give the US military smarter, cheaper tools that are more precise and can help deter the enemy and/or help counter their attacks. And we feel passionate about that.
John Koetsier: That’s cool. I think that financial inequity that you’re talking about is you’re shooting down a $1,000 or $5,000 drone with a $100,000 missile. Let’s say you’re using a Patriot system or something like that. So that makes no sense whatsoever.
And you can wear down an opponent if you beat them essentially financially, or economically, by doing that.
I don’t really understand the Mars reference — because if you put a city there, what are you defending it against? There’s not much there. What’s your thinking there?
Sankaet Pathak: There’s not much there yet. I think there’s going to be a city that we build. There’s going to be a city that China builds.
But more importantly, conceptually, what I was interested in is the US Department of Defense also builds a lot of infrastructure in very remote places.
John Koetsier: Yes, it does.
Sankaet Pathak: Huge. So to me, that was the bigger area of interest. I was like, okay, that’s the skill I really want to learn: how do you build and maintain infrastructure of that size and scale in areas that are remote? That’s what I meant more so by it.
John Koetsier: I understand. That makes a ton of sense. And that wins wars, honestly. That’s logistics. And that’s equipment that you can get there, that’s people that you can maintain and keep safe.
World War II — the Seabees were critical to the US war in Asia, essentially against Japan, and in the Philippines and stuff like that. Building bridges, roads, airfields — all those very dangerous things. You’re not a combat fighter, but you’re still at risk.
So it’s interesting that most of the things that you’ve talked about in terms of humanoid robots for the military is moving things, checking things, bringing things, building things. It’s not pulling a trigger.
Sankaet Pathak: I mean, at some point it would probably be pulling the trigger, especially in the scenarios — the use cases — I’m talking about, where you’re kind of like the first body in. Because if you’re first body in and you’re docile, then the enemies are not going to really expose themselves. So you have to be first body in and deadly.
And we are currently in the process of doing our first pilot around that — for building-breaching type operations.
John Koetsier: Wow.
Sankaet Pathak: So we do think those things will become important for robots to be able to do as a partly useful tool in the military.
John Koetsier: Will that be fully autonomous? Will it be partial? Will there be a rider — essentially in VR or something like that? How do you see that?
Sankaet Pathak: It’s going to be pretty much like what we do with drones today. They’re not fully autonomous. They are autonomous on how, but not on what.
John Koetsier: Yes.
Sankaet Pathak: How to move, how to maneuver, how to fly — autonomous. Where to shoot, where to go — not autonomous. And the same is going to apply here.
John Koetsier: Wow. So the future of war is video games.
Sankaet Pathak: The future of war is video games, for sure.
John Koetsier: Interesting. This has huge implications for power and war, right? How does this change wars? Does the richest country win? Does the most productive country win? Like, if Ukraine can make 100 million drones, they win that war, right? If China can outproduce us on humanoid robots, they win any future conflict, right? This is interesting changes in military doctrine.
Sankaet Pathak: Or it serves as a mutually assured destruction scenario, and then nobody wins and everybody pacifies themselves.
John Koetsier: That’s the best-case scenario.
Sankaet Pathak: Yeah. But that’s happened with nuclear weapons already.
John Koetsier: Yes. Yes.
Sankaet Pathak: I don’t want to get too complacent — that’s why you usually don’t talk about it as much publicly — but it does seem like the biggest impact nuclear weapons had was it toned down the kind of force we fight with.
You don’t see us and Russia go really head-to-head for that reason.
John Koetsier: No. Proxies.
Sankaet Pathak: Yeah. You see proxy wars. And I have my opinions on the proxy war thing, which is probably for another day.
I do think you can create a similar disincentive on just your regular military might. I do generally agree that peace through strength is a thing. I wish mammals and humans were not built that way — where they operate that way — but they sadly do.
So unless core biology changes, I do think the way to not get bullied is to be like, okay, I think I’m going to get beat up if I believe this person.
So I do think there’s an element to that. And I do think if the US military had like 100,000 robots that they could just demonstrate work, it would by and large end a lot of wars before they start.
John Koetsier: Wasn’t that a Teddy Roosevelt thing? Speak softly and carry a big stick.
Sankaet Pathak: Exactly.
John Koetsier: Something along those lines. Well, Sankaet, this has been fascinating. Super, super interesting to hear the progress that you’re making, what you’re working on, and how you see the future. Thank you for this time.
Sankaet Pathak: Yeah. Thanks for having me.