John Collison:
Okay. I’m very excited to talk about Zipline. So previously you were a professional rock climber. You studied biochemistry and computer science in Harvard, you had a previous robotics startup. Then in 2014 you started Zipline. Most people who do know Zipline probably know you for the medical deliveries in Rwanda and broader in Africa. You now also have operations. You just announced Chipotle and burrito deliveries in the United States, and you have a delivery take place every minute, and you have a hundred million miles flown.
Keller Cliffton:
Think like 110 or 120 now. Yeah.
John Collison:
Okay. Yeah. So is that the good potted Zipline summary? What would you add?
Keller Cliffton:
What would I add? I mean, when we were starting to build Zipline in 2013, 2014, our backgrounds were in robotics and automation and software. And it really seemed to us like it should be possible to build an automated logistics system for Earth. When it comes to robotics, you really want boring and repetitive. That’s ideal. Constrained problems, boring and repetitive. And logistics is about as boring and repetitive as it gets, but it’s this underlying infrastructure that affects everything in the world. It affects how people get access to healthcare. It affects how people even choose where they live. It affects how much time people get with their families versus having to spend time driving around and buying things. And so yeah, the idea was simple. The execution has turned out to be a bit more complicated than we were expecting.
John Collison:
Yes, and you’ve made the point that what does a UPS truck weigh? Five, five,
Keller Cliffton:
6,000 pounds.
John Collison:
Yeah, yeah, yeah. It’s like we’ve taken for granted that we just have these massive vehicles, trundling around our suburban streets is the best answer for delivering things, and it’s not obvious, couldn’t be improved from there.
Keller Cliffton:
I think the best way to think about that is that when you’re batch processing things and delivering things over the course of a day, that works pretty well. You can load tons of packages into a car and then the car truck over time can deliver things slowly to lots of homes. But obviously the thing we’ve seen over the last five or seven years is the rise of instant delivery. There are now five and a half billion instant deliveries happening every year just in the US, and this is where we’re using a 4,000-pound gas combustion vehicle driven by a human sitting in the front moving the steering wheel back and forth to deliver something to your home that weighs on average five pounds.
John Collison:
Yes.
Keller Cliffton:
So actually you do not have to be a physicist to realize this is a really weird way.
John Collison:
It’s a bit over-engineered…
Keller Cliffton:
… or under-engineered, depending on how you think about it. But this, it creates traffic in our neighborhoods. It creates pollution in our neighborhoods. It creates safety issues in our neighborhoods. It’s phenomenally expensive actually, and it’s pretty slow.
John Collison:
And obviously the competing vision is the sidewalk robots that some people are pursuing. But I presume you think for many of these applications, they’re just too slow, and you actually can’t fit that many of them before they start clogging up sidewalks. Is that basically the argument you make?
Keller Cliffton:
My personal opinion is that we should be giving space back to humans, not taking space away from humans. So we have already, I mean, I grew up, maybe you grew up, we played in the streets, we played street hockey. I wouldn’t really let my kids go play street hockey today. So something has mentally changed where that doesn’t seem like as safe a space anymore. Sidewalks maybe are the last place that are meant to be for humans. The idea that, okay, now the robots are going to take those over too, and humans are going to have to jump out of the way as these sidewalk robots go back and forth, doesn’t make a lot of sense to me.
John Collison:
Why has the drone delivery vision taken so long where people were talking about tackle copter on Hacker News in 2012? The big Amazon announcement was what year?
Keller Cliffton:
2013.
John Collison:
And so everyone has felt like it’s just around the corner for, well, longer than a decade now. Is it the FAA’s Beyond Visual Line of Sight rules? Is it like the tech was surprisingly hard? Why is this happening now? You guys just announced Chipotle burrito deliveries. Why is that happening in 2025 and not 2015?
Keller Cliffton:
I think it’s similar to the question of why do we not have autonomous cars? When certain leaders were saying, “Oh, it’s right around the corner in 2015 or 2016.” I think that with any advanced technology, there’s always, what do they call it? The trough of disillusionment curve. It’s kind of like you have the early hype, people realize something should be possible. It’s really exciting, and then it goes, completely craters, because people realize it’s really hard, and then everybody basically is like, “Well, that was a stupid idea. It’s never going to work.” And when people say it’s a stupid idea, it’s never going to work, that’s when the actual hard work begins. And then there’s a decade of building all of the fundamental underlying infrastructure and technology that’s required to actually make something work in the real world. And work in the real world is kind of like an easy way of putting it. But the reality is, whether you’re talking about autonomous cars or autonomous aircraft, you have to figure out not just safety and reliability. You have to figure out regulatory, you have to figure out manufacturing and maintenance and supply chain, all of the software systems that are required to run behind the hardware to really make it reliable. And then you also need a good customer experience on top of all that. And then ideally, you also have to make the unit economics work.
John Collison:
So it’s the classic thing of producing convincing demo in hardware is relatively easy, but then getting something actually useful operationally is where the last 10% of the work is another 2x or 3x from there.
Keller Cliffton:
I mean, Elon, I remember watching him for ten years, say again and again, people would always say, “Oh, Ford’s going to do what you do. GM is just going to do it and crush you.” And he just kept saying, “Prototypes are easy, production is hard, prototypes are easy. Production is hard.” I think it’s deeply true in hardware. It’s very easy for people to, and we’ve seen it again and again and again, specifically with drone delivery. Someone will buy a quadcopter off the shelf, duct tape a Snickers bar to the bottom of it, manually pilot it when the weather is good, two miles landed. It’s like, “We just did drone delivery!” And then get a journalist to cover that. But the reality is that what customers really want is teleportation, and they don’t care at all about drones. This is kind of a counterintuitive thing that we’ve learned. It’s like when we’re working with all these ministries of health or these governments or big companies like Walmart or big hospital systems in the US like Cleveland Clinic or many others, they all want to be able to just teleport things directly to patients or directly to a customer’s home in a way that either saves a life or transforms a customer experience. We could be using whatever. They’re not interested in drones. They don’t care about drones. They don’t want to deal with any of that complexity. All they want to know is something goes from point A to point B fast enough to solve a real problem. And so I think that a lot of the technology kind of sits behind a curtain in order to enable a customer experience that is just jaw-dropping and changes the way people live their lives.
John Collison:
So demos are easy, production is hard. You got started in production and really ramped up in Rwanda. How did that come about where people, I can understand maybe you might want to go outside the US for a more favorable regulatory environment, but most people will go to Mexico or Canada. How did you get to Rwanda? And maybe you can tell the story of scaling up operations there.
Keller Cliffton:
Yeah, hard to frame. I mean, I think that when we were starting to build Zipline in 2013, we spent about two years building the initial version of the technology and everybody told us this was a stupid idea. Everybody told us there was no way it was going to work.
John Collison:
Was it hard to get funding for?
Keller Cliffton:
Oh, incredibly difficult. I remember we were begging people to invest $250,000 in the company.
John Collison:
What was the first big break in terms of who invested?
Keller Cliffton:
Sequoia led our Series A and took this bet on us. This was actually before we started building Zipline though.
John Collison:
Oh, wow.
Keller Cliffton:
Yeah.
John Collison:
This is for your previous startup?
Keller Cliffton:
Yeah, well same company, actually.
John Collison:
Oh, but it pivoted. I didn’t know that.
Keller Cliffton:
Yeah, most people don’t. And then as we were starting to build Zipline, I just think there was so much skepticism. I mean, it’s such a crazy idea. What do you know about aviation? Nothing. What do you know about healthcare? Nothing. What do you know about operating in Africa? Nothing. Okay, well…
John Collison:
It’s like hard tech, operating in Africa…
Keller Cliffton:
Yeah. Let’s stay in touch.
John Collison:
Smile and nod…
Keller Cliffton:
Yeah, exactly. I think that our intuition was like we needed a country that was as desperate as we were desperate as a startup, maybe that sounds bad.
John Collison:
Which is like a threshold question. Why didn’t you start operating in the United States?
Keller Cliffton:
There just wasn’t a regulatory pathway to do so.
John Collison:
Okay. Just the ban on drones. Yeah.
Keller Cliffton:
Totally impossible to do it in the US. And also we really wanted to find, we had this instinct that to convince a country to do this at scale, we needed to focus on lifesaving use cases. And so we really wanted to find a national healthcare system rather than a fragmented private one you have in the US. So we were looking for a national healthcare system, healthcare challenges that we felt we could really clearly solve, and a country that would be entrepreneurial and fast moving alongside us.
And so that to us had us looking at countries in East Africa. We’re also talking to countries in Latin America, and interestingly, I remember this meeting with the minister of health of Rwanda. I mean we were beyond clueless. It’s just like a 24-year-old showing up in East Africa. We build robots. We think that this could work. And I remember this meeting with her where I was saying, “Hey, we’ll deliver blood and vaccines and cancer products and all these different things. We could do it all instantly in a way that would be transformative and reduce waste and save lives.” I remember she looked at me and just said, “Keller, shut up. Just do blood.” And as she was explaining to me, 50% of blood transfusions are going toward moms with postpartum hemorrhaging. 30% are going toward kids under the age of 5. And it is a super hard problem from a logistics perspective. You have packed red blood cells, platelets, cryoprecipitates, all these different types of each A, B, AB, and O. They have all different storage requirements and shelf lives. Platelets only last six days, for example, when you get them out of a donor into a patient.
So they gave us 21 hospitals, and they said, “Just do blood.” And that’s what we did for the entire first year. And it was a disaster, by the way.
John Collison:
Why?
Keller Cliffton:
Because we had no idea what we were doing. I think apropos of the flag behind us, we didn’t know how to integrate with a national healthcare system. We didn’t know how to integrate with a civil aviation authority, and we definitely didn’t know how to build an aerospace-grade product that was reliable yet. So we launched and we were pulling constant all-nighters. We were flying back and forth between the US and Africa. We had the entire Rwanda team hired, and the system only served one hospital for the first nine months reliably. And we were just trying everything we could do to make the system work reliably for that one hospital.
John Collison:
Why Rwanda out of all the sub-Saharan African countries you could’ve picked?
Keller Cliffton:
We can answer that question now. We have a much better understanding. The reality is then we weren’t smart enough. We just got lucky. The answer now is Rwanda really is kind of like the Singapore of Africa. It has a very technocratic leader with a very technocratic leadership team, where they’re taking really smart bets on technology and moving quickly, and it’s a very, very entrepreneurial country.
John Collison:
It’s got a much stronger civil service, yeah.
Keller Cliffton:
Super strong civil service. I can’t even describe to you. I mean many times I would be on calls with a minister or a chief of staff at 9:00 on a Saturday, 11:00 p.m. on a Saturday. I mean moving lightning fast. That’s kind of what I mean. There are many countries that are moving very fast because they know that the only way they’re going to get ahead is by being aggressive and entrepreneurial and making bets.
John Collison:
So it was messy at first. When did you crack operations in Rwanda and you felt like, wow, this is really working.
Keller Cliffton:
So after nine months of all-nighters, total desperation, everything that could break did break—it started working reliably for that one hospital, which was called Kabgayi Hospital. By the way, that hospital was only five miles away. So it was like we were doing the easiest possible thing and failing at it. It didn’t feel great for nine months.
John Collison:
But when did Zipline get to the point where Rwanda would definitely not let you rip it out? Like, oh, we rely on this.
Keller Cliffton:
So after nine months, it started working well for that hospital and then we rapidly added the other 20. So we were actually serving all 21 hospitals in the contract by the end of the first year, and then they asked us to expand to 50 hospitals and then 200 and then 500. Then we went from delivering blood to vaccines, cancer products, infusions, transfusions; today we’re like the sole distributor of 500-plus health products in the public healthcare supply chain. And then we started expanding outside Rwanda. Today we serve 5,000 hospitals and health facilities. It’s become the largest commercial autonomous system on Earth.
John Collison:
If this model makes total sense, which I think what you’re saying makes sense, there are various healthcare products where you may need them anywhere. You can’t predict where you’re going to need them and you need to get them in a time-critical fashion. Why haven’t other countries followed? I mean you’re starting to do this in other countries, but why as of 2020 was every country not beating down a path to your door saying, “Oh yeah, we’ll take 10 Zipline systems for our health service.” It’s kind of interesting given that all countries’ healthcare needs are kind of the same, a lot of national healthcare systems, you would think more people would want to replicate the system.
Keller Cliffton:
Yeah, I would say they kind of were because starting in 2022, so it took us three years to stabilize and grow it in Rwanda. And then it was also very helpful. We had a number of big academic institutions come in and want to do studies. The University of Pennsylvania came in and did this big study across all the hospitals we serve and showed a 51% reduction in maternal mortality. This is pretty profound.
John Collison:
Pretty causal.
Keller Cliffton:
Very causal. They have their control group and their noncontrol group. There are interesting different reasons for it. A lot of it is around bloat availability, but also part of it is just that increased trust in the health system. More moms give birth in hospitals when they know that the hospital is going to have access to the products they need. By the way, when we were starting to build Zipline, we were told this was such a stupid idea, but if we had known we were going to get a 5% reduction in maternal mortality, we would’ve told you, hell yes, this is completely worth it…
John Collison:
Victory condition.
Keller Cliffton:
Victory condition. So the idea of a 51% reduction is something—don’t listen to the experts, I guess is a good takeaway. And so from there we also got these additional studies showing 70% reduction in vaccine waste, 32 percentage point decrease in zero-dose children, so these are kids who never get vaccinated. And so we actually did see Zipline then launched in Nigeria and Côte d’Ivoire and Ghana and Kenya. We started launching a lot of different countries in around 2020. The system started to scale very fast.
John Collison:
Have you seen any induced demand when healthcare systems have Zipline available to them? So you can imagine we can fulfill existing needs better, we need blood for a surgical procedure and we don’t have it. And now with Zipline we have it, but have you seen healthcare systems start to do things that they weren’t previously doing because they have this cool logistic system available to them?
Keller Cliffton:
For sure. I mean that’s the main thing that we are observing happening. I think there’s a lot going on that’s all pretty interesting and maybe we didn’t anticipate all of it. I would say one big thing to realize is with a health system, you’re typically balancing access against waste.
So if you want to make access really high, you send a ton of medicine out to the last mile, and then access gets better. But now you start throwing tons of stuff out because all this stuff has shelf lives and a lot of it is very expensive and cold chain dependent. If you want to reduce waste, you keep more stuff centralized, but now you have more people dying when they can’t get access to the thing that they need. So one of the really crazy things with Zipline was we were able to show that by keeping things centralized and turning the entire logistics system over to a just-in-time system, you can actually move both up simultaneously. That’s a big idea.
Another idea is what you’re saying, which is like the induced demand, you make it possible to vaccinate kids for example, at a different layer of the healthcare system. In Rwanda we now serve, I think it’s like 400 or 500 health posts in the country. And so these are the layer below primary care facilities. It means that kids can get vaccinated way closer to where they live, more kids get vaccinated, just more convenient. The last interesting thing that we see is, and this is on the opposite side of induced demand, is actually when you had a logistics system that wasn’t very reliable, you actually had a lot of systems that started to exhibit hoarding behavior, which is kind of weird. But basically what would happen is that there was never enough blood to go around, and these health systems would have to place their order for blood every quarter.
John Collison:
So they would over-order.
Keller Cliffton:
They would over-order because they know they’re going to be cut back. And so they would order twice what they needed on the assumption that the government’s going to send them half of what they actually asked for. And then that’s just like a brinksmanship kind of thing. The numbers get bigger and bigger and you lose all signal in the system. And so the really cool thing that we’ve seen is actually a lot of the demands have gone down. They’re actually confident that they’re going to get exactly what they’re asking for.
John Collison:
Yeah, it’s the just-in-time framework, but for a healthcare system. What is the business of what you’re doing in Africa? Are you paid by the healthcare system? Is this NGO funded? How does the business side of things work?
Keller Cliffton:
10 years of building the infrastructure in Africa, we always felt really strongly that we wanted this to be sustainable and we want it to be led by the countries. There’s a lot of work that gets done in Africa by NGOs, they’re funded by the US or by the international organizations. And I think what’s kind of most amazing about Zipline is it’s been entirely country led and almost entirely country funded. So we sign contracts directly with the ministries of health, the governments themselves, and then we help those governments save lives and save money. So typically we can, for any given ministry of health or health system, we can save millions of dollars. Plus you can show extraordinary healthcare impact and that makes it a pretty easy decision for any country.
John Collison:
Yeah, it’s one of these amazing, better, faster, cheaper tech use cases or tech value props where you go to a health ministry and you say, “We’ll give you a better logistics system that delivers stuff faster and is cheaper than how you’re currently moving things around.” And I presume that’s hard to pass up. Running a healthcare system is expensive.
Keller Cliffton:
Running a healthcare, but yeah, exactly. These products are super expensive, whether it’s vaccine or cancer products, they’re all cold chain dependent, short shelf life, very expensive and there is phenomenal waste. It’s mind blowing.
I remember, and one of the things that actually kind of inspired us on one of my first visits ever, and I won’t name the country, it wasn’t Rwanda. We went to this USAID-funded medical product warehouse, distribution warehouse. And we were going inside and you can tell it’s like, whoa, these systems are not good. By the way, all donor-funded and there’s a separation from the customer I think is the way to think about it. But we were in the warehouse, it’s kind of a disaster. It doesn’t look like things are going well. And then we walk out the back of the warehouse, and it’s just a field of boxes as far as the eye can see. And by the way, it all says “from the American people” on it. And I was like, what is this? And they were like, “Oh don’t worry about that. Don’t worry about that.” It’s like, no, really what the heck? It was vast scale. And they’re like, that’s all the medicine that expired last year. And so I think we had this kind of aha moment. It’s like, whoa. And the crazy thing is that was happening in a country that was having massive shortages and healthcare outcomes where people couldn’t get access to these products, which you could kind of see the two points to just connect with a line.
John Collison:
So what’s the lesson? The generalized lesson here, is it that existing aid supply chains are pretty inefficient right now and there’s a lot of improvement possible?
Keller Cliffton:
I think maybe one big generalized takeaway is that logistics is so boring and so in the background, we never think about it. If you’re living in, I don’t know, the Bay Area, you’re ordering stuff on Amazon, you have it delivered to you 30 minutes later or an hour later, whatever it is, it’s kind of like running water. You don’t really appreciate it until you don’t have it. There are parts of the world that have crappy access to logistics and then there is about five or six billion people on Earth who have no access to logistics. And I think that is maybe the big realization is that maybe we think of it as something that’s just ubiquitous or something you don’t worry about. But the reality is that if you don’t have that strong layer of logistics, everything else is phenomenally difficult and expensive to do. So you mentioned the public healthcare institutions. They’re all building verticalized supply chains in these countries, and it’s many different supply chains. It’ll be a different supply chain for AIDS drugs, different supply chain for maternal mortality, different supply chain for kids or neglected tropical diseases, and those systems don’t run very efficiently.
John Collison:
In how many countries today do you run your medical logistics network?
Keller Cliffton:
Eight.
John Collison:
Why is that not 80? Again, it feels like something every healthcare system should want, and it’s pretty proven at this stage where it’s been working well in Rwanda and other places for several years.
Keller Cliffton:
You sound like our board, John. You’re really putting the pressure on. What are our revenue growth numbers, and why are we not growing faster? Exactly. I think that there are a few things. I mean one is that we’ve always been pretty clear that it’s one thing to save the lives of the people you’re delivering to, but it’s really important that we be safe for the people who are flying over. Zipline now has about 120 million commercial autonomous miles and zero safety incidents, and that’s definitely hard to do. That has come from, I think the company being very just focused and step by step doing things in the right way, testing, validating, and making sure never to get over our skis.
John Collison:
But the laws of aerodynamics are not different in Rwanda versus Algeria versus the UK.
Keller Cliffton:
So that’s part of it. But then also it’s kind of like, I don’t know, sometimes we talk about it internally as a pig traveling through a snake. It’s kind of all about bottlenecks, scaling up hardware. I mean we were really only ready for a lot of scale. I would say 2020 was when we started expanding into new countries. We then went from one to eight countries, went from a hundred health facilities to 5,000 today. So the system has been scaling very, very fast. The other thing that did happen is that, 2022, when you see interest rates spike in the US, it’s like, oh man, that’s really inconvenient and annoying to us in the US. In Africa, it creates economic Armageddon.
John Collison:
And it feels like if you’re running a medical system, what you’re describing about having—I view you as not being in the drone business, but in this line of business you are in the medical inventory management business and you just have a much better inventory management solution because you’re able to do this just-in-time stuff. And there are a few medical systems that don’t want that.
Keller Cliffton:
I just think it’s becoming more and more clear that this is the infrastructure of the future. This is a smart thing for countries to be investing in. And so there is a lot that’s going on even just this year that is leading to massive growth both in terms of within countries, us adding new distribution centers and us adding new medical products and then also adding new countries to the list.
John Collison:
This medical logistics system we’re describing is based on your Platform 1 aircraft. And the way the system works, as I understand it, is you have a fixed-wing, battery-powered drone that is launched off a catapult, you [shoot] it off the catapult and then it flies from a distribution center, flies to the hospital, it drops the payload with a parachute attached, and they go grab it, and then the drone returns home, and I presume it just lands in the grass and you take it and plug it back in. That’s roughly how Platform 1 works, right?
Keller Cliffton:
Yeah. Doesn’t land in the grass. I mean we actually use something that is kind of inspired by aircraft carriers.
John Collison:
Oh, you catch it?
Keller Cliffton:
Yeah, we actually catch it out of the air.
John Collison:
Okay.
Keller Cliffton:
It looks kind of crazy.
John Collison:
Or it’s like the thing in Batman where they’re recovering the…
Keller Cliffton:
What thing from Batman? You mean the sky hook from Mission Impossible?
John Collison:
Yeah. There’s also a sky hook in Batman.
Keller Cliffton:
Sometimes people think sky hook. Basically the vehicle is flying by at about 100 kilometers an hour, and we can snatch it out.
John Collison:
Oh, I’ve seen that.
Keller Cliffton:
Called the snap catch…
John Collison:
It’s a big pole…
Keller Cliffton:
Big poles.
John Collison:
It’s like something out of a cartoon.
Keller Cliffton:
Every time you see it. Yeah, it really does look cartoonish.
John Collison:
Cartoonish energy.
Keller Cliffton:
Yeah, it does look weird. And every time you see it you’re kind of like, hallelujah. It’s amazing that that worked, but now it works at a thousand a day. Yeah, we got away with another one! But the cool thing about autonomous systems is that we’re using guidance navigation control algorithms that are hyper-precise. When you take the human out of it, you can actually achieve levels. Not that we want to take human pilots out. I realize I was sitting with a talented pilot, but robots are really good at flying vehicles. And so actually you can achieve levels of precision that seem very counterintuitive. Like the aircraft has a one-centimeter tail hook on the back of it, and we reliably catch that one-centimeter tail hook every single time as the vehicle’s coming by.
John Collison:
When you say robots are better pilots, I take no offense at that. Have you read The Right Stuff by Tom Wolfe?
Keller Cliffton:
I haven’t actually.
John Collison:
I really enjoyed it. Tom Wolfe, fabulous writer, all his stuff would recommend and certainly The Right Stuff maybe for listeners of this podcast is on the history of the Mercury space program. So before the Apollo program, the initial space program, but they describe how when they were making the spacecraft, their initial plan that was totally obvious to them was they would make them autonomous.
This is back in the ’60s, but it was still kind of patently obvious to them that they would make them autonomous. They faced rebellion from all the shit-hot Air Force fighter jocks in their program. Well, they just didn’t want to be passive passengers. We sent monkeys up to space, and monkeys don’t, we’re no better than monkeys just sitting there. So they added manual controls for the spacecraft afterwards, to the design, because the pilot…
Keller Cliffton:
…the controls weren’t connected to anything, but they could at least it’s like in an elevator, right? You can push the close button.
John Collison:
Exactly. But very funny. And so anyway, that was back in the ’60s, so yeah, totally. Yeah.
Keller Cliffton:
But a slightly more serious note on that, which is that today, the majority, I think you probably well know, the majority of airplane crashes are caused directly by humans. It’s actually a bit sobering, but also I think it is just a sign that the safety of these systems has progressed so dramatically that at this point the last unsafe thing in the cockpit is the human themselves.
John Collison:
So I have some questions on the tech and safety and how you do things. How do you avoid birds?
Keller Cliffton:
We don’t. Birds avoid us.
John Collison:
Okay.
Keller Cliffton:
Yeah. Turns out birds have pretty good detect-and-avoid systems of their own. And so they hear us coming.
John Collison:
How fast do you fly?
Keller Cliffton:
About a hundred kilometers an hour.
John Collison:
So you’re much smaller than airplanes.
Keller Cliffton:
We’re not sucking birds into it in the same way that a jet would.
John Collison:
How do you avoid airplanes?
Keller Cliffton:
So we avoid airplanes by basically it’s like a multilayer deconfliction strategy. First of all, we’re flying below the commercial aviation floor. So in the US that’s about 400, 500 feet. So we’re never conflicting with the kinds of normal airplanes that people who are listening would ever be flying in. We are also then on ADS-B, so we’re receiving ADS-B. So as long as something is squawking on ADS-B, we know it’s there and can avoid it. We also then have cameras on board and then we also do things like NOTAM, so old term, but “notice to airmen,” basically telling people when we’re operating in certain places. So all of these layers together combined to get us a safety profile that is super obvious that the FAA can improve.
John Collison:
Super cool. Then what, can you tell if a battery’s caught on fire?
Keller Cliffton:
So the cool thing about the way the aircraft is designed is that there are 43 major subassemblies on the aircraft. Every one of those subassemblies is designed from scratch by Zipline. So super vertical integration, and we really take inspiration from the way that planes are designed. We’ll look at the way a flight compute on a Boeing 787 is designed and they’re of course using multimillion power 20-year-old components. We can talk about why that’s so screwed up in a sec if you want. But the architecture is usually fundamentally sound. And then what we’ll usually do is just use cell phone components to build that architecture in a way that’s about one-thousandth of the cost but just as safe. And so you’re kind of using a lot of those aerospace paradigms of being able to fail over and building redundancy into every flight critical system. So, the aircraft, we have redundant nodes of the battery. One node can fail, the vehicle can still fly itself home. We have redundant flight compute architecture. So you can have a flight compute fail, you can fail over the second flight computer and fly home. You can actually reach into the aircraft while it’s flying and cut any wire with scissors. The plane will still fly itself home.
John Collison:
Because it’s wired, redundantly.
Keller Cliffton:
Redundant CAN bus. So basically we’re passing redundant CAN bus on the aircraft means that we’re not, you can lose any wire. We also have redundant control services.
John Collison:
Do you do that because you’re paranoid about midair things or like midair collisions or because, just as a general redundancy strategy?
Keller Cliffton:
General redundancy, I mean you’re not actually worried about someone reaching into the airplane with scissors, usually; that would be a hard thing to pull off. But what you are worried about is vibration and connectors shaking loose. That’s a core failure mode for aircraft.
John Collison:
So going back to the operations in like we’re talking about these Platform 1 operations that you’re doing in Africa and the medical logistics use case, what we described sounds pretty simple, but as you say, demos are easy, engineering is, or production is hard. What was hard about getting from 90% reliability to 99.9% reliability? Like, concretely, what kind of stuff were you solving?
Keller Cliffton:
I think that with hardware, everything that can go wrong does go wrong. First, you have to get a design that works and makes sense, and I would say…
John Collison:
Because you have earlier design iterations that were wrong in subtle ways.
Keller Cliffton:
Unbelievable or not-subtle ways, in retrospect. Wow, we’re like complete idiots.
John Collison:
What were the silly early design…
Keller Cliffton:
Yeah. One really good example with Platform 1 was when the aircraft would land, we would swap the battery. The advantage of swapping the battery was we could then have the aircraft back on the launcher and back in the air three minutes later. We put the GPS on the flight computer, which would seem to make sense as a place for GPS…
John Collison:
But it has to reacquire GPS every single time.
Keller Cliffton:
So see, you’re smarter than we were basically. So the problem was that you get an order, and it’s like wow, there’s an emergency. A patient needs this blood transfusion, and so every second counts to immediately grab the aircraft. We put the battery in…
John Collison:
Give me a minute…
Keller Cliffton:
The GPS turns on, and now the GPS is going to take 20 minutes to acquire all the satellites necessary for the vehicle to pass preflight and launch.
And so we realized how stupid this was and immediately iterated and put the GPS on the battery itself. So it didn’t make sense from a bomb cost perspective, because now you need two GPS units for every aircraft that’s flying because one is sitting on the battery charger, but it really makes up for it because you’ve just doubled your cycle time on the aircraft. So there are a million examples. I mean another example, this stuff is so unfancy, it’s almost embarrassing to describe, but when we initially launched Platform 1, the aircraft had 43 different kinds of fasteners on it. Turns out that it is extraordinarily difficult to keep 43 fasteners in stock at all these different distribution centers where you’re doing constant maintenance actions on them. So the next version of the aircraft had two total kinds of fasteners on the entire aircraft. So a lot of this is like, it’s not fancy, but it’s hard to learn without, you just kind of learn by doing.
John Collison:
Is there a component that’s been really annoying? If you could wave a magic wand and get a better version of some component, what would you ask for a better version of?
Keller Cliffton:
I mean every component we have iterated on many, many times, I would say today probably more than anything is servos. And in fact, Zipline went and designed its own servo from scratch. We also ended up going and designing almost every component of the aircraft. It was just kind of a matter of when it made sense to do it. But over time, almost every component of the aircraft, every major subassembly has been designed completely from scratch.
John Collison:
Why do servos frustrate you? Servos for context are actuators basically, things that turn electrical power into mechanical movement.
Keller Cliffton:
I think servos are hard just because there was no industry. I think the way to think about, people think about drones, and they’re like, oh yeah, it’s a drone. But the reality is you have a $1,000 quadcopter that will take pictures and fly for 20 minutes, and the reliability and safety of it really doesn’t matter at all. And then you have a $50 million Predator aircraft, which is actually surprisingly also very unreliable. Famously I read this statistic, which is 50% of US military drones have crashed due to operator error.
John Collison:
Oh my God.
Keller Cliffton:
Read these stories. They didn’t realize they were flying upside down. It is crazier than you think.
John Collison:
The DJI inbuilt collision avoidance is now much better.
Keller Cliffton:
Not working. I don’t know…
John Collison:
The new ones will.
Keller Cliffton:
Yeah, I think that the reality is those are two different sides of the spectrum. And what Zipline really needed was, we think a lot about automotive grade, which is actually almost think of it like a sacred area of cost and quality because automotive grade is super. It is generally very low cost and very, very high quality and reliability. And that’s really where servos are. You can get servos that are totally crappy that you’d use in consumer toys, and then you can get extremely expensive servos designed in Germany and they cost $1,000 each, and they’re way over-engineered. But what we needed was something that was extremely cost effective and it would last for a million cycles, and that doesn’t exist in the world.
John Collison:
And so you’re now producing your own servos.
Keller Cliffton:
So we designed it completely from scratch, and now that’s what we use in Platform 2.
John Collison:
That’s wild. Will you end up just selling drones for other applications? Did you know Boeing and United used to be the same company? It was full vertical—vertically integrated aviation where…
Keller Cliffton:
I did not know that. Wow, I feel like I’m embarrassed.
John Collison:
The airframer and the airline and then they were split up in that there was a big antitrust. Antitrust was emerging as a domain kind of at the time, and as part of that they were forced to split it up anyway. You are the Boeing and United, right? You provide the service and you produce the aircraft. Might you ever just sell the aircraft?
Keller Cliffton:
I don’t think so, for a number of reasons. I mean I think the most important one is kind of what I was alluding to before, which is none of our customers care about drones. If we were out there trying to sell drones, I mean it just is not solving the problem that people actually care about. Walmart, Cleveland Clinic, the government of Rwanda, none of these customers want to be managing a fleet of autonomous aircraft.
John Collison:
Oh no, I don’t think you’d sell the drones to your current customers. I think there’d be new customers where maybe I want to build a crop-dusting service and your drones would be the best for doing crop dusting. And so I buy your drones, and they have all the safety equipment and the high-quality servos and things like that, but you’re not currently in the crop-dusting market.
Keller Cliffton:
Yeah, I mean the second reason I was going to say is I just don’t think selling hardware in that way is a very good business.
Keller Clifton:
Okay.
Keller Cliffton:
But I also think the third thing to consider is that it’s a bit like asking someone, it’s like, “Oh, Jeff Bezos, early days of the internet, are you also going to build other internet businesses?” It’s like the internet is just a technology platform, and Amazon was building obviously a new kind of retail company. And I think that drones are a technology platform. Zipline is a logistics company. We are really focused on building automated logistics for Earth in a way that you want to build the first logistics system that serves all people equally. We would happily use any different kinds of tech. If someone invents teleportation tomorrow, we’ll switch to that because that would be better.
John Collison:
Ah, you’re a logistics platform, and the big unsolved space was drones. Will you ever do a modality that is not drones?
Keller Cliffton:
Very possibly.
John Collison:
What would that be?
Keller Cliffton:
We have lots of ideas.
John Collison:
Have a few more sips.
Keller Cliffton:
Yeah, exactly. I think that ultimately if you really want to be a new kind of logistics provider, you probably have to be even more integrated. Certain things are not going to go via the air, certain things are going to go via the ground.
John Collison:
Okay, so you can get into multimodal…
Keller Cliffton:
Yeah, you should be multimodal. UPS is multimodal. Perhaps Zipline should be multimodal.
John Collison:
Keller is describing here how Zipline got its start in Africa, but they’re now up and running with drone deliveries in the United States. And this is a technology pattern you see again and again—drone delivery, mobile payments, WhatsApp, QR codes, these are all technologies that started in emerging markets first, and it’s going to be the same with stablecoins. So here at Stripe, we’re all in on stablecoins. We recently acquired Bridge, the leading stablecoin orchestration platform, and Privy, the best crypto wallet infrastructure. We’re already seeing a ton of stablecoin adoption, especially with companies that want to build cross-border financial products. We’re working with companies like Ramp to build a global stablecoin-backed spend card, SpaceX for global treasury management, and Shopify to enable payments on Shopify stores from every country. And then breakout startups like Chipper Cash and DolarApp, they’re offering dollar access to people in Africa and Latin America.
But we don’t think stablecoins are going to stop there. Stablecoins are a programmable, low-cost, high-throughput financial system that’ll have applications everywhere. I know from my experience when I talk to CEOs, they’re all interested in the applications of stablecoins. Stablecoin supply is up 50% in the past year alone. They’re definitely not a flash in the pan. So if you’re a company building with stablecoins or you’re just thinking about your strategy there, come talk to Stripe.
We have been describing Platform 1, which is like we said, fixed-wing, catapult-launched, sky-hook recovered, operating for the last decade in Rwanda. You now have just launched Platform 2, which is vertical takeoff and landing, looks more like a traditional multirotor drone that people might picture when they hear a drone, though there’s also a wing?
Keller Cliffton:
Yeah, it’s a hybrid.
John Collison:
Okay, so I forget what you call these, but yeah, there’s a wing and there’s—
Keller Cliffton:
Vertical takeoff and landing. So VTOL…
John Collison:
Yeah, VTOL fixed-wing thingy. And you are using this for deliveries in suburban America. Many questions on this. First off, why construct a whole new modality? It seems to me that what you were doing with the precision airstrikes of little packages under parachutes would also work for my burrito. So why not do Platform 1 with my burrito?
Keller Cliffton:
I mean in fact it did. The crazy thing is that when we initially started building Zipline, we kind of thought, oh, we’re going to have to build this very fancy complex technology. It’ll have to be a vertical takeoff and landing fixed-wing vehicle. We’ll have to be able to deliver things very precisely and gently. And then we were talking to all these early customers, we realized—and this is the power of just going and talking to customers directly—They only cared about cost and speed. They said they wanted to cover maximum range, they want it to be fast, and they want it to be really cheap. And so we kind of ended up thinking like, hey, perhaps we should just design this very simple fixed-wing system that would be, to a certain degree, very unfancy. When you see the way we’re delivering there, it’s just like this cardboard box with a wax-paper parachute. It is amazing to me that that system worked that well and scaled for 10 years to 5,000 hospitals and health facilities and became the largest commercial autonomous system on Earth. And when we say a hundred million commercial autonomous miles, I don’t think people fully get, that’s going to the moon and back 200 times.
John Collison:
And I actually, I always find comparisons of, to the moon and back, really not—
Keller Cliffton:
Not that helpful? Okay, here’s another…
John Collison:
Not that helpful. No one knows where the moon is! I presume your average… is order of a hundred miles in Africa. And so you’re basically talking a million flights, I think.
Keller Cliffton:
It’s like 1.5 million.
John Collison:
Yeah, 1.5 million flights. So I think that’s a pretty good number.
Keller Cliffton:
So anyway, not to digress, but it’s kind of amazing to me that Platform 1 scaled as far as it did; this delivered experience was super unfancy. And ultimately as we were scaling that crazy through 2020 and 2021, there were all these big customers in the US who all were kind of saying, “Hey, we want teleportation. We would use that here in the US.” And so I think that this was kind of our realization that it was time to expand basically and say, okay, “If we were going to build a product that can deliver directly to homes, what does that need to look like?” And the design considerations for that are pretty different in our opinions. We need to integrate really well with existing buildings, build distribution centers like we did with Platform 1. We need to be able to basically plug into the side of any existing hospital or primary care facility or warehouse or even restaurant. And then from there you need to be able to deliver to homes in a way that was fully automated, zero emission, 10 times as fast. And most importantly, we want it to be silent. We think that people, this is a kind of important thing. I think sometimes people think drone delivery, “Oh my god, it’s going to be so annoying.” You have this nonstop buzzing of drones. That is how people experience drones today. But the reality is that it doesn’t have to be that way. You can design systems that are actually way quieter than receiving a delivery via car.
John Collison:
And the crazy thing, sorry, that I forgot to mention in my description of Platform 2 is that the drone does not come. The aircraft does not come anywhere near your house. It sits up at 300 feet above your house and then long-lines the payload down to you in a little droid thingy and that drops it off. But as a result, the drone is hovering 200 feet above you, but I presume it’s just inaudible, basically.
Keller Cliffton:
Yeah, it’s nearly inaudible.
John Collison:
Yeah.
Keller Cliffton:
It looks crazy. I was talking to an investor the other day.
John Collison:
It sounds crazy.
Keller Cliffton:
Yeah, I know. And I was talking to an investor the other day, he wrote us this nice email and he said, “Hey, we really want to meet you because we think Zipline has built a fundamentally jaw-dropping consumer experience.” I do think when people come and see it for the first time, it’s pretty wild. You just never really experienced something like zipping down to you from a football field over your head.
John Collison:
It’s like a science fiction. Yeah.
Keller Cliffton:
It’s kind of science fiction. But the advantage is, that droid that you described, that little robot on the end, is actually controlling its position in x and y axis. So this is really important. It’s often windy, it’s going to be blowing. You have different environmental conditions you’re trying to control for, and we want to be able to deliver with dinner-plate-level accuracy to any home. And so this enables us, even when the wind is blowing really, really hard, this enables us to very reliably, safely, and precisely deliver something even in a very, very tight area.
John Collison:
Okay, so the reason that you needed to develop a platform for suburban deliveries in the United States was 1) the place you’re delivering from, the Chipotle, you probably can’t fit a catapult there, at least not in every one.
Keller Cliffton:
I’m pretty sure you can’t.
John Collison:
And some of ’em you could, but there’s a parking lot, but not reliably. And then for the target, again, it might be accurate enough, but the new system is definitely accurate enough.
Keller Cliffton:
Yeah, I mean, in fact, the very first Walmart store that we ever integrated, we actually integrated with Platform 1. And that was even more shocking to me. The customers loved basically having their stuff bombard them from 30 feet up. I mean it’s honestly crazy. I think that was the clearest signal to me of like, man, if people were willing to put up with that delivery experience to get something delivered autonomously, wait until they see Platform 2.
John Collison:
So the reason that we haven’t had drone delivery in the United States for a long time is the FAA did not provide any rulemaking on Beyond Visual Line of Sight operations. And they’re your regulators, so you probably can’t criticize them in public. I’ll criticize them for you. I think the FAA was slow and roundly criticized for being slow in not having any commercial drone frameworks. The old Part 107 stuff for hobby drone operators and line of sight is the test where it has to be a light drone, less than 55 pounds, and you have to be there controlling with your remote control, which obviously is not viable for commercial drone operations. So for the longest time they sat on commercial drone operations and didn’t do anything. Just now in the last month, we’ve seen Sean Duffy, the secretary of transportation and acting head of the FAA, announced new Part 108, Beyond Visual Line of Sight commercial drone rules, or at least the notice of proposed rulemaking. What do you think of the current FAA and the rules they’re proposing?
Keller Cliffton:
Yeah, I mean, first of all, Part 108 is kind of codifying something that has already been happening over the last three to four years, which is I think that about four years ago—on an ad hoc basis and via exemptions—which is that about four years ago, Zipline started seeking regulatory approval in the US. We were able to go to the FAA and say, “Hey, we have a hundred million commercial autonomous miles and zero safety incidents. We think at this point you have the data you need.” And they agreed. And so very much to their credit, this is what really unlocked Zipline starting to scale aggressively in the US. And we’ve done that via exemptions. There are things called like 44807, which is a different process rather than getting an aircraft type certified. If you really want to talk about something that’s broken, we could talk about type certification—
John Collison:
Of drones or aircraft?
Keller Cliffton:
Both.
John Collison:
Okay.
Keller Cliffton:
Yeah. I mean so far—
John Collison:
Did you see the MOSAIC stuff?
Keller Cliffton:
I am not an expert…
John Collison:
It’s amazing! They fixed Part 23…
Keller Cliffton:
It is really cool.
John Collison:
…certification by making Part 22 much bigger.
Keller Cliffton:
A type certification has been this interesting blocker, which is if you wanted to operate commercially in the US, you couldn’t operate under 107, or if you wanted to operate commercially, nonvisualized site, you had to operate under Part 135. I fear that we’re getting, we’re really in the weeds now, but I guess the main point here is that you needed to get an aircraft type certified. And this is kind of the definition of a very painful regulatory process. It takes five to ten years to get aircraft type certified, and it’s really based on this idea of, a regulator is going to look at every component of the aircraft and independently verify that that component is meeting certain standards that were set by the regulator. That might, I mean, imagine trying to do that in a world where every year you have a radically new and better version of technology. It just doesn’t work.
John Collison:
Totally. Yes. You guys are doing progressive hardware development and constantly figuring out that two fasteners is better than 43 and other such innovations.
Keller Cliffton:
You just have really, really fast iteration. And so in a world where type certification takes five to ten years, you’re dead on arrival. And that’s kind of what we saw for five or ten years. None of this innovation happened in the US. I think the good news is that, in 2020, the FAA kind of woke up to this, they realized like, hey, this is going to become one of the most important industries in the world. The US needs to be a front-runner here. We cannot be a laggard. And the FAA actually started to move super fast. I mean the FAA has been, I know you think maybe I’m just saying this, but the FAA has actually been a great partner. There are so many people there who want to win and who want to see the US win.
John Collison:
I think the FAA is full of awesome people trying really hard. I think [progress] has been slow, and I think exemptions and waivers are not a substitute for a good regulatory framework.
Keller Cliffton:
Yeah. Well there you go. I mean, 108 really is, I think there have been these exemptions made that luckily enabled Zipline to invest aggressively in the US over the last four years and experience pretty explosive exponential growth, which we can talk about. But 108, it is cool to see that now be codified. That is just the beginning of a process, right? Because now the FAA has nine months to actually operationalize and execute based on that rulemaking. And we’ll see how that goes. But I think it’s super important. I think you and I probably both have grown up reading about the SR-71 program or a lot of these programs to design airplanes. There was a world where the US was iterating incredibly fast in aviation and aerospace, and we’ve largely lost that. Our grandparents got to grow up in a world where they could fly on supersonic jets. We can’t. I think people have this sense that the world moves inexorably forward. It doesn’t. Didn’t you have a tweet about this, maybe? I think about it every day. Actually, no. Sometimes we literally move backwards. And in aviation, what plane did you learn to fly on?
John Collison:
I learned to fly on a Cessna 152.
Keller Cliffton:
And how old was that aircraft?
John Collison:
Yeah, it was from the 1970s, I want to say. Yeah.
Keller Cliffton:
What the hell?
John Collison:
No, I was actually just ogling the Carbon Cub UL, which uses a new Rotax engine. So there’s this awesome new engine technology out of Germany, this company Rotax, that makes them there. And it hasn’t been in the US certified market because it hasn’t been worthwhile to go get the engine certified. And now they’re selling it in the LSA regime because…
Keller Cliffton:
Light sport aircraft.
John Collison:
Exactly. It is possible there. And so I think we’re just going to see this wave of interesting new, because it’s the kind of pent-up technologies that haven’t been certified yet, such as Rotax engines, which are better everywhere.
Keller Cliffton:
Have we lost like 98% of the audience? I talked about 108 and 135, I guess. But just to share.
John Collison:
No, but it’s specific, it matters.
Keller Cliffton:
Just share one thing to try to make it even more general-interest. What really inspires me is you think about Skunk Works, right? And the teams that built a lot—
John Collison:
Is the Lockheed Martin factory that made or kind of group, tiny group that made all the fun news Delta aircraft, F-104 and stuff like that.
Keller Cliffton:
SR-71 even more famously, like the U—
John Collison:
U-2.
Keller Cliffton:
Yeah, exactly.
John Collison:
And they didn’t make the F-104, they made the F-111, sorry.
Keller Cliffton:
Yeah, and just to give a specific example, a couple examples. One, the first modern passenger aircraft that was ever built, the DC-3, I’m going to get everything right. DC-3 was built originally, the entire design cost for that aircraft was like $2.5 or $3 million. So even when you convert into today’s dollars, that’s like $75 million to design a passenger airliner. The 787 costs $32 billion to develop. Oh yeah, DC-3 was like napkin-sketched to finish aircraft in two years. The 787 took 20, 25 years, $32 billion. It’s literally become 500 times as expensive to build an airplane as it was for our grandparents. And I think the crazy thing is you’re going to be like, oh, well, it’s like the 787 is a much more complicated airplane. The iPhone is also much more complicated than was five years ago.
John Collison:
Yes, yes. So if you think about the causes, I think there’s a great point to that. The 787 certification process is so much more tortuous than the DC-3. If you think about why that is, I think there is a few obvious things you could point to. One is, and the one people would point all the time is, we have much less tolerance for unsafety than we used to. The second is the certification approaches are much more prescriptive. And so they require that you do specific things to prove how you’re going to meet various tests and stuff like that. And then maybe you could say that just the paper turnaround times have gotten slower. There’s a lot of back and forth. If you could wave a magic wand to fix aircraft certification, either across drones or manned aircraft, again, you’re put in charge of the FAA, what would you actually do?
Keller Cliffton:
I think the answer is actually really obvious. And the good news is, I think this is largely already happening, which is that we need to move away from prescriptive regulatory requirements that are like, “Okay, now show me the life-support system for the pilot. Should we lose whatever atmospheric pressure in the cabin?” It’s like, it’s a drone. There is no pilot, and then the FAA looks at you, like, I don’t know, we can’t move forward. I think there are a lot of things are just, you have all these prescriptive rules about how we’re going to—
John Collison:
Need performance-based…
Keller Cliffton:
You need performance-based. So basically everything needs to transition to a world where the FAA is setting a statistical level of safety. And from there, any company can basically self-certify against that level of safety. And you can imagine different ways of what kind of data you might have to provide to prove that you’re meeting that level of safety. But there are a lot of different ways to build something that’s safe. And what you want is engineering innovation. You want people trying new things. If you try something new, and it’s an amazing innovation, it’s suddenly blowing it out of the water with a prescriptive, absolute regulatory framework, you lose all of that because it’s basically like the FAA is just looking to make sure every subsystem, that’s the way we’ve been designing airplanes for a hundred years.
John Collison:
The subsystem by subsystem stuff is crazy rather than performance-based.
Keller Cliffton:
It’s really painful. So anyway, I know this might be a little, whatever, shop talk or whatever, but I think that it’s really important. And I also think you can probably see similar versions of this battle playing out in many different parts of the US government. When you look at the way we build nuclear power plants, we created, what is it called? The nuclear, is it the—
John Collison:
Nuclear Regulatory Commission?
Keller Cliffton:
Nuclear Regulatory Commission, started in, I think, 1978. The commission has started to ensure that power plants are safe, and since then we’ve built zero power plants, I think. Or is it like one? I think it’s zero.
John Collison:
Yeah. Yeah, it is definitely the rates slowed down.
Keller Cliffton:
It’s a bit painful! The commission achieved its goal though.
John Collison:
Yeah.
Keller Cliffton:
Because we built zero, and that’s painful as an American.
John Collison:
Maybe you’re describing failure to engage with tradeoff space in an intellectually honest way where there’s a tradeoff between safety and cost, and you just have to actually address those questions, and there’s a tradeoff between speed and safety, these things.
Keller Cliffton:
And the cool thing in aviation is that there’s now 50 years of pent-up technology that has totally not been put to work at all. It’s actually very similar. I think it’s why SpaceX is this enormous success because they were like, oh yeah, we’ll just actually use modern microcontrollers or modern semiconductors. Whereas ULA was building on a semiconductor architecture that I think was like 30 years old. They just didn’t want to certify a new flight computer.
John Collison:
And so you’ve launched in Texas now, just Texas, in the United States?
Keller Cliffton:
Well, we launched originally in Bentonville, and now we’re expanding in Dallas.
John Collison:
So you’re expanding in Dallas and delivering for people like Walmart and Chipotle, and where else should people be ordering from in Dallas?
Keller Cliffton:
I mean, you can just go to the Zipline app.
John Collison:
Okay.
Keller Cliffton:
Yeah, and we have all the different storefronts on there. You go to Zipline—
John Collison:
But those are the biggest ones?
Keller Cliffton:
Yep, those are the biggest, and I mean to give a sense right now we’re launching—
John Collison:
Yeah, give me your explosive growth stats.
Keller Cliffton:
I mean, it’s kind of crazy. It’s really only the last three months, but we’re now launching a Walmart Supercenter every week. Those stores are ramping extraordinarily fast. I mean, I don’t know exactly how much I’m allowed to share. It’s probably more for Walmart to share…
John Collison:
Just give us some more of the magnitude.
Keller Cliffton:
I mean, in order of magnitude, the service is becoming quite large relative to the way that people receive.
John Collison:
How many drones [are] you flying in the Dallas area?
Keller Cliffton:
So right now we are integrated with about 10 Walmarts, and we’re launching 1 a week, and each Walmart will have somewhere between 8 and 20 Zips flying from it.
John Collison:
So it gives you hundreds of drones now delivering in the Dallas area.
Keller Cliffton:
And I mean the crazy thing is the service.
John Collison:
And can do hundreds of deliveries a day? Tens, high tens?
Keller Cliffton:
Yeah, high tens is the way I think about it.
John Collison:
So you have the capacity for a thousandish, thousands of deliveries.
Keller Cliffton:
Yep. That’s the way I think about it. And that’s growing rapidly. We’re actually expecting to get to 10,000 deliveries a day in Dallas. That’s actually the company’s goal. We’re going great. Moving as fast as we can toward that. Looking at the statistics, the service has been growing 25% to 30% week over week for the last three months. And the interesting thing, the last couple of things, I mean really it’s only the last three months that, because we only launched in Dallas on April 1 or April 4, I think. So seeing it grow like that, the crazy thing is that about a month ago we were starting to have all these capacity issues. We can’t do as many deliveries. It’s growing really fast. So we decided to turn off all the marketing. We turned off the in-app notifications, we turned off the field service marketing. Basically all the demand generation marketing we’re not doing and the service just, I mean we hit new records…
John Collison:
Because it’s just a much better product.
Keller Cliffton:
Yeah, Saturday we hit a new record, all-time daily delivery record and then on, two days ago, and then yesterday we blew that record out of the water by 20%. It’s crazy. So I think that, a couple takeaways, one is that it’s amazing how quickly science fiction becomes—
John Collison:
Normal.
Keller Cliffton:
Entitlement.
John Collison:
Yeah. We see this with Waymo in San Francisco.
Keller Cliffton:
It’s exactly like Waymo where people get in and they’re like, “Oh my God, autonomous car. Insane.” And then three minutes later they’re scrolling on Instagram, and I mean when you go to Dallas and just—
John Collison:
What’s the version of that in Dallas?
Keller Cliffton:
I think the version in Dallas is that for the first day that we launch a new Walmart, you’ll see people on the hoods of their cars just watching, like people are kind of taking note, and then three days later people are just totally going about their normal lives, do not care. You’ve got this installation of autonomous aircraft, they’re just coming and going nonstop at crazy volume.
I think the other example is I visited a 78-year-old grandma who’s ordered 340 times from Zipline over the last year.
John Collison:
I was going to ask, are you seeing induced demand?
Keller Cliffton:
This is induced demand.
John Collison:
So people are ordering more stuff.
Keller Cliffton:
It’s crazy. She’s been I think the county clerk for 40 years straight, and she’s just showing me, she’s like, “Oh yeah, let’s place an order right now.” And she’s around on her iPhone putting everything into her basket. She’s like, double-click the side of the phone, Face ID, Apple Pay. She’s like, “It’s on its way. It’ll be here in eight minutes.”
John Collison:
And she’s 78.
Keller Cliffton:
She’s 78. And she’s like totally living in the future. And you talk to moms who are like, “Oh yeah, this buys me back three or four hours a week. I can just be with my kids or take care of the family rather than be out in a car trying to shop or get something.” So I think a lot of families we’re seeing now ordering three to four times a week on average, where they’ll do a big grocery run once a week, and then they can do three to four. Sometimes they talk about ’em as top-up visits, where it’s like, if you don’t know what you’re making for dinner or there’s something unexpected or you have people over, you just get everything you need delivered autonomously.
John Collison:
Why can we not have this in California? Why are you in Texas and not California? I would’ve thought from a regulatory point of view, this federal preemption of the airspace. And so I would’ve thought all places are equal. Why is Texas a more hospitable environment for this than California?
Keller Cliffton:
You could ask the same question about Rwanda. I mean, I think that Zipline has probably been—
John Collison:
Rwanda had a different regulatory framework. Drones are banned in the US, whereas drones aren’t banned in California.
Keller Cliffton:
No. However, a lot of what we do is still about a partnership with the community and the government.
John Collison:
I see. You need a permit for the base or whatever.
Keller Cliffton:
Still need a permit, permit to build. Yep, exactly. We’re still permitting the ground infrastructure, for example. It’s a bit like Tesla building supercharger infrastructure in the US.
John Collison:
So who do I need to write a letter to? Is it by mayor? Is it by local representative?
Keller Cliffton:
We’re working on it. Yeah, we’re working on it. But to be honest, as someone who lives in the Bay area, and so 100% feels your pain, it’s like why is this not serving my house right now? The vast majority of technology has first focused on serving whatever the elite cities on the coast of the United States. I actually think it’s okay that for once technology is going.
John Collison:
Because this works better in a low density area…
Keller Cliffton:
It’s going the opposite. Well, it’s actually, I mean Dallas is very high density. I mean it’s not Manhattan, but it’s probably denser than San Francisco. I wouldn’t say it’s as much about density. It is a lot about, we typically want to launch somewhere where people are freaking psyched to have us and where there isn’t, I do think there’s a little bit of a flavor. I mean, you had Kyle on here and seeing what crews went through in San Francisco with the conning of the cars sometimes are
John Collison:
Our war on electric scooters here in San Francisco. Hey, but we have a new mayor, Daniel Lurie…
Keller Cliffton:
Hey, I’m not judging. I know he’s doing awesome. He’s doing great. I think that, look, I just think that sometimes cities might, if they just assume that they’re always going to have the best technology and they don’t have to do anything about, or I don’t know. I think that there are a lot of cities and states out there that are 100% aggressively courting infrastructure, new solutions. They want high-paying jobs, and they’re unapologetically excited about new technology.
John Collison:
So you’re going to go to the cities that want you first and then you can get around to the cities that are more ambivalent.
Keller Cliffton:
I think what’s going to happen is that, at some point, again, there are certain cities that are just extremely forward-leaning. An example—I brought, one of our board members from San Francisco leads one of the biggest funds in the world here in San Francisco to Dallas. And he was looking at what we were doing, and I could see the smoke coming out of his ears. He was like, “Oh my god, you guys are just doing things.” We went to a site where we were building charging infrastructure and we had literally broken ground three days previously, there was a huge hole in the ground, and he was like, “So when is this going to be online?” And I was like, “It’s going to be online in 15 days, before the board meeting,” we had a board meeting in 17 days, and he was like, “That’s insane. It is not going to be ready for the board meeting. What are you talking about?” And it was actually ready five days before the board meeting. Basically different kinds of cities have different requirements, and there are places in the US where you can get things done super, super fast.
John Collison:
Yeah.
We are here in South San Francisco just down the road from your aircraft-manufacturing facility. Tell me about that. How do you build them? How many are you building? How many will you build?
Keller Cliffton:
The most important thing is the speed of the engineering iteration. I think you’ll appreciate this. That’s something I think that’s been a very well-understood concept in software companies, not so well in hardware, way harder to do in hardware. And so the way to do that is full vertical integration and then having engineering practically on top of manufacturing so that as you’re rolling hardware changes through onto the line, you have the engineers who design those changes working with manufacturing engineers and then working with the people on the line itself to figure out how to make that work and get it into production. And so even though it seems like a bit of a weird thing to say, yeah, we’re producing all these aircraft here in South San Francisco, the value of us being fully integrated in that way, what it does for the speed at which the plane can iterate and evolve the product is way more valuable than like, oh, it’s more expensive to hire someone in South San Francisco than it is in Nevada or something like that. So it’s an easy trade. So yeah, this facility, and we just expanded to 160,000 square feet, it’ll be capable of building around 50,000 aircrafts a year at scale, which should get us to near-national scale in the United States.
John Collison:
50,000 aircrafts a year. And how many aircrafts will a medium-sized city require to support it?
Keller Cliffton:
I think it’s around a thousand. It depends on the city. Cities come in a lot of different shapes and sizes.
John Collison:
Yeah. Okay. Yeah. So 50,000 a year is really a lot.
Keller Cliffton:
Again, it depends what you’re comparing to. If you’re comparing to iPhones, not really, but if you’re comparing to airplanes, then yeah, really a lot compared to the Cessna that you learned to fly in.
John Collison:
Yeah. Okay. And so you’ll be able to, in your current production facility in South San Francisco, produce 50,000 drones a year.
Keller Cliffton:
Three shifts, 24-7 operations. We already operate our test sites 24-7.
John Collison:
What do you do with the test sites? What do you test?
Keller Cliffton:
Yeah, I think when people come visit the test sites, I think that’s one of the things about Zipline that is most mind-blowing to people. And I think it’s, it doesn’t really look like anything people have seen before. People often ask us, they’re like, wow, 120 million commercial autonomous miles and zero safety incidents. Why don’t other people do that? It turns out that designing something that’s safe is hard. You have to think—
John Collison:
Reusable rocket. That’s a good idea.
Keller Cliffton:
Yeah, exactly. It’s like, yeah, everybody should just do it. It turns out some of these things are hard to do, and I think the way you do it is it’s all just like none of this is fancy. It’s just practical problem solving. And so for us it was having three major layers of testing and validation. So the first layer is software testing. We do a huge amount of testing at ten or 100 times real-time speed to test all the different software systems that we are constantly iterating. And there’s ten times as much software as there is hardware.
John Collison:
So this is the airplane avoidance you were describing. It’s like you input a signal of like, oh, you’re near an airplane and check that it avoids the airplane.
Keller Cliffton:
Yeah, autonomy, computer vision, guidance, navigation controls, communications, architecture, all the backend cloud services. These can all be, you can do a huge amount of software testing because I think the other major way to think about what is the problem we’re trying to solve, do you know how long it takes Boeing to do a software update?
John Collison:
Long time, I’m guessing. No.
Keller Cliffton:
You don’t want to guess?
John Collison:
I’m too horrified.
Keller Cliffton:
Come on, guess.
John Collison:
Like, from when to when?
Keller Cliffton:
For example, from when they realize that their planes are falling out of the air and killing people to when they can then have the software update out. And so the airplane’s back in the air.
John Collison:
I mean those, order of a year?
Keller Cliffton:
Three years.
John Collison:
Oh, that was that long?
Keller Cliffton:
Yeah. Again, this is part of the problem of how the FAA regulates software with those kinds of aircraft, but Zipline right now is doing a full new version. We’re releasing new software versions to the entire global fleet every 30 days. And so validating that, basically, it might be 50 to 100 bug fixes, maybe 20 to 50 performance improvements, and then five to 20 major feature additions or upgrades. And that gets validated through the software-testing layer. Then it gets validated through HITL, which is Hardware in the Loop testing, it’s basically all of the avionics of the aircraft assembled on shelves that are kind of like the brain of the aircraft connected to the matrix. They’re all operating in simulation. We can detect like 95% of what’s going to go wrong on an aircraft we can detect in HITL before we then go to the final layer, which is the test sites. The test sites have hundreds of aircraft operating 24-7, so they never stop, even Saturday at 3:00 a.m. you could go to the test site and you’ll see the aircraft are just there and there are humans overseeing them.
John Collison:
You’re describing your build manufacturer-testing infrastructure in the United States. One of the top discussed topics right now is American dynamism, the re-emergence of American manufacturing. And when people talk in DC about the importance of American manufacturing and the criticality of the supply chain in the United States, the example they often point to as a supply chain that we have outsourced is obviously drones, where there is not as serious…
Keller Cliffton:
I thought you’re going to say semiconductors.
John Collison:
Oh, well semiconductors.
Keller Cliffton:
Semiconductors is even more scary.
John Collison:
Semiconductors gets talked about a lot. Yeah, yeah, for sure. But there’s the headlines about Intel and everything, but people also talk a lot about drones and just how much more scaled up the Chinese drone industry is and the US drone industry. People also talk about the fact that it’s not competitive to manufacture in the United States anymore because supply chain. Did you read the Apple in China book? It was like a pretty good.
Keller Cliffton:
I’ve heard of it. I haven’t read it yet.
John Collison:
It’s great. And it’s like a good history of Apple manufacturing in China, and they describe kind of the emergence of it, but in particular, people sometimes thought of China as just low wage. Whereas the book correctly describes just how much agility there was in adapting supply chains and being able to manufacture there. So you are a company manufacturing at scale in the United States. What do you think of this topic?
Keller Cliffton:
To put a finer point on it, in a week, China makes more drones than the US makes in a year.
John Collison:
That is a finer point.
Keller Cliffton:
It’s a little spooky. And obviously you don’t have to be a deep student of global events to understand that, wow, this technology is clearly kind of a remaking, not just industries, but remaking warfare and military power.
John Collison:
Yeah. And people watch the war in Ukraine where it is now all drone-determined from an airspace.
Keller Cliffton:
Totally, a hundred percent. Everything is attritable and you wouldn’t want humans on the battlefield at all. And I think this is a good news, bad news thing. So the bad news is, yeah, it’s true. China has massive manufacturing advantages and massive scale in this area that the US does not currently have. But that is mainly for plastic quadcopters that cost a thousand dollars. It’s definitely a different class of aircraft. When you look at where this is all going, the systems that are really going to matter, whether it’s military or logistics, they aren’t thousand-dollar quadcopters. They’re more complicated technologies. They require really strong integration of software AI solutions with the hardware itself. And again, the hardware is only 10% of the complexity of the overall solution. So in those areas, the US is actually leading, the US is leading both in terms of I think drone delivery and logistics, but also leading on the software side.
John Collison:
That’s right. Those areas being…
Keller Cliffton:
I would say more advanced hardware software integrations that are required for really good products for military and really good products for logistics.
John Collison:
Okay.
Keller Cliffton:
So when we’re talking about the 50x, that’s because you have DJI in China or they’re building tens of thousands for drone light shows, those kinds of things. That is not necessarily the battlefield that is going to determine who really is the front-runner in the industry that matters the most. And so I think that it’s too early to say, I think China has obvious advantages, and that they have massive manufacturing scale. You mentioned the Apple supply chain book. That is a huge reason. The fact that Apple outsourced so much of that manufacturing and so much of that knowledge and expertise is now in Shenzhen. Now Shenzhen’s ability to build phone-adjacent products, which I think you could definitely describe drones, even cars you could describe as phone-adjacent now as Tesla will tell you, or Tesla could attest. It’s actually those supply chains that make it possible to then branch out and build a lot of these other products way faster. So that’s the advantage that China has. The advantage the US has is a lead in AI, a lead in more advanced hardware and software integrations. I don’t know who’s going to win.
John Collison:
Yeah, the point the book makes is that Apple’s direct investment in China is much larger, even in inflation just terms, than the Marshall Plan in Europe or it’s just a huge investment in manufacturing capability.
Keller Cliffton:
I was thinking about that on the way here, how that’s going to look with the benefit of hindsight.
Corporations sometimes make the easy short-term decision to maximize profit instead of the more strategic decision of how do we set ourselves up for long-term success. And it really does not seem like it would be that hard to achieve some of the company goals that a company like Apple wanted to achieve, but also ensure that you maintained manufacturing dominance and kept it at home, or at least ensure that you kept that as a strategic option. The fact that we’re now in a position where we don’t have the ability to build our own smartphones period, and where, depending on what happens with Taiwan, we may not have the ability to even build modern computers. It’s spooky.
John Collison:
Last question. You’re one of the success stories that people hold up in hard tech where you’ve built a product that is a successful business line. You found a good way to start that wasn’t too capital-consumptive, you were able to get a real product out there in the field and refine it from there. And now you have a product that people love in several countries. What advice would you have for hard tech founders starting out today?
Keller Cliffton:
It certainly seems to me like the whole American dynamism, what do you call it? Paradigm that we kind of see, I think it’s making things easier. When I moved to the Bay Area in 2013 with the goal of building Zipline, the most exciting company in all of Silicon Valley, probably in the entire US, was a company doing file sharing. It was Dropbox, and they had the best engineers, the best growth story, the most ambitious vision. And it was, I think at that time, like a $5 billion company, which is more money than I could possibly imagine. And that was the pinnacle of what you could achieve in terms of building a startup in Silicon Valley. And so it’s interesting to think that in just 12 years, the scope and scale of human ambition has radically increased. I mean, whether it’s Tesla or SpaceX or Stripe or OpenAI or Brian Armstrong was on here, right? He’s working on NewLimit, they’re trying to make humans immortal, and a lot of these ideas are working and scaling. There are people building nuclear-fusion power plants. I think that’s kind of probably underappreciated, just how much change has happened in just 12 years. And I think that, I guess to answer the question, my biggest advice is that I think especially when it comes to hard tech, the whole challenge is just designing something incredibly simple and getting it into the real world quickly and learning by doing. It’s very easy to be in a bunker or an R&D lab, building cool technology, really hard to force yourself into the real world and then learn by doing. I think that Zipline, we never thought we were the smartest team. We never thought we were certainly not the best funded team, but we always had this deep conviction that we were the most practical team, and we had this mission that really inspired us. We knew exactly what we needed to do. It was like, deliver to those 21 hospitals, and we forced ourselves to launch something that to even our own eyes, we were quite embarrassed. It’s like, “Oh my God, is that, are people going to put up with that delivery experience?” And it turned out that even that very simple version of the product scaled very fast. And so I think that it sounds obvious when you say it, and yet it’s extremely rare, especially in hard tech companies.
John Collison:
Keller, thanks for coming by.
Keller Cliffton:
Yeah, thanks, John.