AI transcript
0:00:00 [MUSIC]
0:00:06 Pushkin.
0:00:06 [MUSIC]
0:00:11 >> Hey everybody, I’m Kai Rizdal, the host of Marketplace,
0:00:13 your daily download on the economy.
0:00:16 Money influences so much of what we do and how we live.
0:00:19 That’s why it’s essential to understand how this economy works.
0:00:23 At Marketplace, we break down everything from inflation and
0:00:26 student loans to the future of AI so that you can understand what it all means for you.
0:00:32 Marketplace is your secret weapon for
0:00:34 understanding this economy, listen, wherever you get your podcasts.
0:00:37 [MUSIC]
0:00:41 Every day around the world, more than 1,000 weather balloons are launched into the sky.
0:00:46 The balloons float high up into the atmosphere, sending back information.
0:00:50 You know, temperature, wind speed, air pressure, etc.
0:00:54 And then, a couple hours later, the balloons pop.
0:00:58 This is basically the way weather balloons have worked for decades.
0:01:02 And the information the balloons send back is really useful for weather forecasts.
0:01:06 But the information the balloons send back is also pretty limited.
0:01:10 Because the balloons only stay in the sky for a couple hours, they don’t fly very far.
0:01:15 If we could figure out how to make balloons stay up for longer,
0:01:19 they could blow in the wind and travel thousands of miles.
0:01:22 They could travel across continents and across oceans and send us back a lot more
0:01:27 data and give us a much clearer picture of what weather is coming our way.
0:01:31 [MUSIC]
0:01:37 I’m Jacob Goldstein, and this is What’s Your Problem,
0:01:39 the show where I talk to people who are trying to make technological progress.
0:01:43 My guest today is Kai Marshland.
0:01:46 He’s the co-founder and chief product officer of Windborne Systems.
0:01:50 Kai’s problem is this.
0:01:52 Can you build weather balloons that stay in the air for weeks or months instead of hours?
0:01:57 And can you pair the data from those balloons with AI to make weather forecasts
0:02:02 that are way better than anything we have today?
0:02:05 Kai got interested in balloons back in 2015.
0:02:08 He was a freshman at Stanford, and he joined the Student Space Initiative,
0:02:12 which sounds kind of fancy, but in fact,
0:02:15 was basically a bunch of college kids trying to build their own weather balloons.
0:02:19 [MUSIC]
0:02:23 The first half dozen flights of our balloons were complete into total failures.
0:02:29 When do things start working?
0:02:31 That would be back in June of 2016.
0:02:35 This is the end of my freshman year, and I think our longest flight that year had been 14 minutes.
0:02:43 So shorter than a normal weather balloon.
0:02:48 – Buy a lot. – Buy a lot.
0:02:50 And also, the person who had started this project, who then became my co-founder,
0:02:59 had just graduated and was about to leave what we thought never to be seen again.
0:03:05 And school has ended.
0:03:07 We can’t even get into the building because our key cards have stopped working,
0:03:12 so we have to tailgate someone in to get our balloon launch equipment out,
0:03:17 go out to the middle of nowhere in the central valley of California,
0:03:22 and if this launch didn’t succeed, Windborn would not exist.
0:03:28 – Okay. – So I can guess what happened next based on that fact,
0:03:33 but tell me what happened.
0:03:34 So it got to be about 4 a.m.
0:03:38 And we’re out in the middle of this park, really cold, and the sprinklers turn on.
0:03:45 They start slowly swiveling around towards our delicate electronics.
0:03:52 And so I have to hurl myself onto the sprinklers, getting soaked, just to save our electronics.
0:03:58 – It’s like the nerd version of throwing yourself on a grenade. – Exactly.
0:04:02 – I really felt like I was in an action movie sequence. – Yeah, good.
0:04:06 But that sacrifice really paid off because with that flight,
0:04:12 we managed to set the world record for weather balloon endurance.
0:04:17 – How long did it stay up? – It stayed up for 76 hours.
0:04:21 – Wow. And your previous best was 14 minutes? – Exactly.
0:04:25 So you’re in college. How do you build weather balloons that are better than whatever
0:04:29 the federal government, NOAA, is launching our state-of-the-art weather balloon people?
0:04:34 So the first big thing is that some new consumer electronics have come out.
0:04:41 In particular, things like microcontrollers and lightweight low-power satellite communications.
0:04:50 And we realized that we can use these innovations to make a balloon smarter and control its altitude.
0:04:59 – What’s a microcontroller? – A microcontroller is a super small computer
0:05:05 that you can put in any device to give it a brain.
0:05:10 – And they’re cheap. – They’re dirt cheap.
0:05:13 Anytime you hear the word “internet” of things, that means microcontrollers.
0:05:17 – And nobody else was doing this at the time?
0:05:19 Like all the big, well-funded weather agencies weren’t sticking microcontrollers on their weather balloons?
0:05:27 – Well, with these big agencies, they are really incentivized to keep things stable.
0:05:35 Because weather is this really important thing that billions of people depend on.
0:05:40 And so they’re not going to say, “Hey, what crazy experiments can we run?”
0:05:46 – Their incentive is just like, “Don’t screw it up.” – Exactly.
0:05:48 – “With the way we did it yesterday, don’t screw it up and you won’t get fired.” – Exactly.
0:05:52 – And so specifically, what was it that you were doing that no one had done before?
0:05:58 – Yeah, so it’s a very simple concept that is really hard to make work in practice.
0:06:05 The way we control our altitude is when it gets too high up, we vent some of our gas.
0:06:12 So it has less lift. When we fall down too far, we drop some ballast, so it stops falling.
0:06:19 But the hard part about this is it has to function at -70 degrees Celsius.
0:06:25 That’s the temperature of dry ice. You have to have a lot of onboard software.
0:06:31 That’s where the microcontrollers come in to do things like understand,
0:06:36 is this just turbulence that you’re hitting? Or is it actually a real change in lift that we
0:06:41 need to drop ballast or vent gas to account for? And in fact, we stayed up 24/7 every single time
0:06:49 it talked to us every five minutes. We had it play an air horn sound because we were,
0:06:55 of course, falling asleep in the middle of the night to see what it was doing.
0:07:00 – So, okay, so you set the record, you stay up for, what, three days watching it? How far did it go?
0:07:06 – It flew all the way across the country and landed out in the Atlantic Ocean.
0:07:13 But the real reason why we decided to form a company around this was because we realized
0:07:21 the impact this could have. We got into this from the engineering side. The lands are fun.
0:07:27 But we realized, wait a second, 85% of the Earth’s atmosphere is invisible to humanity.
0:07:34 Weather is this crazy, unpredictable thing, and we can solve that. And weather is so much more
0:07:42 than just do you bring an umbrella to work tomorrow. It is the single most immediately
0:07:49 destructive aspect of climate change. And improving the weather forecast, it, of course,
0:07:55 helps with adapting to climate change, things like predicting where a hurricane will make
0:08:00 landfall. But it can also help with preventing climate change in the first place. If you have
0:08:06 a better weather forecast, you can better route ships and planes to save a huge amount of fuel.
0:08:13 You can accelerate the transition to renewables because you know when the wind will be blowing
0:08:20 or the sun will be shining. And we looked at this and said, no one else has our balloon technology.
0:08:26 If we don’t do this, no one will. And if we walk away from this opportunity, we can’t live with
0:08:32 ourselves. – I mean, I buy that the stakes are high. I buy that it could be very helpful.
0:08:39 But like, why wouldn’t anybody else do it? – Yeah, a lot of people are trying to improve
0:08:47 the weather forecast. But what’s unique about Windborne is the combination of our balloon
0:08:55 technology with our AI weather forecasts. Because really, those are the two levers
0:09:01 to pull to increase the accuracy of a weather forecast. – You mentioned that 85% of the world
0:09:08 lacks good weather data, which was surprising to me the first time I heard it. But I would
0:09:14 think like, can’t satellites do it? That’s kind of, I guess, my naïve-ish thought. It’s like,
0:09:20 I know there’s a lot of satellites looking down at the Earth all the time. – I’m actually a big
0:09:24 fan of the space industry. I used to work at SpaceX. But the problem is the laws of physics
0:09:29 fundamentally limit what satellites can measure. Take, for example, pressure. It turns out pressure
0:09:35 is extremely important for predicting the weather because it determines where the winds are blowing,
0:09:42 how the weather systems are moving. But a satellite fundamentally cannot measure pressure from space
0:09:49 because it’s not in the atmosphere. It can’t see what’s going on. – So you start the company,
0:09:54 you’re good at making weather balloons that go up and stay up for a long time and that you can
0:10:00 steer. I also read that you launched thinking that you could collect data for a tenth the cost
0:10:08 of existing alternatives, which sounds compelling, but as I understand was not compelling enough.
0:10:14 Is that right? – It is. We’re now at 150 times more data per dollar than alternatives. – Okay.
0:10:21 Was there some reason one tenth the cost was not cheap enough? – Yeah. When we started the company,
0:10:27 we thought it would be cheap enough. But it turns out when you really look at the scale of weather
0:10:34 data collection you need to really understand the atmosphere, you want 10,000 balloons aloft
0:10:42 concurrently. That’s a balloon every 60 miles in the atmosphere. And in order to get to that level
0:10:52 of scale, it’s the difference between your company spending $100 million a year and spending a billion
0:10:59 dollars a year. One was a lot more in reach. – How do you drive down the cost to one 150th
0:11:05 the cost? I mean, I’m sure it’s many, many, many incremental efficiency gains, but what’s
0:11:11 an example of one? – Yeah. Well, one of the big pieces is improving the software that flies on it
0:11:19 so that the balloon can fly for longer. I talked about in the student group that first flight
0:11:25 lasting 76 hours, now our longest flight can fly for over 40 days. – Wow. – And the longer you fly,
0:11:34 well, the hardware cost stays the same, so that means you’re collecting a lot more data. – Yeah,
0:11:40 yeah. So basically make the balloon stay up for longer is the fundamental way that you make it
0:11:44 cheaper. – Exactly. – And how do you make it fly longer? – Yeah. Well, one of the big things is
0:11:52 improving the software to better decide when do you vent gas, when do you drop ballast,
0:11:57 or really one of the other big things is just making everything smaller and lighter,
0:12:03 because the smaller it is, the lighter it is, the less chance of a leak, the less ballast you have
0:12:10 to use, and things have just shrunk down so far. It really surprises me sometimes when I’m like,
0:12:19 “Wait, that tiny thing, the size of a dime, replaces the thing that was the size of a dinner
0:12:26 plate before?” – Yeah, that’s amazing. So okay, so let’s talk about where you are today. Let’s
0:12:32 talk about sort of how it works. How big are the balloons? – So the balloon is two pieces. There’s
0:12:39 the envelope, that’s the bag that holds all the gas, the balloon part of the balloon, and then
0:12:47 there’s the main unit, which is the electronics and the ballast. – The stuff. – Exactly. – The balloon
0:12:55 and the stuff. How big is the balloon part of the balloon? – The balloon part of the balloon
0:12:59 is five and a half meters tall, so that’s two and a half, three times the height of person,
0:13:06 depending on how tall you are. – Yeah, so tall. It’s a big balloon. And how big is the stuff?
0:13:14 How big are the sensors in the part that’s not the balloon? – Yeah, that stuff is, I guess this is
0:13:21 an audio thing, so I would say about this big. – The size of a basketball? – Yeah, it’s roughly
0:13:30 the size of a basketball. It weighs just over four pounds, so about the weight of a large duck.
0:13:37 And it’s kind of long and skinny. It kind of looks like a fish. Think of it as a trout attached to a
0:13:49 giant bag. – How many of your balloons are in the sky right now about? – Yeah, there are a few dozen
0:13:57 aloft right now. We launch around 100 a month and are quickly ramping that up. – So where do you
0:14:03 launch your balloons from? – We launch them every day from three continents. South Korea, Palo Alto,
0:14:09 New York, and Cabo Verde. – Cabo Verde is just off the off the west coast of Africa, is that right?
0:14:14 – You got it. One of the reasons why we have launch site in Cabo Verde right now,
0:14:19 that’s right where a lot of hurricanes are forming. And so the fact that we’re collecting
0:14:24 data around there is going to be really impactful for better predicting the path of these hurricanes.
0:14:31 – So they start there and then they essentially travel across the Atlantic and into the Americas.
0:14:36 And if you can understand what’s going on there, ideally you can sort of understand the hurricane
0:14:42 where it’s going to go in the dream scenario, even just as it’s becoming a tropical storm.
0:14:47 – Exactly. Yeah, NOAA did an analysis of the impact our data had on the 2022 hurricane season.
0:14:55 And it made the forecasts for Hurricane Fiona about 20% better.
0:15:01 – Now it’s time for a few ads. After the ads, Kai and I will talk about AI, of course. We’ll also
0:15:12 talk about Windborne’s business, what they’re selling. And we’ll talk about the company’s
0:15:16 quest to build balloons that can stay in the air for months at a time and make multiple trips around the world.
0:15:23 – When it comes to ensuring your company has top-notch security practices,
0:15:37 things can get complicated fast. Vanta automates compliance for SOC2 ISO 27001 and more,
0:15:44 saving you time and money. With Vanta, you can streamline security reviews by automating questionnaires
0:15:50 and demonstrating your security posture with a customer-facing trust center. Over 7,000 global
0:15:56 companies like Atlassian, FlowHealth and Quora use Vanta to build trust and prove security in real time.
0:16:03 Listeners can claim a special offer of $1,000 off Vanta at Vanta.com/special. That’s V-A-N-T-A.com/special
0:16:12 for $1,000 off Vanta. – Hey, everybody. I’m Kai Rizdal, the host of Marketplace, your daily
0:16:20 download on the economy. Money influences so much of what we do and how we live. That’s why it’s
0:16:26 essential to understand how this economy works. At Marketplace, we break down everything from
0:16:32 inflation and student loans to the future of AI so that you can understand what it all means for you.
0:16:38 Marketplace is your secret weapon for understanding this economy. Listen wherever you get your podcasts.
0:16:43 What is your business right now? What are you selling and who are you selling it to?
0:16:52 Right now, our business has two pieces to it. We are selling the observations of the atmosphere we
0:16:59 collect to the government right now so that those governments can use this data to improve your
0:17:06 weather forecast. Just a couple of weeks ago, we announced our AI-based weather model, which is
0:17:15 the world’s most accurate global weather model, bar none. – I feel like you need a little TM when
0:17:20 you say the world’s most accurate global weather model, bar none. – We really do. – We really do.
0:17:25 So let’s talk about that. Is that claim validated? I know there was news about that just this year,
0:17:30 and congratulations. But also, how do I know that’s true, respectfully? – I love that question
0:17:36 because anytime a weather company makes a claim, you should look at it closely.
0:17:42 First off, we’ve published our results on our website, so you can dig into some of the raw
0:17:50 numbers there. Second, we’re in the process of submitting to weather bench, which is these
0:17:58 benchmarks run by Google, which were the previous holders of this record.
0:18:03 So we’re actively talking with them right now about submitting our results, getting them
0:18:10 validated and live up there. It’s just a process that takes a little while. – So let’s talk about the
0:18:17 model you built. Generally, when I talk to people who are working on AI, they talk a lot
0:18:23 about data, right? And it seems like to some significant degree, the models are kind of
0:18:29 commoditized, not quite, but like models are pretty similar, it seems in many settings at
0:18:34 least, one to the other. And the differentiator often ends up being the data. Is that the case
0:18:40 in this instance? I mean, do you think you’re the best if you’re the best because you have all this
0:18:44 data from all your balloons? – I think that your spot on data is the real moat here because
0:18:52 what really sets us apart is the ability to have all of this data that no one else has
0:18:58 and use that both for training the model, but then also for running the model in real time.
0:19:05 Data isn’t just about training, it’s about essentially giving your model the prompt of
0:19:10 what’s going to happen next. And you need new data every single day for that.
0:19:17 – Just one thing to clarify. So if you’re selling the data to governments, are they not
0:19:25 sharing it widely, like they’re paying for the data, but not everybody can get at the data?
0:19:30 Is that why it’s a moat for you? – Yeah, so what they’re using the data we sell them for is
0:19:37 putting it into conventional physics-based weather models that they then release. So
0:19:45 they can’t redistribute the data itself, but the general public can benefit from its use in
0:19:53 these weather models. – And so just to distinguish, and this is a distinction that’s not particular
0:19:58 to your company, but you mentioned the traditional physics-based weather models. I mean, I think
0:20:04 it’s worth spending just a moment here to distinguish between the classic weather model,
0:20:12 the sort of pre-AI weather model, and the AI weather model. What’s the basic difference between
0:20:18 those two? – Yeah, so a conventional physics-based weather model, it takes the initial state of
0:20:23 the atmosphere and then runs a bunch of fluid dynamics on it to simulate what’s going to happen
0:20:32 to all of these different fluids. – This is a sort of kind of classic, sort of feels like kind of
0:20:37 19th century deterministic, give me the initial conditions and I’ll tell you what’s going to
0:20:43 happen in the future. It’s that, right? It’s sort of rules-based. – Exactly, exactly. And that is
0:20:53 nice because you can see the exact physics that is being used, but you need compute clusters that
0:21:00 cost hundreds of millions of dollars in order to run this because the atmosphere is so big.
0:21:07 And by contrast, a AI-based weather model can run on just a gaming laptop. And what it does,
0:21:18 by contrast, is effectively picking out statistics. – I mean, it’s machine learning,
0:21:24 I presume. It’s pattern matching. – Exactly. Our model is a transformer-based architecture,
0:21:30 the same thing that powers chat GPT. – Are AI models in general better at this point than the
0:21:36 physics-based traditional models? – It depends on the use case, but if you’re talking about
0:21:41 hurricanes, yeah. We ran a case study on Hurricane Ian and our model would have predicted landfall
0:21:52 by 200 kilometers closer. – I mean, that’s another one where I’m intrigued by what you said, but I’m
0:21:59 very eager for… I’m not eager for this hurricane season to come, but it will resonate more with
0:22:05 me when you do that prospectively, right? – Of course. – I mean, will you this year be predicting
0:22:12 where hurricanes are going to make landfall? – We will be predicting that for public safety reasons.
0:22:18 We aren’t going to be… – You won’t tell me. – Yeah. – You will, but you won’t tell me. – We
0:22:23 will tell me NOAA hurricane capture. – Fair. – You, of course, don’t want to say, “Hey,
0:22:30 don’t listen to NOAA about these evacuation orders.” – No, no. Fair. That is very responsible.
0:22:37 So basically, you’re selling data, you are likely soon to be selling forecasts. That’s
0:22:44 sort of the business. On the technical side, what is the frontier? What are you trying to figure
0:22:50 out that you haven’t figured out yet? – Yeah. Well, one of the big areas for innovation is
0:22:56 on the AI modeling side. We were kind of surprised that our models did as well as they did, because
0:23:05 there are a lot of, quite frankly, obvious things that we haven’t done. Things like just
0:23:12 increasing the amount of compute we’re using to train these models. We use something like a 15th
0:23:19 as much as Google did. So what happens when you train it for longer? – I mean, I’ll say, on the AI
0:23:26 side, it’s not obvious that it’s optimal for you to be doing the AI as well. There’s a universe
0:23:33 where you are optimizing the balloons and the data, and then someone else, like Google, who has
0:23:39 all the computers and all the AI brains that knows what to do with doing the AI side of it.
0:23:46 Yes and no. I think that where that falls apart is coupling our data with the models much more
0:23:55 closely, in that there’s so much to be done in terms of figuring out how to better take advantage
0:24:02 of our data in particular, and also things like saying, based on this weather model, where should
0:24:10 we be flying our balloons to improve the forecast? And we’re already using our own AI weather forecasts
0:24:17 to do that flight plan optimization to figure out where our balloons are going to fly. So it’s
0:24:23 really this beneficial effect where the better our weather forecast, the better we can fly our
0:24:28 balloons, and we can then target our balloons to fly to the places that will most improve the
0:24:35 weather forecast. That’s a good feedback loop. There’s also a lot to do on the balloon side of
0:24:40 things. I wish I could tell you about our project that will increase flight time by another factor
0:24:49 of 10. Another factor of 10? Yeah. So what are you at now? What’s the like median flight time?
0:24:57 Medium flight time is seven to ten days, depending on how we’re targeting it. And so you think you’re
0:25:04 going to get to 90 days? Yep, we do. Around the world in 80 days? Yeah. You say you wish you
0:25:12 can tell me like you feel like you’re about to do it. Yeah. Tell me without telling me what you
0:25:17 can’t tell me. Okay, I’ll do my best. So we have had some very successful flight tests where we have
0:25:28 demonstrated the ability to fly without using any finite resources. So what is the key finite
0:25:38 resource, the lift gas? What is it? Helium or hydrogen? What do you use to keep the balloons up?
0:25:44 Our balloons are compatible with both helium and hydrogen, so which we use is dependent on the
0:25:50 country. When you say flying without any finite resource, that’s what I think of. Yep. Okay.
0:25:56 Yeah, that and ballast. Oh, right. So you want it? Should I guess? There’s heat. There’s the sun.
0:26:07 I’ll leave it there and just say… If you get that, could you fly forever? Is that the dream?
0:26:14 Someday, someday. We’re definitely excited about it. I guess 90 days for a weather balloon is sort
0:26:20 of forever, right? It really is. And so yeah, we’ve had some very successful flight tests.
0:26:26 What’s the longest flight you’ve had to this point? 40 days. 40 days. Okay. Long. Yeah, long.
0:26:33 And that was without using this technology. And was that like a fluke? It was definitely
0:26:39 on the side of the bell curve, but we’ve had multiple month-long flights.
0:26:45 What… Why do your balloons usually, what do you say, fall? Crash sounds a little extreme. Crash?
0:26:52 Fall? Yeah, it’s that they run out of ballast and lift and gas. And so these finite resources
0:27:00 have been used up. You don’t have any more ballast to drop. You don’t have any more gas to vent.
0:27:06 And so you can no longer control your altitude. What happens when it falls, crashes, comes down?
0:27:14 So by the time it falls, it has used up all of the ballast. So instead of weighing four pounds,
0:27:23 it weighs about 200 grams. 200 grams is what, half a pound-ish? Yes. I’d looked it up 0.44 pounds.
0:27:37 Okay. By the way, when you’re getting rid of ballast, is that just like sand or something?
0:27:41 You’re just like sprinkling sand out? Exactly. Okay. Yeah. Dust in the wind.
0:27:47 So it’s half a pound. So it wouldn’t hurt if it hit me on the head. Is that part of the reason
0:27:52 half a pound is important? Would it hurt if it hit me on the head? Has it hit anybody on the head?
0:27:56 It’s never hit anybody. We do landing simulations and control where it lands to direct it towards
0:28:04 unpopulated areas. But remember, it also has this envelope attached to it, which acts like a parachute.
0:28:13 So it actually falls very slowly. It’s very light and not an issue. And then what happens? Then you
0:28:20 have this big deflated balloon sitting on the ground or floating in the ocean. And is it just
0:28:25 what happens to it? Yeah. So I love that question because it’s a chance to talk about how our balloons
0:28:34 can reduce the amount of waste going out into the world compared to the half a million that are
0:28:39 launched every year of which only a fifth are recovered. Just answer directly. Okay. I appreciate
0:28:45 that you want to contextualize it. Okay. But first, let’s talk about what happens with your balloons
0:28:50 and then feel free to provide the bigger content. Sounds good. Sounds good. Yeah. Yeah. So our balloons
0:28:56 are out in the world and we recover as many of them as we can. In the long term, we’re aiming to
0:29:04 recover essentially all of them because we’ll direct them to specific landing sites. I mean,
0:29:10 the ocean has got to be tough. I presume when they fall in the ocean for the most part, you
0:29:15 don’t recover them. Is that right? Exactly. And that’s one of the reasons why improving endurance
0:29:20 is so exciting because if you circumnavigate multiple times, well, on the last leg, you bring
0:29:26 it down in a field, have somebody drive around and collect all of them. There are about half a
0:29:31 million weather balloons launched each year and only a fifth of those are recovered. So when each
0:29:39 balloon flies for only two hours, well, you have to launch a lot more balloons to collect the same
0:29:46 amount of data. Yeah. So we really see this as an opportunity to reduce the amount of stuff that’s
0:29:51 going out there in the first place. Right. So there will actually be fewer balloons
0:29:57 going into the world if you succeed. Exactly. So you’re at this point where you’ve learned a
0:30:03 lot. It feels like you’re kind of on the precipice of a lot more. And I want to talk about sort of
0:30:09 two futures. One is a future where your company doesn’t succeed for any number of reasons with
0:30:16 which you’re surely more familiar than I. And then the other is if you do succeed, right? So
0:30:21 in the sort of sad version where you don’t succeed, what are some reasons it might not work out?
0:30:25 Yeah. I think that probably the biggest reason is that scaling up hardware is hard. We need to
0:30:34 increase manufacturing by a factor of 100 effectively. And so it’s scaling up that final
0:30:41 assembly by a factor of 100 is complicated and hard and just kind of classic, hard building a
0:30:47 business, gets more capital intensive. Exactly. You have to spend the money before you can get the
0:30:52 money. Exactly. Okay. So that’s a easy to imagine sad outcome. Let’s talk about the happy outcome.
0:31:01 Like it works. You scale up your balloons, stay up for months at a time. What’s the world look
0:31:09 like? What are you doing in that scenario? We want everybody to see twice as far into the future
0:31:15 when it comes to weather. So making the 10-day forecast as accurate as the current 5-day forecast,
0:31:21 making the 20-day forecast as accurate as the 10-day forecast. So we want people to see twice
0:31:27 as far into the future. We want to pinpoint where hurricanes are going to make landfall
0:31:33 a week in advance. And we want day-to-day weather that businesses rely on to be really accurate.
0:31:43 Things like never having to cancel a flight last minute because a bomb cyclone popped up.
0:31:52 Saving fuel with all of your shipping because you know where there will be headwinds.
0:32:00 We want to accelerate the transition to renewables. And we want weather to go from this
0:32:06 crazy, unpredictable source of uncertainty to something that humans just know about.
0:32:13 We’ll be back in a minute with The Lightning Round.
0:32:28 Hey everybody, I’m Kai Rizdal, the host of Marketplace, your daily download on the economy.
0:32:33 Money influences so much of what we do and how we live. That’s why it’s essential to understand
0:32:39 how this economy works. At Marketplace, we break down everything from inflation and student loans
0:32:45 to the future of AI so that you can understand what it all means for you. Marketplace is your
0:32:50 secret weapon for understanding this economy. Listen wherever you get your podcasts.
0:32:57 Okay, let’s do The Lightning Round.
0:33:00 Would you rather have it be too hot or too cold?
0:33:06 Too cold any day.
0:33:08 Okay, the next one is multiple choice. What is your favorite movie prominently featuring balloons?
0:33:17 Wizard of Oz, Around the World in 80 Days, Up, The Red Balloon, or None of the Above?
0:33:23 It’s Gotta Be Up, brilliant movie. Blimps, Overrated or Underrated?
0:33:30 So here I’ve got to make a distinction between blimps and zeppelins because I actually used
0:33:37 to work at a zeppelin company. Is that the Larry Page zeppelin company?
0:33:43 It sure is. Perhaps the only zeppelin company.
0:33:46 And so is the distinction that a zeppelin has a rigid frame?
0:33:53 Got it in one.
0:33:53 Okay, so how about this? Let me reformulate it to see if I get it this time.
0:33:59 Airships, Overrated or Underrated?
0:34:02 Underrated. I think that zeppelins are so cool. I don’t have a great mission-driven answer here
0:34:13 other than, you know, I think it’s amazing. I read too many books growing up which prominently
0:34:20 featured airships. Isn’t the Larry Page airship company dream some kind of cargo?
0:34:27 Like the idea that for like really heavy things, giant airships would be an efficient way to
0:34:32 move cargo? Full disclosure, it’s been a long time since I worked there, so I can’t speak to that
0:34:38 company in particular. But yeah, faster deliveries than a ship, more efficient deliveries than a plane.
0:34:47 Aha, it’s like a niche, an unfilled niche. What’s something besides the
0:34:53 weather that AI will be really good at predicting?
0:34:56 Oh, great question.
0:34:58 Let’s see. I don’t have a good answer off the top of my head. I think that it will be
0:35:10 really interesting for economics in terms of taking short-term data, various real-time indicators,
0:35:21 and making predictions about what the final readings are going to be.
0:35:26 I’m sure a lot of very smart people are working on that, and they may in fact already have good
0:35:32 models and they’re not telling us. I think that’s likely.
0:35:36 Is there anything else you want to say?
0:35:38 I think that covers it. Yeah, great to be on the show.
0:35:41 Thanks for your time. It was lovely to talk with you. Good luck.
0:35:45 Thanks. And if you ever are in Palo Alto, you’re welcome to come see a balloon launch.
0:35:50 Great. So do you still, you launch from, where do you launch from that air force base or like,
0:35:58 or just from? From our parking lot.
0:36:00 The yard, are they easy? Yeah, from the parking lot, that’s cool.
0:36:03 Yeah.
0:36:03 How much space do you need to launch a weather balloon?
0:36:07 Not that much, about 50 by 50 feet. It’s really just a matter of making sure there’s
0:36:13 nothing that the balloon will blow into right after releasing it.
0:36:16 Seems like it’ll be fun to see. Yeah, I like a balloon.
0:36:21 Kai Marshland is the co-founder and chief product officer of Windborne Systems.
0:36:29 Today’s show was produced by Gabriel Hunter Chang and edited by Lydia Jean Cotte.
0:36:35 It was engineered by Sarah Brugier. You can email us at problem@pushkin.fm.
0:36:41 I’m Jacob Goldstein, and we’ll be back next week with another episode of What’s Your Problem.
0:36:51 Hey, everybody. I’m Kai Rizdal, the host of Marketplace, your daily download on the economy.
0:37:02 Money influences so much of what we do and how we live. That’s why it’s essential to understand
0:37:07 how this economy works. At Marketplace, we break down everything from inflation and student loans
0:37:13 to the future of AI so that you can understand what it all means for you.
0:37:17 Marketplace is your secret weapon for understanding this economy. Listen wherever you get your podcasts.
0:00:06 Pushkin.
0:00:06 [MUSIC]
0:00:11 >> Hey everybody, I’m Kai Rizdal, the host of Marketplace,
0:00:13 your daily download on the economy.
0:00:16 Money influences so much of what we do and how we live.
0:00:19 That’s why it’s essential to understand how this economy works.
0:00:23 At Marketplace, we break down everything from inflation and
0:00:26 student loans to the future of AI so that you can understand what it all means for you.
0:00:32 Marketplace is your secret weapon for
0:00:34 understanding this economy, listen, wherever you get your podcasts.
0:00:37 [MUSIC]
0:00:41 Every day around the world, more than 1,000 weather balloons are launched into the sky.
0:00:46 The balloons float high up into the atmosphere, sending back information.
0:00:50 You know, temperature, wind speed, air pressure, etc.
0:00:54 And then, a couple hours later, the balloons pop.
0:00:58 This is basically the way weather balloons have worked for decades.
0:01:02 And the information the balloons send back is really useful for weather forecasts.
0:01:06 But the information the balloons send back is also pretty limited.
0:01:10 Because the balloons only stay in the sky for a couple hours, they don’t fly very far.
0:01:15 If we could figure out how to make balloons stay up for longer,
0:01:19 they could blow in the wind and travel thousands of miles.
0:01:22 They could travel across continents and across oceans and send us back a lot more
0:01:27 data and give us a much clearer picture of what weather is coming our way.
0:01:31 [MUSIC]
0:01:37 I’m Jacob Goldstein, and this is What’s Your Problem,
0:01:39 the show where I talk to people who are trying to make technological progress.
0:01:43 My guest today is Kai Marshland.
0:01:46 He’s the co-founder and chief product officer of Windborne Systems.
0:01:50 Kai’s problem is this.
0:01:52 Can you build weather balloons that stay in the air for weeks or months instead of hours?
0:01:57 And can you pair the data from those balloons with AI to make weather forecasts
0:02:02 that are way better than anything we have today?
0:02:05 Kai got interested in balloons back in 2015.
0:02:08 He was a freshman at Stanford, and he joined the Student Space Initiative,
0:02:12 which sounds kind of fancy, but in fact,
0:02:15 was basically a bunch of college kids trying to build their own weather balloons.
0:02:19 [MUSIC]
0:02:23 The first half dozen flights of our balloons were complete into total failures.
0:02:29 When do things start working?
0:02:31 That would be back in June of 2016.
0:02:35 This is the end of my freshman year, and I think our longest flight that year had been 14 minutes.
0:02:43 So shorter than a normal weather balloon.
0:02:48 – Buy a lot. – Buy a lot.
0:02:50 And also, the person who had started this project, who then became my co-founder,
0:02:59 had just graduated and was about to leave what we thought never to be seen again.
0:03:05 And school has ended.
0:03:07 We can’t even get into the building because our key cards have stopped working,
0:03:12 so we have to tailgate someone in to get our balloon launch equipment out,
0:03:17 go out to the middle of nowhere in the central valley of California,
0:03:22 and if this launch didn’t succeed, Windborn would not exist.
0:03:28 – Okay. – So I can guess what happened next based on that fact,
0:03:33 but tell me what happened.
0:03:34 So it got to be about 4 a.m.
0:03:38 And we’re out in the middle of this park, really cold, and the sprinklers turn on.
0:03:45 They start slowly swiveling around towards our delicate electronics.
0:03:52 And so I have to hurl myself onto the sprinklers, getting soaked, just to save our electronics.
0:03:58 – It’s like the nerd version of throwing yourself on a grenade. – Exactly.
0:04:02 – I really felt like I was in an action movie sequence. – Yeah, good.
0:04:06 But that sacrifice really paid off because with that flight,
0:04:12 we managed to set the world record for weather balloon endurance.
0:04:17 – How long did it stay up? – It stayed up for 76 hours.
0:04:21 – Wow. And your previous best was 14 minutes? – Exactly.
0:04:25 So you’re in college. How do you build weather balloons that are better than whatever
0:04:29 the federal government, NOAA, is launching our state-of-the-art weather balloon people?
0:04:34 So the first big thing is that some new consumer electronics have come out.
0:04:41 In particular, things like microcontrollers and lightweight low-power satellite communications.
0:04:50 And we realized that we can use these innovations to make a balloon smarter and control its altitude.
0:04:59 – What’s a microcontroller? – A microcontroller is a super small computer
0:05:05 that you can put in any device to give it a brain.
0:05:10 – And they’re cheap. – They’re dirt cheap.
0:05:13 Anytime you hear the word “internet” of things, that means microcontrollers.
0:05:17 – And nobody else was doing this at the time?
0:05:19 Like all the big, well-funded weather agencies weren’t sticking microcontrollers on their weather balloons?
0:05:27 – Well, with these big agencies, they are really incentivized to keep things stable.
0:05:35 Because weather is this really important thing that billions of people depend on.
0:05:40 And so they’re not going to say, “Hey, what crazy experiments can we run?”
0:05:46 – Their incentive is just like, “Don’t screw it up.” – Exactly.
0:05:48 – “With the way we did it yesterday, don’t screw it up and you won’t get fired.” – Exactly.
0:05:52 – And so specifically, what was it that you were doing that no one had done before?
0:05:58 – Yeah, so it’s a very simple concept that is really hard to make work in practice.
0:06:05 The way we control our altitude is when it gets too high up, we vent some of our gas.
0:06:12 So it has less lift. When we fall down too far, we drop some ballast, so it stops falling.
0:06:19 But the hard part about this is it has to function at -70 degrees Celsius.
0:06:25 That’s the temperature of dry ice. You have to have a lot of onboard software.
0:06:31 That’s where the microcontrollers come in to do things like understand,
0:06:36 is this just turbulence that you’re hitting? Or is it actually a real change in lift that we
0:06:41 need to drop ballast or vent gas to account for? And in fact, we stayed up 24/7 every single time
0:06:49 it talked to us every five minutes. We had it play an air horn sound because we were,
0:06:55 of course, falling asleep in the middle of the night to see what it was doing.
0:07:00 – So, okay, so you set the record, you stay up for, what, three days watching it? How far did it go?
0:07:06 – It flew all the way across the country and landed out in the Atlantic Ocean.
0:07:13 But the real reason why we decided to form a company around this was because we realized
0:07:21 the impact this could have. We got into this from the engineering side. The lands are fun.
0:07:27 But we realized, wait a second, 85% of the Earth’s atmosphere is invisible to humanity.
0:07:34 Weather is this crazy, unpredictable thing, and we can solve that. And weather is so much more
0:07:42 than just do you bring an umbrella to work tomorrow. It is the single most immediately
0:07:49 destructive aspect of climate change. And improving the weather forecast, it, of course,
0:07:55 helps with adapting to climate change, things like predicting where a hurricane will make
0:08:00 landfall. But it can also help with preventing climate change in the first place. If you have
0:08:06 a better weather forecast, you can better route ships and planes to save a huge amount of fuel.
0:08:13 You can accelerate the transition to renewables because you know when the wind will be blowing
0:08:20 or the sun will be shining. And we looked at this and said, no one else has our balloon technology.
0:08:26 If we don’t do this, no one will. And if we walk away from this opportunity, we can’t live with
0:08:32 ourselves. – I mean, I buy that the stakes are high. I buy that it could be very helpful.
0:08:39 But like, why wouldn’t anybody else do it? – Yeah, a lot of people are trying to improve
0:08:47 the weather forecast. But what’s unique about Windborne is the combination of our balloon
0:08:55 technology with our AI weather forecasts. Because really, those are the two levers
0:09:01 to pull to increase the accuracy of a weather forecast. – You mentioned that 85% of the world
0:09:08 lacks good weather data, which was surprising to me the first time I heard it. But I would
0:09:14 think like, can’t satellites do it? That’s kind of, I guess, my naïve-ish thought. It’s like,
0:09:20 I know there’s a lot of satellites looking down at the Earth all the time. – I’m actually a big
0:09:24 fan of the space industry. I used to work at SpaceX. But the problem is the laws of physics
0:09:29 fundamentally limit what satellites can measure. Take, for example, pressure. It turns out pressure
0:09:35 is extremely important for predicting the weather because it determines where the winds are blowing,
0:09:42 how the weather systems are moving. But a satellite fundamentally cannot measure pressure from space
0:09:49 because it’s not in the atmosphere. It can’t see what’s going on. – So you start the company,
0:09:54 you’re good at making weather balloons that go up and stay up for a long time and that you can
0:10:00 steer. I also read that you launched thinking that you could collect data for a tenth the cost
0:10:08 of existing alternatives, which sounds compelling, but as I understand was not compelling enough.
0:10:14 Is that right? – It is. We’re now at 150 times more data per dollar than alternatives. – Okay.
0:10:21 Was there some reason one tenth the cost was not cheap enough? – Yeah. When we started the company,
0:10:27 we thought it would be cheap enough. But it turns out when you really look at the scale of weather
0:10:34 data collection you need to really understand the atmosphere, you want 10,000 balloons aloft
0:10:42 concurrently. That’s a balloon every 60 miles in the atmosphere. And in order to get to that level
0:10:52 of scale, it’s the difference between your company spending $100 million a year and spending a billion
0:10:59 dollars a year. One was a lot more in reach. – How do you drive down the cost to one 150th
0:11:05 the cost? I mean, I’m sure it’s many, many, many incremental efficiency gains, but what’s
0:11:11 an example of one? – Yeah. Well, one of the big pieces is improving the software that flies on it
0:11:19 so that the balloon can fly for longer. I talked about in the student group that first flight
0:11:25 lasting 76 hours, now our longest flight can fly for over 40 days. – Wow. – And the longer you fly,
0:11:34 well, the hardware cost stays the same, so that means you’re collecting a lot more data. – Yeah,
0:11:40 yeah. So basically make the balloon stay up for longer is the fundamental way that you make it
0:11:44 cheaper. – Exactly. – And how do you make it fly longer? – Yeah. Well, one of the big things is
0:11:52 improving the software to better decide when do you vent gas, when do you drop ballast,
0:11:57 or really one of the other big things is just making everything smaller and lighter,
0:12:03 because the smaller it is, the lighter it is, the less chance of a leak, the less ballast you have
0:12:10 to use, and things have just shrunk down so far. It really surprises me sometimes when I’m like,
0:12:19 “Wait, that tiny thing, the size of a dime, replaces the thing that was the size of a dinner
0:12:26 plate before?” – Yeah, that’s amazing. So okay, so let’s talk about where you are today. Let’s
0:12:32 talk about sort of how it works. How big are the balloons? – So the balloon is two pieces. There’s
0:12:39 the envelope, that’s the bag that holds all the gas, the balloon part of the balloon, and then
0:12:47 there’s the main unit, which is the electronics and the ballast. – The stuff. – Exactly. – The balloon
0:12:55 and the stuff. How big is the balloon part of the balloon? – The balloon part of the balloon
0:12:59 is five and a half meters tall, so that’s two and a half, three times the height of person,
0:13:06 depending on how tall you are. – Yeah, so tall. It’s a big balloon. And how big is the stuff?
0:13:14 How big are the sensors in the part that’s not the balloon? – Yeah, that stuff is, I guess this is
0:13:21 an audio thing, so I would say about this big. – The size of a basketball? – Yeah, it’s roughly
0:13:30 the size of a basketball. It weighs just over four pounds, so about the weight of a large duck.
0:13:37 And it’s kind of long and skinny. It kind of looks like a fish. Think of it as a trout attached to a
0:13:49 giant bag. – How many of your balloons are in the sky right now about? – Yeah, there are a few dozen
0:13:57 aloft right now. We launch around 100 a month and are quickly ramping that up. – So where do you
0:14:03 launch your balloons from? – We launch them every day from three continents. South Korea, Palo Alto,
0:14:09 New York, and Cabo Verde. – Cabo Verde is just off the off the west coast of Africa, is that right?
0:14:14 – You got it. One of the reasons why we have launch site in Cabo Verde right now,
0:14:19 that’s right where a lot of hurricanes are forming. And so the fact that we’re collecting
0:14:24 data around there is going to be really impactful for better predicting the path of these hurricanes.
0:14:31 – So they start there and then they essentially travel across the Atlantic and into the Americas.
0:14:36 And if you can understand what’s going on there, ideally you can sort of understand the hurricane
0:14:42 where it’s going to go in the dream scenario, even just as it’s becoming a tropical storm.
0:14:47 – Exactly. Yeah, NOAA did an analysis of the impact our data had on the 2022 hurricane season.
0:14:55 And it made the forecasts for Hurricane Fiona about 20% better.
0:15:01 – Now it’s time for a few ads. After the ads, Kai and I will talk about AI, of course. We’ll also
0:15:12 talk about Windborne’s business, what they’re selling. And we’ll talk about the company’s
0:15:16 quest to build balloons that can stay in the air for months at a time and make multiple trips around the world.
0:15:23 – When it comes to ensuring your company has top-notch security practices,
0:15:37 things can get complicated fast. Vanta automates compliance for SOC2 ISO 27001 and more,
0:15:44 saving you time and money. With Vanta, you can streamline security reviews by automating questionnaires
0:15:50 and demonstrating your security posture with a customer-facing trust center. Over 7,000 global
0:15:56 companies like Atlassian, FlowHealth and Quora use Vanta to build trust and prove security in real time.
0:16:03 Listeners can claim a special offer of $1,000 off Vanta at Vanta.com/special. That’s V-A-N-T-A.com/special
0:16:12 for $1,000 off Vanta. – Hey, everybody. I’m Kai Rizdal, the host of Marketplace, your daily
0:16:20 download on the economy. Money influences so much of what we do and how we live. That’s why it’s
0:16:26 essential to understand how this economy works. At Marketplace, we break down everything from
0:16:32 inflation and student loans to the future of AI so that you can understand what it all means for you.
0:16:38 Marketplace is your secret weapon for understanding this economy. Listen wherever you get your podcasts.
0:16:43 What is your business right now? What are you selling and who are you selling it to?
0:16:52 Right now, our business has two pieces to it. We are selling the observations of the atmosphere we
0:16:59 collect to the government right now so that those governments can use this data to improve your
0:17:06 weather forecast. Just a couple of weeks ago, we announced our AI-based weather model, which is
0:17:15 the world’s most accurate global weather model, bar none. – I feel like you need a little TM when
0:17:20 you say the world’s most accurate global weather model, bar none. – We really do. – We really do.
0:17:25 So let’s talk about that. Is that claim validated? I know there was news about that just this year,
0:17:30 and congratulations. But also, how do I know that’s true, respectfully? – I love that question
0:17:36 because anytime a weather company makes a claim, you should look at it closely.
0:17:42 First off, we’ve published our results on our website, so you can dig into some of the raw
0:17:50 numbers there. Second, we’re in the process of submitting to weather bench, which is these
0:17:58 benchmarks run by Google, which were the previous holders of this record.
0:18:03 So we’re actively talking with them right now about submitting our results, getting them
0:18:10 validated and live up there. It’s just a process that takes a little while. – So let’s talk about the
0:18:17 model you built. Generally, when I talk to people who are working on AI, they talk a lot
0:18:23 about data, right? And it seems like to some significant degree, the models are kind of
0:18:29 commoditized, not quite, but like models are pretty similar, it seems in many settings at
0:18:34 least, one to the other. And the differentiator often ends up being the data. Is that the case
0:18:40 in this instance? I mean, do you think you’re the best if you’re the best because you have all this
0:18:44 data from all your balloons? – I think that your spot on data is the real moat here because
0:18:52 what really sets us apart is the ability to have all of this data that no one else has
0:18:58 and use that both for training the model, but then also for running the model in real time.
0:19:05 Data isn’t just about training, it’s about essentially giving your model the prompt of
0:19:10 what’s going to happen next. And you need new data every single day for that.
0:19:17 – Just one thing to clarify. So if you’re selling the data to governments, are they not
0:19:25 sharing it widely, like they’re paying for the data, but not everybody can get at the data?
0:19:30 Is that why it’s a moat for you? – Yeah, so what they’re using the data we sell them for is
0:19:37 putting it into conventional physics-based weather models that they then release. So
0:19:45 they can’t redistribute the data itself, but the general public can benefit from its use in
0:19:53 these weather models. – And so just to distinguish, and this is a distinction that’s not particular
0:19:58 to your company, but you mentioned the traditional physics-based weather models. I mean, I think
0:20:04 it’s worth spending just a moment here to distinguish between the classic weather model,
0:20:12 the sort of pre-AI weather model, and the AI weather model. What’s the basic difference between
0:20:18 those two? – Yeah, so a conventional physics-based weather model, it takes the initial state of
0:20:23 the atmosphere and then runs a bunch of fluid dynamics on it to simulate what’s going to happen
0:20:32 to all of these different fluids. – This is a sort of kind of classic, sort of feels like kind of
0:20:37 19th century deterministic, give me the initial conditions and I’ll tell you what’s going to
0:20:43 happen in the future. It’s that, right? It’s sort of rules-based. – Exactly, exactly. And that is
0:20:53 nice because you can see the exact physics that is being used, but you need compute clusters that
0:21:00 cost hundreds of millions of dollars in order to run this because the atmosphere is so big.
0:21:07 And by contrast, a AI-based weather model can run on just a gaming laptop. And what it does,
0:21:18 by contrast, is effectively picking out statistics. – I mean, it’s machine learning,
0:21:24 I presume. It’s pattern matching. – Exactly. Our model is a transformer-based architecture,
0:21:30 the same thing that powers chat GPT. – Are AI models in general better at this point than the
0:21:36 physics-based traditional models? – It depends on the use case, but if you’re talking about
0:21:41 hurricanes, yeah. We ran a case study on Hurricane Ian and our model would have predicted landfall
0:21:52 by 200 kilometers closer. – I mean, that’s another one where I’m intrigued by what you said, but I’m
0:21:59 very eager for… I’m not eager for this hurricane season to come, but it will resonate more with
0:22:05 me when you do that prospectively, right? – Of course. – I mean, will you this year be predicting
0:22:12 where hurricanes are going to make landfall? – We will be predicting that for public safety reasons.
0:22:18 We aren’t going to be… – You won’t tell me. – Yeah. – You will, but you won’t tell me. – We
0:22:23 will tell me NOAA hurricane capture. – Fair. – You, of course, don’t want to say, “Hey,
0:22:30 don’t listen to NOAA about these evacuation orders.” – No, no. Fair. That is very responsible.
0:22:37 So basically, you’re selling data, you are likely soon to be selling forecasts. That’s
0:22:44 sort of the business. On the technical side, what is the frontier? What are you trying to figure
0:22:50 out that you haven’t figured out yet? – Yeah. Well, one of the big areas for innovation is
0:22:56 on the AI modeling side. We were kind of surprised that our models did as well as they did, because
0:23:05 there are a lot of, quite frankly, obvious things that we haven’t done. Things like just
0:23:12 increasing the amount of compute we’re using to train these models. We use something like a 15th
0:23:19 as much as Google did. So what happens when you train it for longer? – I mean, I’ll say, on the AI
0:23:26 side, it’s not obvious that it’s optimal for you to be doing the AI as well. There’s a universe
0:23:33 where you are optimizing the balloons and the data, and then someone else, like Google, who has
0:23:39 all the computers and all the AI brains that knows what to do with doing the AI side of it.
0:23:46 Yes and no. I think that where that falls apart is coupling our data with the models much more
0:23:55 closely, in that there’s so much to be done in terms of figuring out how to better take advantage
0:24:02 of our data in particular, and also things like saying, based on this weather model, where should
0:24:10 we be flying our balloons to improve the forecast? And we’re already using our own AI weather forecasts
0:24:17 to do that flight plan optimization to figure out where our balloons are going to fly. So it’s
0:24:23 really this beneficial effect where the better our weather forecast, the better we can fly our
0:24:28 balloons, and we can then target our balloons to fly to the places that will most improve the
0:24:35 weather forecast. That’s a good feedback loop. There’s also a lot to do on the balloon side of
0:24:40 things. I wish I could tell you about our project that will increase flight time by another factor
0:24:49 of 10. Another factor of 10? Yeah. So what are you at now? What’s the like median flight time?
0:24:57 Medium flight time is seven to ten days, depending on how we’re targeting it. And so you think you’re
0:25:04 going to get to 90 days? Yep, we do. Around the world in 80 days? Yeah. You say you wish you
0:25:12 can tell me like you feel like you’re about to do it. Yeah. Tell me without telling me what you
0:25:17 can’t tell me. Okay, I’ll do my best. So we have had some very successful flight tests where we have
0:25:28 demonstrated the ability to fly without using any finite resources. So what is the key finite
0:25:38 resource, the lift gas? What is it? Helium or hydrogen? What do you use to keep the balloons up?
0:25:44 Our balloons are compatible with both helium and hydrogen, so which we use is dependent on the
0:25:50 country. When you say flying without any finite resource, that’s what I think of. Yep. Okay.
0:25:56 Yeah, that and ballast. Oh, right. So you want it? Should I guess? There’s heat. There’s the sun.
0:26:07 I’ll leave it there and just say… If you get that, could you fly forever? Is that the dream?
0:26:14 Someday, someday. We’re definitely excited about it. I guess 90 days for a weather balloon is sort
0:26:20 of forever, right? It really is. And so yeah, we’ve had some very successful flight tests.
0:26:26 What’s the longest flight you’ve had to this point? 40 days. 40 days. Okay. Long. Yeah, long.
0:26:33 And that was without using this technology. And was that like a fluke? It was definitely
0:26:39 on the side of the bell curve, but we’ve had multiple month-long flights.
0:26:45 What… Why do your balloons usually, what do you say, fall? Crash sounds a little extreme. Crash?
0:26:52 Fall? Yeah, it’s that they run out of ballast and lift and gas. And so these finite resources
0:27:00 have been used up. You don’t have any more ballast to drop. You don’t have any more gas to vent.
0:27:06 And so you can no longer control your altitude. What happens when it falls, crashes, comes down?
0:27:14 So by the time it falls, it has used up all of the ballast. So instead of weighing four pounds,
0:27:23 it weighs about 200 grams. 200 grams is what, half a pound-ish? Yes. I’d looked it up 0.44 pounds.
0:27:37 Okay. By the way, when you’re getting rid of ballast, is that just like sand or something?
0:27:41 You’re just like sprinkling sand out? Exactly. Okay. Yeah. Dust in the wind.
0:27:47 So it’s half a pound. So it wouldn’t hurt if it hit me on the head. Is that part of the reason
0:27:52 half a pound is important? Would it hurt if it hit me on the head? Has it hit anybody on the head?
0:27:56 It’s never hit anybody. We do landing simulations and control where it lands to direct it towards
0:28:04 unpopulated areas. But remember, it also has this envelope attached to it, which acts like a parachute.
0:28:13 So it actually falls very slowly. It’s very light and not an issue. And then what happens? Then you
0:28:20 have this big deflated balloon sitting on the ground or floating in the ocean. And is it just
0:28:25 what happens to it? Yeah. So I love that question because it’s a chance to talk about how our balloons
0:28:34 can reduce the amount of waste going out into the world compared to the half a million that are
0:28:39 launched every year of which only a fifth are recovered. Just answer directly. Okay. I appreciate
0:28:45 that you want to contextualize it. Okay. But first, let’s talk about what happens with your balloons
0:28:50 and then feel free to provide the bigger content. Sounds good. Sounds good. Yeah. Yeah. So our balloons
0:28:56 are out in the world and we recover as many of them as we can. In the long term, we’re aiming to
0:29:04 recover essentially all of them because we’ll direct them to specific landing sites. I mean,
0:29:10 the ocean has got to be tough. I presume when they fall in the ocean for the most part, you
0:29:15 don’t recover them. Is that right? Exactly. And that’s one of the reasons why improving endurance
0:29:20 is so exciting because if you circumnavigate multiple times, well, on the last leg, you bring
0:29:26 it down in a field, have somebody drive around and collect all of them. There are about half a
0:29:31 million weather balloons launched each year and only a fifth of those are recovered. So when each
0:29:39 balloon flies for only two hours, well, you have to launch a lot more balloons to collect the same
0:29:46 amount of data. Yeah. So we really see this as an opportunity to reduce the amount of stuff that’s
0:29:51 going out there in the first place. Right. So there will actually be fewer balloons
0:29:57 going into the world if you succeed. Exactly. So you’re at this point where you’ve learned a
0:30:03 lot. It feels like you’re kind of on the precipice of a lot more. And I want to talk about sort of
0:30:09 two futures. One is a future where your company doesn’t succeed for any number of reasons with
0:30:16 which you’re surely more familiar than I. And then the other is if you do succeed, right? So
0:30:21 in the sort of sad version where you don’t succeed, what are some reasons it might not work out?
0:30:25 Yeah. I think that probably the biggest reason is that scaling up hardware is hard. We need to
0:30:34 increase manufacturing by a factor of 100 effectively. And so it’s scaling up that final
0:30:41 assembly by a factor of 100 is complicated and hard and just kind of classic, hard building a
0:30:47 business, gets more capital intensive. Exactly. You have to spend the money before you can get the
0:30:52 money. Exactly. Okay. So that’s a easy to imagine sad outcome. Let’s talk about the happy outcome.
0:31:01 Like it works. You scale up your balloons, stay up for months at a time. What’s the world look
0:31:09 like? What are you doing in that scenario? We want everybody to see twice as far into the future
0:31:15 when it comes to weather. So making the 10-day forecast as accurate as the current 5-day forecast,
0:31:21 making the 20-day forecast as accurate as the 10-day forecast. So we want people to see twice
0:31:27 as far into the future. We want to pinpoint where hurricanes are going to make landfall
0:31:33 a week in advance. And we want day-to-day weather that businesses rely on to be really accurate.
0:31:43 Things like never having to cancel a flight last minute because a bomb cyclone popped up.
0:31:52 Saving fuel with all of your shipping because you know where there will be headwinds.
0:32:00 We want to accelerate the transition to renewables. And we want weather to go from this
0:32:06 crazy, unpredictable source of uncertainty to something that humans just know about.
0:32:13 We’ll be back in a minute with The Lightning Round.
0:32:28 Hey everybody, I’m Kai Rizdal, the host of Marketplace, your daily download on the economy.
0:32:33 Money influences so much of what we do and how we live. That’s why it’s essential to understand
0:32:39 how this economy works. At Marketplace, we break down everything from inflation and student loans
0:32:45 to the future of AI so that you can understand what it all means for you. Marketplace is your
0:32:50 secret weapon for understanding this economy. Listen wherever you get your podcasts.
0:32:57 Okay, let’s do The Lightning Round.
0:33:00 Would you rather have it be too hot or too cold?
0:33:06 Too cold any day.
0:33:08 Okay, the next one is multiple choice. What is your favorite movie prominently featuring balloons?
0:33:17 Wizard of Oz, Around the World in 80 Days, Up, The Red Balloon, or None of the Above?
0:33:23 It’s Gotta Be Up, brilliant movie. Blimps, Overrated or Underrated?
0:33:30 So here I’ve got to make a distinction between blimps and zeppelins because I actually used
0:33:37 to work at a zeppelin company. Is that the Larry Page zeppelin company?
0:33:43 It sure is. Perhaps the only zeppelin company.
0:33:46 And so is the distinction that a zeppelin has a rigid frame?
0:33:53 Got it in one.
0:33:53 Okay, so how about this? Let me reformulate it to see if I get it this time.
0:33:59 Airships, Overrated or Underrated?
0:34:02 Underrated. I think that zeppelins are so cool. I don’t have a great mission-driven answer here
0:34:13 other than, you know, I think it’s amazing. I read too many books growing up which prominently
0:34:20 featured airships. Isn’t the Larry Page airship company dream some kind of cargo?
0:34:27 Like the idea that for like really heavy things, giant airships would be an efficient way to
0:34:32 move cargo? Full disclosure, it’s been a long time since I worked there, so I can’t speak to that
0:34:38 company in particular. But yeah, faster deliveries than a ship, more efficient deliveries than a plane.
0:34:47 Aha, it’s like a niche, an unfilled niche. What’s something besides the
0:34:53 weather that AI will be really good at predicting?
0:34:56 Oh, great question.
0:34:58 Let’s see. I don’t have a good answer off the top of my head. I think that it will be
0:35:10 really interesting for economics in terms of taking short-term data, various real-time indicators,
0:35:21 and making predictions about what the final readings are going to be.
0:35:26 I’m sure a lot of very smart people are working on that, and they may in fact already have good
0:35:32 models and they’re not telling us. I think that’s likely.
0:35:36 Is there anything else you want to say?
0:35:38 I think that covers it. Yeah, great to be on the show.
0:35:41 Thanks for your time. It was lovely to talk with you. Good luck.
0:35:45 Thanks. And if you ever are in Palo Alto, you’re welcome to come see a balloon launch.
0:35:50 Great. So do you still, you launch from, where do you launch from that air force base or like,
0:35:58 or just from? From our parking lot.
0:36:00 The yard, are they easy? Yeah, from the parking lot, that’s cool.
0:36:03 Yeah.
0:36:03 How much space do you need to launch a weather balloon?
0:36:07 Not that much, about 50 by 50 feet. It’s really just a matter of making sure there’s
0:36:13 nothing that the balloon will blow into right after releasing it.
0:36:16 Seems like it’ll be fun to see. Yeah, I like a balloon.
0:36:21 Kai Marshland is the co-founder and chief product officer of Windborne Systems.
0:36:29 Today’s show was produced by Gabriel Hunter Chang and edited by Lydia Jean Cotte.
0:36:35 It was engineered by Sarah Brugier. You can email us at problem@pushkin.fm.
0:36:41 I’m Jacob Goldstein, and we’ll be back next week with another episode of What’s Your Problem.
0:36:51 Hey, everybody. I’m Kai Rizdal, the host of Marketplace, your daily download on the economy.
0:37:02 Money influences so much of what we do and how we live. That’s why it’s essential to understand
0:37:07 how this economy works. At Marketplace, we break down everything from inflation and student loans
0:37:13 to the future of AI so that you can understand what it all means for you.
0:37:17 Marketplace is your secret weapon for understanding this economy. Listen wherever you get your podcasts.
Kai Marshland is the co-founder and chief product officer at WindBorne Systems. Kai’s problem is this: How do you build weather balloons that can stay in the air for months at a time, and pair the data gathered by the balloons with AI to make weather forecasts that are way better than anything we have today?
See omnystudio.com/listener for privacy information.