This Startup Fused Human Brain Cells with Silicon Chips | E2295

This Week In Startups is made possible by: Deel https://deel.com/twist Quo https://quo.com/TWiST LinkedIn Jobs https://LinkedIn.com/twist Today's show: Cortical Labs is the world's first company selling biological computers. Their CL1 fuses lab-grown human neurons (derived from stem cells, not actual folks) with silicon hardware to create Synthetic Biological Intelligence (SBI). Founder Dr. Hon Weng Chong walks us through how the system works and why neurons are more efficient than GPUs at reinforcement learning. (Also… is this computer alive?) PLUS Pyka co-founder and CEO Michael Norcia explains the various uses for his autonomous aircraft, from crop-spraying drones in Brazil to a a hybrid-electric defense UAV for the military. Guests: Cortical Labs: ****https://corticallabs.com/ Dr. Hon Weng Chong on X: https://x.com/dr1337 Pyka: https://www.flypyka.com/ Pyka on Instagram: https://www.instagram.com/flypyka/?hl=en Further Reading: 2022 Pong paper in Neuron: https://www.cell.com/neuron/fulltext/S0896-6273(22)00806-6 2017 Paper: "Attention is All You Need"; https://arxiv.org/abs/1706.03762 The "Barista Test" for Artificial Intelligence: Chris Rourk: https://medium.com/predict/the-turing-test-is-so-last-century-the-barista-test-for-artificial-general-intelligence-faf91034fa8c Notable Links: Playing "DOOM" on CL1: https://www.youtube.com/watch?v=yRV8fSw6HaE DayOne Data Center: https://dayonedc.com/ NeurIPS 2026 Conference: https://neurips.cc/ Neuralink: https://neuralink.com/ CliniCloud Digital Stethoscope and Thermometer: https://www.design-industry.com.au/clinicloud Air Force Research Laboratory (AFWERX): https://afwerx.com/ Joby Aviation: https://www.jobyaviation.com/ Prime Movers Lab: https://www.primemoverslab.com/ Timestamps: 0:00 What is "biological computing"? 2:49 Cortical's new $30 million raise 4:15 The world's first biological data center 9:48 Deel - Founders scale faster on Deel. Set up payroll for any country in minutes, hire anyone anywhere, get visas handled fast, and get back to building. Visit https://deel.com/twist to learn more. 10:51 Biological computers have a learning advantage 19:43 Quo (formerly OpenPhone) - Quo gives you a clean, modern way to handle every customer call, text, and thread all in one place. Try it free at https://quo.com/TWiST 29:15 LinkedIn Jobs - Hire right, the first time. Post your first job and get $100 off towards your job post at https://LinkedIn.com/twist 38:46 From paper airplanes to Group 4 UAVs 52:20 Introducing the DropShip defense drone 58:28 How regulations block US drones 1:00:40 Why Pyka builds everything in-house Subscribe to the TWiST500 newsletter: https://ticker.thisweekinstartups.com Check out the TWIST500: https://www.twist500.com Subscribe to This Week in Startups on Apple: https://rb.gy/v19fcp Follow Lon: X: https://x.com/lons Follow Alex: X: https://x.com/alex LinkedIn: ⁠https://www.linkedin.com/in/alexwilhelm Follow Jason: X: https://twitter.com/Jason LinkedIn: https://www.linkedin.com/in/jasoncalacanis Check out all our partner offers: https://partners.launch.co/ Great TWIST interviews: Will Guidara, Eoghan McCabe, Steve Huffman, Brian Chesky, Bob Moesta, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland Check out Jason's suite of newsletters: https://substack.com/@calacanis Follow TWiST: Twitter: https://twitter.com/TWiStartups YouTube: https://www.youtube.com/thisweekin Instagram: https://www.instagram.com/thisweekinstartups TikTok: https://www.tiktok.com/@thisweekinstartups Substack: https://twistartups.substack.com

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  I'm a little worried about how we're going to tell people we're fusing neurons with computers.

• 0:04Speaker 01

  The world's first biological data center. When they compared it against their reinforced learning systems, the neurons we had were 5,000 times more sample efficient.

• 0:13Speaker 00

  Has anyone complained that you're tinkering a bit with the edges of humanity?

• 0:17Speaker 01

  The Vatican were worried about this. You do not want to create conscious systems because ethically, a conscious system has the ability to suffer. And we do not want any suffering to come about from any technology.

• 0:28Speaker 00

  This week in startups is brought to you by LinkedIn. Post your job for free at LinkedIn.com/Twist, then promote it to get access to LinkedIn Jobs' new AI assistant. Quo, formerly OpenPhone, gives you a clean, modern way to handle every customer call, text, and thread all in one place. Try it free and get 20% off your first six months at quo.com/Twist. And Deal. Founders scale faster on Deal. Set up payroll for any country in minutes, hire anyone, anywhere, get visas handled fast, and get back to building. Visit deal.com/Twist to learn more.

• 1:05Speaker 00

  Hello and welcome back to Twist. Now, today, we're talking to a company we spoke to in 2025 because I thought they were one of the most interesting startups in the entire world. Called Cortical Labs, they are trying to fuse silicon, chips that we all know and love, with human neurons, bringing the biologic and the synthetic together to create a new type of computer, a biological computer. I was so tickled by the idea, I had to talk to them. But since that first conversation, Cortical Labs has built out data centers of its biological computers, which means that we now have real, robust capacity to kind of bring humans and computers together. So to tell us more about what's been going on over in the realm of Cortical Labs, please welcome back to the show, it's my dear friend Han. Han, how you doing?

• 1:46Speaker 01

  Good, thank you. It's great to chat again, Alex. And congratulations, I heard you had a new baby.

• 1:51Speaker 00

  Yes, that's why I've been extra tired these last four months, but we're powering through with just the grace of coffee. All right, so Han, last time we talked, you had just put out your CL1, which was the first kind of like fully-contained biological computer with neurons and chips, and you were selling them, I think- Think for something like thirty five thousand dollars apiece. So before we get deep into the tech, just for the business folks out there, how has that product performed in market?

• 2:18Speaker 01

  We've kind of exhausted our entire stock of thirty units that we had capped. So that's good. And you can work out how much that that ended up becoming.

• 2:27Speaker 00

  Roughly a million.

• 2:28Speaker 01

  Yeah, and we have actually, so I'm in the US right now, partly because we we're fundraising, but at the same time, I've also, you know, CEO is essentially chief everything officer, right? I'm also now the company courier. I just dropped off a unit at Johns Hopkins with some of the folks there. I just came from Boston, so Mass General got a unit. Another one just got dropped off at UCSF. So now there is about five US institutions with a CL one device, the other one being Dartmouth. They got an early preview early last year.

• 2:59Speaker 00

  So tell folks what the CL one is, and maybe because you have one with you, show us a little bit for folks on the video version, what it looks like.

• 3:07Speaker 01

  Yeah, absolutely. So maybe before I show it, like to give a very brief summary, the CL one is our attempt to build a computing platform that allows researchers and developers to get going with biological computing very easily. It saves you the effort of building your own hardware, writing your own software, and you just program these things with Python and, you know, they're rack mountable three U unit systems. I have one here to show, so excuse me if the video is a little bit jerky. And this is a CL one, actually, maybe from the front here, you can see what they look like from the sun.

• 3:45Speaker 00

  It looks like a very long space age toaster.

• 3:48Speaker 01

  Yes, so they're actually very similar to rack, rack multiple sleds that some of the GPUs come in the same form factor. So this is three U, so in a server rack cabinet, this will take up about three rows. In a forty five U system, we can pack twenty of these units. And we have about a hundred and twenty, so there are six racks in our lab in Australia that we're calling the world's first biological data center. You know, there's somebody came by and They're like, hey, don't you sell biological computing, uh, computers? And I was like, yeah. And they're like, well, what do you call a place with a lot of computers? I, I don't know, data center. They're like, well, now you've got a biological data center. I was like, huh, that's really nifty.

• 4:31Speaker 01

  But I'm just going to show you what it actually looks like. So we open up the top. This is the neural chamber. So this is where we load up the compute unit. So neurons go into this chip. Um, there's a life support system that is connected to it. There's a heating element underneath here that keeps it at a nice thirty seven degrees Celsius at about a hundred Fahrenheit. We close this latch and think of this as our Neuralink, right? So this is the neural interface. There's a computer at the back. And that all of this stuff here is life support. So we build mechanisms to, to keep the system flowing, to give nutrients to the, the brain to remove the waste. So we have

• 5:12Speaker 01

  pumps here like the heart. We have a feeding and a waste reservoir. So kind of think of it like a stomach and a bladder. These are filtration units like the kidneys. We have a gas mixer like the lungs. And also the really interesting thing is you also have your traditional and non-traditional I O units. So we look at the back here. This is USB C, USB A, Ethernet, but you also have gas inputs. So, uh, filtered room air, CO2, nitrogen. And we have a waste, uh, gas outlet here to relieve the pressure. We jokingly call it the fart valve.

• 5:52Speaker 00

  I'm just laughing because we're talking about, you know, rack mounting and how much space it takes up in a server rack. And then you're like, and here are the lungs and here's the way it expels waste. It's such an interesting combination of things that I understand, but never see together. Now, how many neurons can you pack into a C L one? And then in, uh, compute friendly terms, how much power is that that you bring to bear?

• 6:15Speaker 01

  Yeah. So in a C L one, you can go all the way up to a million, two million neurons, if you so wish. People grow organoids on these things. And they have several million neurons. Um, for the cortical cloud and the, the offerings. That we have, we go down to about 200,000 neurons A because that's a number that, you know, we can grow pretty easily and keep them alive quite well. Makes it commercially viable, but also we found that you can actually get some learning and training with that. So.

• 6:45Speaker 00

  So I want to make an analogy here because I think that everyone's very familiar with the idea of parameters in an, in an L L N. You know, like 105 B would be 105 billion parameters. If it's mixture of experts, yes, fewer than 105 billion would be activated, blah, blah, blah. But we kind of get that number. So 200,000 neurons, a million neurons, same number to me. I have no idea if that's a lot, not very many. So. How many neurons do I have upstairs? And how many do you need to actually have the C L one function as a biological computer?

• 7:15Speaker 01

  Yeah, I can't remember exactly the number, but I think it's like 100 billion neurons that you have in your, in your brain. So, you know. We have that and then a ton more like trillions of synapses. So the closest thing that, you know, you can analogize what we have here is maybe a cockroach. Or, or, you know, a fly kind of thing.

• 7:35Speaker 00

  So that actually doesn't, tell me why I'm wrong here, hon, but. Cockroach is not famous for their intelligence, famous for their durability, but I mean, not, not for being super smart, but when you combine that number of neurons with a, a, a, a chip, a silicon chip. What, what's like the, the multiplication factor that we get from bringing those two things together?

• 7:57Speaker 01

  Yeah, actually, so cockroaches are actually pretty smart, along with, with, with bees and flies. You, you, you know, you, you, you're not wrong. In some cases, they're not intelligent in the way we view like human intelligence, like they're not going to solve calculus, right? But what they do solve really well is have you ever tried killing a fly? They're really hard to kill.

• 8:18Speaker 00

  So hard. Yeah. So quick.

• 8:19Speaker 01

  They're so quick and they're so agile. They are, they almost like predict your actions ahead of time. And so, you know, this whole thing, and I think, you know, the industry needs to get the terminology right. We have it called the super intelligence, like, you know, G P T, whatever, 5.5 is super intelligent. Is it generally intelligent?

• 8:38Speaker 00

  No.

• 8:38Speaker 01

  No, right? Because. You know, the what is it? Steve Wozniak has the best like test so AGI. Can you walk into a stranger's kitchen and make yourself a cup of coffee? Everything is different. Everything is new. You'll have to experiment with it, right? You've never seen it before. We don't have that yet. So I would say that biology, even very simple organisms like, you know, a fly has generalized intelligence, something that none of our machines have. So, you know, we're hoping to exploit those properties and even very simple systems. And, you know, maybe we can get a lot of stuff done there without having to go into the realms of, you know, human intelligence and all the baggage that comes with it, like consciousness and so forth.

• 9:17Speaker 00

  So when I, when I think about biological computing, I presume we're talking about a lot of use cases in drug testing and drug discovery, you know, biosimilars, blah, blah, blah, blah, all the stuff that's kind of flesh and blood. But do you foresee a future in which the, probably not the CL one, maybe the CL three, manages to bring silicon and neurons together in a way that creates a computer that is better than today's GPUs and TPUs and so forth at certain types of calculations that we use in the technology world versus the biology world?

• 9:49Jason Calacanis?

  Founder scale faster on Deal. That's the deal. You can grow your company without borders and you can set up payroll for any country in minutes. Hire anyone, anywhere, like a modern startup or large company does. And Deal is gonna get all the visas handled fast. So you can get back to building. There's a great talent war that's going on right now. And you need people with superpowers for your startup to be competitive, to beat your competitors, to get your products to market. But anytime you try to grow your team with overseas hires, oh my Lord, you've got to reinvent the wheel. You gotta navigate a tangled web of international laws, regulations, and you can't get these things wrong, folks. You want to onboard new staffers in other countries? You want to get them set up on your network, nice and secure, IT access, all that good stuff? You want to manage their benefits?

• 10:33Jason Calacanis?

  Trust me, this is all a nightmare unless you partner with Deal. They are the people stack for startups. They're gonna take care of all the onboarding, payroll, HR, IT, benefits. Everything you need, quickly, in one place, done perfectly. So visit deal.com/twist. That's D E

• 10:50Speaker 01

  Yeah, actually, there is one that is already like proving to be uh better than CPUs or GPUs, and that's in reinforcement learning. So um this this is some very novel work. Uh it's still getting written up, so I don't really want to jinx it, but we're hoping to get it published later this year at NURIPS, um uh which you know is is also happening down in my part of the world in in Sydney. Um and uh you know, what we've discovered is uh doing some work uh with a bit of really strong research partner. Uh the question was, can these neurons exhibit uh goal-seeking behavior or pathfinding, right? And the the answer is yes, uh but not only was the answer yes, which you know shocked us.

• 11:33Speaker 01

  They the secondary thing that they discovered was when they competed against their uh reinforcement learning systems, their benchmarks, the GPU-based uh systems. Um the neurons we had were 5,000 times more sample efficient than their GPU-based systems. What that means is that for every step a biological system is doing, it takes 5,000 times, 5,000 steps more to do on a GPU system. The saving grace with GPU reinforcement learning is that you can just accelerate time. So you just runtime 5,000 times faster than um than the real world. The caveat with that is you can't accelerate time if you're a robot. You you're operating at the same speed like everyone else in the physical world.

• 12:16Speaker 00

  So if if biological computing is good at reinforcement learning, which as I think everyone listening to this knows is an enormous part of improving AI models today.

• 12:26Speaker 01

  Yep.

• 12:26Speaker 00

  You could end up pretty far outside of the realm of the biology side of of industry. That's very interesting, but I feel like I I I I taking us down the wrong path. Let's back up and talk about um what you guys have built. So you've built a a biological data center. You have 120 units in Melbourne. And you're gonna do one in Singapore next, which I believe can get even bigger.

• 12:46Speaker 01

  Yes, exactly. So um we we're working with a data center company called Day1. Um you know, they've they've partnered with us because A, they they look like the technology for several reasons. A, because it is. Um, a, an alternative way to do compute that doesn't affect their energy budget. So this way they can say, you know, we have a data center operating at a very uh tight window uh specified by the Singapore government of two hundred megawatts. They're going to provide the same chips like everyone else. But on top of that, they're going to provide our compute, which is not affecting any of the energy budget because they don't have to do any special cooling. And we, we, one unit only uses about thirty watts of energy. So a rack is six hundred.

• 13:28Speaker 00

  Which is basically zero for this conversation.

• 13:31Speaker 01

  Correct. So, you know, they're, they're getting more for not much, you know, in terms of uh costs. So um that is one of the uh, the reasons why they've, they've uh partnered with us. But secondarily as well, and I think this is something that is a little bit mind breaking as well. They've built not just the space that has a thousand CR one units, but next to it, a laboratory for us to grow the cells for the compute onsite.

• 13:57Speaker 00

  Oh, so you don't have to ship in your, your stem cell based. I point out, we're not killing people here to steal their brains. Stem cell based neurons, you can just make them, make them, grow them, no?

• 14:07Speaker 01

  Yeah.

• 14:07Speaker 00

  Uh, raise them, raise them?

• 14:09Speaker 01

  Make them, grow them, same thing. Yeah.

• 14:11Speaker 00

  It's okay. I'm sorry. It's, my usual words don't work as well when we translate them over to here. I'm sure you've already gone through this, but for me, it's still novel.

• 14:17Speaker 01

  Yeah.

• 14:18Speaker 00

  Um, so what does that save in terms of like operating costs for you guys to not have to stick them in a cooler and fly them?

• 14:23Speaker 01

  Um, tremendous amount, but it also means that there is no supply chain constraint. And it, it, it decouples your data center from not just being a place where you just buy from a central vendor. And you wait for it to come in and you, you, you, you know, you deploy your chips to one where you're also manufacturing your chips onsite. So you kind of decentralize the entire model where, you know, every data center is self sufficient. Every data center technically is not reliant on one vendor from one country.

• 14:53Speaker 00

  Yeah. Re- remind me how long the neurons live before they need to be replaced?

• 15:00Speaker 01

  Um, so neurons can actually live a very long time if you keep them uh, you know, you know, well kept, so to speak. Um, the, the thing that does require a Replenishing are the tube sets and the main, the main culprit for that is at these filtration units here. Let me see.

• 15:18Speaker 00

  If you're on the audio version, he's pointing at.

• 15:21Speaker 01

  I'm pointing at these cartridges here. You think of them like kidneys and what happens is that over time, these filtration units clog up with large protein growth factors. And they kind of result in a bit like a kidney failure situation. So yeah, we just swap them out and then we get another four to six months.

• 15:38Speaker 00

  Okay, so, so really then it's, uh, what, what do you feed the neurons? I was about to say sugar water as a joke, but I think I might actually be a little bit wrong.

• 15:45Speaker 01

  No, it actually is pretty much sugar, sugar water.

• 15:47Speaker 00

  Okay, so basically you, you've made a, a biological system that keeps a small number compared to our brains of neurons alive that are connected to these chips that do a lot of cool things.

• 15:58Speaker 01

  Yeah.

• 15:59Speaker 00

  Here's my question though, as time goes on, do you think that as chips get smarter, we're going to need more neurons to interface with them? Or is this amount of biological compute on top of a chip enough? And we'll get gains just from improving the chip component of this. I'm trying to kind of figure out like putting some dots on the chart for where we're going.

• 16:19Speaker 01

  Yeah, so it's always a case of like push and pull, right? So for instance, um, if, if, let's say we refer back to GPUs, right?

• 16:27Speaker 00

  Sure.

• 16:27Speaker 01

  We had way more GPUs than what we knew what to do with them before we got to the L L M's. And the, the breakthrough was the algorithm, like, you know, was, was a breakthrough success. And we were like, oh, we need more, more processing power. And so there's always this like push and pull tension between are the algorithms there yet? Or are the, you know, is it a hardware limitation? Right now, we, we see this as a algorithm's limitation because the bottleneck is how do you, how do you best represent digital information that is all on the internet and so forth with analog systems, right? Which is you and I. So, um, that, that I think is, is still being worked on. You know, you have really smart people in Neuralink and Synchron working on that for the B C I side.

• 17:12Speaker 01

  We're kind of working on it as well, but we have an additional challenge, which we have to write information directly into the neurons. So, um, you know, to, to answer your question, we, we, we don't think we've saturated the system yet. And so we have enough capacity with our current setup, but who knows? Maybe somebody cracks another new algorithm and then we're like, okay, we got to get more cells going on the system.

• 17:31Speaker 00

  I'm just really excited about what you're working on because when we think about. The systems we're using to make artificial intelligence smarter, we're throwing the equivalent of bodies at it. We're throwing more GPUs, the bitter lesson, Jevon's paradox, blah, blah, blah, blah, blah.

• 17:48Speaker 00

  But we're working on systems that are so fundamentally much more power hungry, less efficient and less attuned to the problems to try and solve than our brains, the things we already have with us. And so to me, it just seems very logical that as the chips get better and as we make, get better algorithms to make the neurons function as we need them to, we should be able to have two different intelligence curves working in synchronized fashion. And we should get much faster gains. And this is when religion comes into it. So, uh, as background, Han, I was raised in a very conservative Christian church. And so growing up in the 90s, I heard a lot about stem cells and cloning and a lot of just what I would call fear mongering.

• 18:31Speaker 01

  Yep.

• 18:32Speaker 00

  I'm now a non-religious science fiction nerd. So I'm pretty much your biggest fan. But I'm a little worried about how we're going to tell people that we're fusing neurons with computers. Cause I think you've now taken this from, from proof of concept, when you guys played Pong to early commercial with the CL1, to playing Doom recently, that was cool. To now building out data centers in an international format. So this is coming to market. Now we can talk about this sort of thing. Has anyone, the Pope or similar complained that you're tinkering a bit with the edges of humanity? And, and, and is anyone worried?

• 19:09Speaker 01

  Actually, the Vatican were worried about this, but fortunately, but my, my CSO has done an excellent job, you know, engaging with bioethicists and actually, you know, being at the forefront of this, right? And so, um, I, I think, you know, because of the space that we're working in and the fact that, you know, there's a lot of ethics that need to go into it, even just doing research work with, you know, any biomedical aspect to it requires an ethics board. We're very attuned to these kinds of potential criticisms, right? So we try to engage them proactively.

• 19:44Jason Calacanis?

  If you're growing a startup or a small business, you can't sleep on incoming calls, even if they come in after hours. Or on the weekends, that's some of your best customers calling you when they need you. That's why today's episode is brought to you by QUO. Q U O. The smarter way to run your business communications. QUO is the number one top rated business phone system on G2. And it's trusted by more than 90,000 businesses. They're going to bring all of your calls, texts and contacts together in a shared collaborative space. That means your entire team is going to use one shared number, if that's what you choose. And there's no more missed calls and no more disconnected conversations.

• 20:24Jason Calacanis?

  I've been using this product for well over five years across three different business units. It's perfect, it's affordable, it's feature rich. And the team is so sharp, they keep releasing new features that I didn't even know I needed. Money is on the line. Always say hello with QUO. Try QUO for free. Plus get 20% off your first six months when you go to quo.com/twist. That's Q U O dot com slash twist.

• 20:49Speaker 01

  And I think, you know, firstly, the most important thing that we're trying to work on right now is to get everybody in this space, plus the consciousness space to, you know, come together and actually agree on a shared nomenclature, right? Because if you cannot agree on the thing that we're all studying on, that we have no chance of actually making any progress. And all we get is just a lot of fear, but not much understanding. So we're trying to do that. And also, yeah, I think there was a, actually Brett was in discussion with the Vatican, who wanted to find out, you know, what's going on here? And, you know, were there any religious and ethical like issues? And, you know, fortunately, they actually agreed that what we're doing was all right. And it was actually, you know, fine. Yeah. Because, because ultimately, the main question that we all, we need to have is, you know,

• 21:33Speaker 01

  is there a harmonization part of the plan? And also, you know, the principle of double doctrine, is there actually a net good that can come from this technology versus a net negative? And so that's the reason why, you know, you started out with the traditional drug discovery, disease modeling and stuff, because that's still a very important part of the work. That is the primary use case for all these labs. Purchasing these devices, which is, you know, so, you know, in mass general, it's a ultimate dementia researcher. At Hopkins, they're looking at toxicology and

• 22:05Speaker 00

  Yeah.

• 22:05Speaker 01

  Turned into animal testing. At UCSF, they're looking at, um, I think they're looking at movement disorders and a whole bunch of other like really.

• 22:12Speaker 00

  Oh, that was nice.

• 22:14Speaker 01

  Yeah.

• 22:14Speaker 00

  These are all, by the way, this is why this technology, even in its current form is fricking awesome. He just listed three major medical centers in the US that are using this technology now to make our lives better. So like this is the dangers that are hypothetical down the road that we're touching on are not meant to undercut the current usefulness and commercial application and research application of this product. I'm just having fun. Keep going on.

• 22:37Speaker 01

  Yeah, thank you. And so, you know, that, that, that is, it's all happening right now. The compute side of it is by all means the smallest, but the newest. The one that's also most forward risk because you never know what people are going to build. So this is something that we're trying to like keep a hold of. And I think internally at the company, this, it's really important to understand this whole, this discussion of consciousness, because I think we've drawn a red line for us, which you do not cross. The problem is where is this line and we got to figure out where it is, is consciousness. You do not want to create conscious systems because ethically, a conscious system has the ability to suffer and we do not want any suffering to come about from any technology. So that is the stance that we take. We try to do as much as we can when we have the cloud, we can monitor things.

• 23:19Speaker 01

  But, you know, when these things go out of the, you know, into the, into the real world, we don't really know what's going on. But, you know, fortunately, most of our users, actually ninety percent of all purchases have all come from academic R and D institutions that, you know, we, we trust that they, they do the right thing.

• 23:35Speaker 00

  But the, the cloud you guys have built, the data center in Melbourne and or Melbourne or whatever, uh, and then Singapore. Um, you know, we're going to have, if it's two hundred thousand neurons a piece and we're going to have, I think it's up to a thousand in Singapore, you can do the math, it's quite a lot. Um, the reason why it doesn't scare me is it's all kind of like cut up into little pieces, like each one has X number of neurons. But there is the science fiction, you know, question of if you have a data center full of these and they can interact, does that change the math? But the good news is that I think we're pretty far away from having so many biological computers in the world all plugged in together that we have to think about that. So getting back to practical applications here, why go the cloud route as opposed to just selling the devices? Because clearly, if you sold out your first run, plenty of demand.

• 24:20Speaker 00

  Um, why, yeah, so cloud? I'm just curious about why that was the right commercial approach to the market.

• 24:25Speaker 01

  Ultimately, what we really want to do at Chronicle Labs is to, you know, I hate the word democratize, but it is democratize the technology by, by reducing the accessibility barrier to biological computing, right? Because, you know, ultimately, um, you know, I, I always go back to Nvidia, right? Because ultimately, the AI that we have today is actually an accident. It was completely a fluke that it happened, right? Because, you know, everyone's like, yeah, Jensen was so smart. He's so brilliant about this. I was like, yeah, if he was so brilliant, why did it take seven years for when CUDA was made to when Alex Krizowski made AlexNet, right? It was a sheer, sheer serendipitous moment where somebody was Jeff Lintons grad student who was solving for image recognition technology, who had a GPU and you had to program CUDA. So.

• 25:10Speaker 00

  I like to think we would have had some other serendipitous meeting point along the way, but that is, you know, how it happened. It does feel a little tenuous, a little fragile. And you look back, yeah.

• 25:20Speaker 01

  Exactly, right? And so, you know, what, what Jensen did, you know, I think brilliantly at the start was he made CUDA free. He made it so that any GPU, even the, the crappiest gaming GPU could also run CUDA. And so the accessibility barrier was really low, so anyone could get into it. So that's what we're trying to do here with, with the cloud system because, you know, unfortunately, if you ever see a one, you really can't do anything with this unless you have a lab and you can grow cells and you, um, you know, can, uh, keep them alive.

• 25:49Speaker 00

  Right. So you need to have a person who's doing the kidney replacement and the feeding and the waste extraction and so forth.

• 25:54Speaker 01

  Exactly, exactly. But what if you don't have any of that, but you have a great idea you want to experiment with? So the idea of the cloud was born where we said, let's just give it to people with ideas so they can muck around with it. And that's actually the story of Doom. We actually didn't build Doom. Um, the Doom was actually a student developer, um, who, um,

• 26:17Speaker 01

  Participated at a hackathon at Stanford, no biology background, used our APN SDK, and he built something that was really cool. And so we decided to, you know, help him tell the story and word got around. And I think he's now been accepted into a very prestigious incubator program. I shall not name which.

• 26:36Speaker 00

  Could it be called Kai Bombinator or rhyme with that?

• 26:42Speaker 01

  Can either confirm or deny.

• 26:44Speaker 00

  Got it, got it. So, so last time we talked, we were talking about the first application of your technology to a video game, which was Pong. And for the young people, Pong is a game when you have two paddles on either side of the screen, they go up and down and knock a ball back and forth. Yes, we used to think that was fun. It still is, frankly. But you told me that when you, you built the system, you had to kind of like create a reward or punishment mechanism, I forget which, to give it the incentives to learn how to play correctly, right?

• 27:14Speaker 01

  Correct, yep.

• 27:16Speaker 00

  In Pong, not a lot of variables. In Doom, many more. So, do you know how they managed to set up the reward or punishment mechanisms to actually interact with a game of that complexity? Because I'm curious if it was hard or not.

• 27:30Speaker 01

  Yeah, so it pretty much is also the same. They're using audit stimulus versus disaudit stimulus as the reward and punishment signal. There were a few more variables and, you know, honestly, it's kind of funny because I think they didn't, they didn't disincentivize or punish it for wasting ammo because there was no ammo.

• 27:52Speaker 00

  Restrictions.

• 27:53Speaker 01

  Restrictions. And so essentially what it did was it figured out that if it just spam the, the shoot button and just spin around the circle, it would just win. So we, it's kind of funny because that's what happens in, in actual reinforcement systems as well. So when we changed it, you know, it actually was, I guess, punished for wasting ammo and all that stuff. And I started to learn and actually started to have some interesting gameplay from that.

• 28:18Speaker 00

  I just realized why it's very important that you don't create conscious systems because they can feel pain. Because if you're using disordered inputs as a, as a. Punishment mechanism is a little bit harsh, everybody. We're working on the terms, but you wouldn't want to do that to a conscious system.

• 28:33Speaker 01

  Correct, exactly.

• 28:34Speaker 00

  Yeah, okay, that makes that I've connected that and that makes a lot of sense. But here's the takeaway. I mean, okay, look for Doom's great. If you haven't played Doom three, do it. You owe yourself to have that fun. But if a kid, and I say that with love, can build this with your API. To me, it shows that the the technology is not that hard to bring to real world application. Hence the cloud. And that brings me to my last question, which is, how many people can you? Customers can you serve with 120 CL ones in a data center? One, 10,000? I literally have no idea what the range is.

• 29:12Jason Calacanis?

  Adding a new member to your team is a crucial decision. You don't want to rush into a hiring situation that you will regret. But if you're a busy founder, you also don't want to spend a ton of time in the trenches looking for that perfect candidate. Instead, you need a partner and you need a trusted partner. And LinkedIn Hiring Pro is that partner. How do I know this? Because I use LinkedIn Hiring Pro. You want a real world testimony? All right, here you go. We recently hired a new customer success manager. They handle all the advertising accounts here on This Week in Startups. And they ensure that we're keeping all of our wonderful partners happy. LinkedIn helped us connect with an amazing candidate right here in Austin. And he had exactly the combination of experience we were looking for.

• 29:55Jason Calacanis?

  It's magic, it's alchemy, it's everything for me to find a great team member. So hire right the first time. Get started by posting your job for free at LinkedIn.com/twist. Terms and conditions apply. That's LinkedIn.com/twist.

• 30:10Speaker 01

  So we have 120 that we've deployed. Unfortunately, you know, there has been some teething with the biology. So while we do have 120, only about 20 or 30 users are on the cloud at the moment. Actually, 20 users. We have another 10 more that we have to bring online. But that's also because, you know, there's a whole conveyor belt like system of getting sales growing and all that stuff. And it's one of those things where unfortunately, because it takes some time. For it to grow, you what you what you what we see now is the result of decisions made two months ago kind of thing. So they're only starting to come online. We've had this flow sort of start, but yeah, a lot of uh so no, a lot of people have started to come on board.

• 30:50Speaker 01

  And the waiting queue is getting smaller. Um, so yeah, we we, you know, we have that. We actually have a couple of um uh. Corporates and and partners coming in board to do some experimental work.

• 31:01Speaker 00

  Yeah.

• 31:02Speaker 01

  Um, but yeah, you're all, yeah.

• 31:04Speaker 00

  But so when when uh when Doom is running on your biological computers, is that one CL one? Is that five? Like, how much does it take?

• 31:13Speaker 01

  One, it was just one. So we haven't come up with the with the chaining of the systems yet. That's coming in, I think.

• 31:18Speaker 00

  Okay, so each one's still a discreet box. They're okay, that then my question was actually relatively silly. Then you could serve one hundred and twenty customers at once.

• 31:24Speaker 01

  Yeah, exactly. I mean, that's all right. I don't know. That's fine, because you know, this is what we we are, what we're doing at the moment. There has been some internal research work about, you know, where we we're doing uh PDMS microfluidic wells, so we can actually segment out the the surface of the chip and. You know, rather you get all sixty, maybe you only need sixteen. And that way, I don't know if you're familiar with like VMware, maybe you can get a hybrid. Like a virtual machine with multiple people running the same thing. So that way we can increase yield and stuff.

• 31:54Speaker 00

  P uh uh PDMS is a poly dimethyl siloxane?

• 31:59Speaker 01

  Yep, that's.

• 32:00Speaker 00

  Which is also known as a dimethylcon. What what are we talking about?

• 32:05Speaker 01

  It's a material. It's like a gel type of material that that you can make microfluidic devices with. Um, so. Think of it like a like a well. Actually, I'll show you what it looks like. So.

• 32:17Speaker 00

  Okay, because and then define microfluidics for me, because again, I I nod my head like I understand, but we are outside of my comfort zone. This is not SAS economics, so. You're going to have to help me along.

• 32:28Speaker 01

  Okay, so I'm gonna show you this. So this is a microfluidic device. So there are probably several thousand neurons in one well. And they are now segmented into these little sections. And we actually have channels that we can turn on and turn off. So we can actually control this more like a circuit now.

• 32:49Speaker 00

  What we're looking at here, if you didn't know what it was, you would think these were top-down views of water storage tanks with pumps and pipes coming into them. But instead, each one of these little

• 33:01Speaker 01

  Octagons.

• 33:01Speaker 00

  Each of these octagons has a bunch of neurons in it, and the microfluidics allows you to feed and then also transfer information between them?

• 33:07Speaker 01

  Yeah, the microfluidics are these channels. And what they do is they constrain the growth of the long part of the neurons. So they can only communicate with their nearest neighbors. Or we can block them off and say, you can only communicate with your east-west neighbor or your north-south mid. So you have a lot more like circuitry, like control over that.

• 33:28Speaker 00

  Right. And then you could, in theory, segregate to have more than one person using the same computer because you can essentially not let the axons get too busy sharing information. Got it.

• 33:38Speaker 01

  Correct. So this is kind of like a triplet model that we've been experimenting with. So we can say, all right, you can have this one here. You'll have 1, 2, 3, 4 electrodes to work from. And if your workload is pretty low, maybe this is just good enough. And you can actually have, if you want, get 2. And then you can have them sort of network up like this. And then we can share the instance and stuff like that. So there's a lot of things that we're working on internally to try to figure out, hey, can we increase yield? Can we get more interesting, like, you know, learning some of that kind of stuff?

• 34:12Speaker 00

  So how much are you raising? What's your target?

• 34:14Speaker 01

  Yeah, we're raising, oh, and actually, one more thing. We're raising 30, but I also wanted to show you guys this. So this is the cortical cloud.

• 34:22Speaker 00

  Ah.

• 34:23Speaker 01

  Yeah.

• 34:24Speaker 00

  If you're on the audio version, it looks like a standard developer backend, which is a compliment.

• 34:29Speaker 01

  Yes, and then you just hit the instance that you've acquired. And there you go. I'm in New York right now. My lab is in Melbourne. And I am live streaming neural activity from a cell culture that my team have assigned me for demo purposes. And

• 34:44Speaker 00

  What is it doing? What are these dots that I'm seeing? If you're on the audio version, imagine you're looking out the window of a spaceship and you're seeing stars go by. That's what it looks like in slow mo.

• 34:52Speaker 01

  Yep. So the window with stars, this is what we call raster plot. So as we go down, you can see the channels are like increasing in number. And if you look on the right side, that's the topological mapping of it. So that's the surface of the chip. And so every time there's a spike or an action potential, we're marking it on this grid here. And we're also marking on this. So you can actually see there's like a temporal correlation. You can see bursting activity. You can see synchrony on this view. So like you can see this cluster here. There's another cluster there. And you know, after a while, you can watch this. Oh, look, they're getting really active here. You can just watch this and sort of like figure out what's going on.

• 35:33Speaker 01

  And then you can also look at the activity view. So here we can actually see the raw waveform. So this is kind of the same thing that a neural link would be picking up. And you know, if you wanted to like, you know, give them a bit of a poke, we can just say, all right, I'm going to poke the neurons in this region here. And I can click and I've delivered a stimulus across and then that wakes them up a bit going on.

• 35:54Speaker 00

  That's really good. It's not conscious. Otherwise that would be kind of rude. Like I was sleeping.

• 35:58Speaker 01

  I know. Yeah. How rude. Now it's a good dream there. We can see like little spike waveform. So we can actually see the, you know, what, what the shape of them are. So you can verify that they're actually really like action potential spikes. And the cool thing here is that if you're a developer, by the way, I'm on team human here. So we still need developers to think of really cool applications. LLMs are not going to solve this for us. You write all the application code in in Python.

• 36:24Speaker 00

  Yeah.

• 36:24Speaker 01

  And yeah, they're all in Jupiter notebooks. So you specify the what program is actually do with CL1 is.

• 36:33Speaker 01

  Think of them as mini architects or mini matrices, right? So you're building the the matrix of the neurons are are going to be put in. You specify the parameters of the game, the the reward and punishments, the objective functions and so forth. And you just let them, you know, get connected to it. And so, yeah, this is how you can do it. Documentations on docs.corticallabs.com. There's an SDK you can pip install and off you go.

• 37:01Speaker 00

  You know, it's one of the best parts about talking to founders that I get to do, Han. It's just straight up getting to see the future. Usually it feels like I'm looking about 12 feet out. This is one of those chats which I feel like I'm looking 12 years out. Like, there's there's enough optimization, expansion, programming languages, how to teach these things. Like, there's so much here that isn't done yet.

• 37:21Speaker 01

  Oh, yeah.

• 37:22Speaker 00

  I feel like you have like, well, you have your life's work ahead of you. This is gonna be, this is gonna be a lot. Okay, I gotta let you come back on in six months and tell us how the cloud's going and how Singapore is going. Because apparently we're just gonna have you on all the time. But for folks who want to learn more, what's the URL? And is there a job you want to shout out to the audience in case the right person is tuned in?

• 37:43Speaker 01

  Yes, so check us out at corticallabs.com. The cloud is at cloud.corticallabs.com, documentation docs.corticallabs.com. I'm on Twitter as well, dr1337. And also, we are looking for talented developers. Stay tuned to our Twitter feed. There may be a hackathon coming out of MIT soon.

• 38:07Speaker 00

  Are you gonna go to that?

• 38:09Speaker 01

  Yeah, probably will have to go for that one.

• 38:11Speaker 00

  So I I only live like a half an hour away from that area. I wonder if I could, maybe I can speak wide. It'd be fun.

• 38:18Speaker 01

  Absolutely, yeah, for sure. Yeah.

• 38:19Speaker 00

  All right.

• 38:20Speaker 01

  So so stay tuned and yeah, more developers. Chailing the internet, Steve Ballmer, developers, developers, developers.

• 38:27Speaker 00

  At least you're not wearing a sweaty collared shirt. Han, thank you so much for coming back to the show. We'll see you again soon.

• 38:33Speaker 01

  Thank you.

• 38:33Speaker 00

  We're talking about drones. No, not drones that sit on the water or go beneath the water or just fly up in the sky carrying a single hand grenade. No, today, we're going to talk about very large drones, mostly outside of the battlefield context. So I rung up my dear friend, Michael Norcia from PICA, to come on and tell us about why drones are good for agriculture, why they're good for near-term cargo transport, and why they may soon be dropping couches from the sky on your head as long as you're within a radius of 50 meters. Please welcome to the show. It's Michael from PICA. How you doing?

• 39:02Speaker 03

  Awesome. Thank you for having me.

• 39:04Speaker 00

  Doing well. So I'm really excited about your company because I thought of you guys as the company that makes large drones that are powered by batteries that sprays crops. But as it turns out, you guys have really broadened out in the last couple of years and do quite a lot more. But Michael, to get to that point, I feel like we should go back to the future. So take me back to the start of the company, the initial vision, and how you guys literally quite got off the ground.

• 39:26Speaker 03

  Yeah, great question. So honestly, the start of the story begins when I was like four years old. I've been just incredibly passionate about aviation my entire life. So I started with paper airplanes when I was about four, rubber band powered airplanes, little electric planes, pretty much obsessively built aircraft throughout my childhood. Ended up studying physics and then I got to work on a number of different EV tall projects in the Bay Area, three or four of them actually. So mostly passenger carrying, vertical takeoff and landing air taxi concepts.

• 39:56Speaker 00

  Yep.

• 39:57Speaker 03

  Super fun, like really hard technical challenges, but it was clear, this was like almost 10 years ago. There was just no way we were going to certify those aircraft for commercial use for another, like realistically, two decades.

• 40:08Speaker 00

  Yeah.

• 40:10Speaker 03

  That took the wind down my sails. So decided to leave to start PICA. And yeah, that was now almost 10 years ago. So it's a long journey.

• 40:18Speaker 00

  When was the first flight? And how far have you guys gone in terms of like commercial penetration of the agricultural sector with your drone that can handle crop springs? I'm curious if you've managed to gain material market share in that industry.

• 40:32Speaker 03

  Yeah. Yeah. So first flight was actually like a week before Demo Day. We went through Y Combinator. So it was a wild Time. We spent one month designing a sort of 600 pound, what would be about a two passenger autonomous aircraft. We built it in two months, actually a little under two months and flew it on like week 11 of the whole program. A week before demo day. So that was awesome. It actually flew itself. It took off and landed running our own software. You know, great testament to the realities of hardware. Like we had a flying prototype autonomous big UAV. 11 weeks in. That was, I think, literally like 1% of the work that we've done now.

• 41:18Speaker 03

  So the other 99% has been going from, you know, prototype to now a like widely. Deployed aircraft that is operating with customers, literally seven hours, sorry, seven days per week. You know, some our customers are doing 12 to 13 hour shifts. So yeah, where we are today, believe it or not, we're actually the only company on earth that I'm aware of. That have deployed a group four UAS to commercial customers at scale. And interestingly, this is happening in the like rural jungles of Brazil. That is the only place on earth that big drones like this are being flown commercially.

• 41:55Speaker 00

  That's weird. Talk to me about group four. I don't think that's a particularly well-known metric outside of the UAS space.

• 42:00Speaker 03

  Yeah. So it's a more like military classification of drones. It's basically drones that are bigger than 1320 pounds. So, you know, things like the MQ nine Reaper, that's the most common group four UAS.

• 42:15Speaker 00

  All right. Now you said you'd only done 1% of the work when you'd gone from no plane. Or no big drone to having a big drone that can fly. A lot of people out there that watch this or listen to this are software founders and they're probably familiar with if you code up a feature. That does not mean product market fit. It doesn't mean the sales cycle is over. There's still a lot of work to be done. Once you have some code. But for folks outside the hardware space, tell them why it was only 1% of the work because from the outside. Going from no plane to plane, sounds like you made a lot of progress.

• 42:42Speaker 03

  Yeah, totally. So, I mean, the thing with. Our products are a blend of hardware and software. And so really like the last nine years have been this just steady. March of maturing both the hardware and software in parallel.

• 42:58Speaker 03

  The hardware, I think, is the particularly difficult one to mature quickly. So that first aircraft, you know, couldn't actually complete an interesting customer mission. It didn't have the features it needed to spray a crop, but couldn't really move cargo even. So there was a lot of like work there. But, you know, I think the most interesting thing really is actually what's been happening in the last three or four years for us with commercial deployments with customers. So, you know, a prototype can be operated by N number of engineers, it can be operated by your entire company if need be. You know, your uptime requirements are non-existent. If you fly for one hour in one week, you'll be like, hell yeah, we got an awesome video. That's all we need.

• 43:43Speaker 00

  Put it on YouTube immediately, we're raising the next round.

• 43:45Speaker 03

  Exactly. With customers, it's completely different. You know, our customers are very, very upset if the aircraft is down for more than 24 hours. And these customers exist in like extremely remote regions of the world. Like they're literally, in some cases, a six hour drive from the nearest city. And so getting to that sort of level of uptime takes an extraordinary amount of time and effort. So it's, you know, it's deploying multiple aircraft with customers, flying them, learning, retrofitting, fixing, re-engineering over and over and over again, until you achieve, you know, it's like the equivalent of product market fit is a tool that a customer can actually rely on.

• 44:30Speaker 00

  Yeah, yeah, product market product reliability fit, maybe we could call it.

• 44:35Speaker 03

  Yeah, exactly.

• 44:36Speaker 00

  Why are we not using more of your ag drones here in the States? You talked about, you know, Latin America, remote regions. I know that agriculture is different across different areas based on what we're growing and seasons and so forth. But, you know, I've laid drip tape on a farm, I've driven large pieces of equipment. And to me, like going above the crops here would make tons of sense. And your system seems to be very robust and it's got a good spray width. And why can't I use this in Nebraska or why aren't?

• 45:05Speaker 03

  Yeah, yeah. So it's coming, it's coming fast. So the, the things that drew us to Latin America were twofold. One is regulatory. So Brazil actually deregulated agricultural drones about two years ago. It was very convenient for us. We do have commercial approval to operate our aircraft here in the United States. We actually have the largest drone approved for commercial use by the FAA, but it has some cumbersome limitations. We can only fly the drone about four kilometers away from where it took off and landed from, which is not commercially viable today.

• 45:36Speaker 00

  No, that's ridiculous. Why?

• 45:37Speaker 03

  Yeah, and it has to do with like line of sight to the vehicle and...

• 45:42Speaker 00

  Oh, ZipLine told me about this, that if you want to fly outside of line of sight, there's another entire set of regulations on that. But it works in Brazil, right?

• 45:50Speaker 03

  It does, yes.

• 45:52Speaker 00

  Is the air different in Brazil compared to the United States? Does something change to navigation and computing when you cross the border?

• 45:58Speaker 03

  Yeah, actually, it's totally... No, no, it's not. Yeah, so in Brazil, we're operating typically anywhere from like five to 15 kilometers from where we took off and landed from.

• 46:07Speaker 00

  Yeah.

• 46:07Speaker 03

  So I mean, the good news is we have data about this. Like, no one has ever done an operation like this before. And so the FAA understandably is sort of apprehensive about what that looks like. We've been working with them over the last year to actually remove that limitation and expand it to more like what we're doing in Brazil, making really good progress on that. So, like I said, we're coming to the US. We do have one customer already here who has an aircraft, but we're going to have far more in the coming year.

• 46:34Speaker 00

  All right. So I think this is a good time to talk about fuel, because it's going to come up when we go over to drop ship in a minute. You currently make mostly electric drones. And I know that when you sell the AG drone, you send with multiple sets of batteries so you can kind of hot swap them and move them in and out. But if I'm six hours from a town, what I probably don't have is as good of access to the grid. So to me, it's lovely that I don't have to cart fuel in a truck up some mountain roads, but I'm curious about just recharging these damn things out in the woods or mountains.

• 47:03Speaker 03

  Yeah. Yeah, so farms have electricity. Like, you know, some farms are pivot irrigated, for example. They have these giant like circular pivots that go around, big water pumps, blah, blah, blah. So the farm will always have electricity. The question is whether they want to run the electricity from the sort of center point to the actual runway. And I would say half of the customers that we have today do that. The other half use a diesel generator. But what's really remarkable, so the kind of competition for our aircraft is a vehicle called the air tractor. It's a big, you know, roughly 8,000 pound vehicle. It burns roughly 55 gallons of jet fuel per hour.

• 47:45Speaker 00

  Which is not small at today's prices.

• 47:49Speaker 03

  No, yeah, that's a lot. Our, the Pelican, in the worst case, if you're charging it off of a diesel generator, is about two gallons per hour.

• 47:59Speaker 00

  That's a lot better. Even if you're doing, even if you're doing dirty charging, as you might call it, it's still much more efficient. Okay. Now, I know that you guys say on the site that the current ag drone starts at, if memory serves, 550,000. As far as farm equipment goes, that's not super crazy, but it's also not that cheap. So I'm curious about the, the ROI here and kind of like time to repay the purchase. Cause the way that I think about it, more efficient flight, probably need fewer human pilots because it's autonomous and you're using electricity and blah, blah. So I presume there's some savings baked into this. What is the time to recovery for people that are changing over to this type of ag work?

• 48:37Speaker 03

  Yeah, that's a really good question. So it's, you know, depends highly on how, how heavily you utilize the vehicle. So that's sort of one other benefit of Brazil actually is they're just crazy about the way they do agriculture. They have multiple seasons back to back and they just have an intense culture of like work. So, you know, we have customers who, one of our newest customers is planning on running three shifts with the Pelican. And so the thing at, you know, at most could operate literally 24 hours in one day.

• 49:09Speaker 00

  How comfortable would that make you as the guy who has to take the phone call if it goes down while it's working 24 hours a day? Cause that sounds like a maintenance nightmare, but maybe I'm overestimating.

• 49:18Speaker 03

  It's intense, but I mean, we're already, we have customers doing 12 hours a day right now. So the difference between 12 and 24 is, is not that big.

• 49:27Speaker 00

  In time, are we going to see like, uh, smaller farmers be able to like do ride share equivalent of rental for these things to, to handle their cubs? Cause I, I don't think everyone's going to need their own.

• 49:39Speaker 03

  Yeah, correct. Yeah. So in the US, for example, you would need to have a farm that is almost 20,000 acres in order to utilize, to fully utilize the aircraft. And that's huge. That is a, by US standards, very, very big. By Brazil standards. That's actually like midsize. So yes, in the US, it will be much more common to have spray as a service where, you know, contractor owns the vehicle and then services a number of customers in a surrounding radius.

• 50:08Speaker 00

  You know, we can make an acronym out of that. We could call it SAS. No one's ever used that before.

• 50:12Speaker 03

  Yeah, totally. Yeah, maybe our valuation would go up. Anyways, so the, you asked about economics and payback period. So yeah, it depends on the utilization, you know, two to three years roughly. But the thing that people don't understand is Pelican, yes, it is a lower cost solution. Yes, there is a different type of labor you use, a much easier to access labor pool. You can train someone to operate a Pelican in two to four weeks versus a aerial application pilot is, you know, 18 months. But really, the like, the reason people love the Pelican is because of its ability to spray really, really well. And so, you know, the simplest way to understand this is the cost of the chemical that is being sprayed

• 50:58Speaker 03

  is typically about four times the cost of the application. So it's this very precious resource.

• 51:04Speaker 00

  Oh.

• 51:04Speaker 03

  You're, yeah, and you're trying to like deposit this fine mist of chemical very evenly over an entire crop, including around the boundaries of the crop, which is very difficult. So the Pelican has like a bunch of things about it that make it just a like massively superior solution for actually applying the chemical.

• 51:23Speaker 00

  And right now, I mean, as you and I record this, the state of Hermos is still totally blocked off. And I presume that behind this comes out, it'll still be blocked, which is killing fertilizer prices around the world. So people are probably extra price conscious right now about these chemicals they're spraying over the crops.

• 51:37Speaker 03

  Yeah, totally. So input costs are, you know, like chemicals are a very, very significant portion of growing food.

• 51:44Speaker 00

  Man, that's, that's. It's worrying to me that the solution to that, which is apparently you guys, is not as applied as the need for fertilizer is because there's a mismatch between solution and crisis in our food supply. Anyways, for everyone listening to this, if you're confused why we're talking about farming here on Twist, it's because I'm working us towards a point, which is that Pika has put together a not just drone, but also software and the integration with the hardware to make a cool system that allows the company to really quickly build new devices. And this brings into play Dropship, which, as you explained to me before the show, is really the second generation of your cargo plane. So how did you get Dropship up and into the air so quickly? What have you learned? And then when will it reach the market?

• 52:29Speaker 03

  Yeah, great question. So just as general background, Dropship is a dual-use product.

• 52:37Speaker 03

  Within the commercial sector, it's used for logistics. Within the defense sector, it sort of forks yet again. It is used for contested logistics. And then it's also a very versatile multi-mission aircraft, good for carrying large sensor payloads, etc.

• 52:53Speaker 00

  Could it carry bombs in the little container? Sorry to be rude, but I'm curious.

• 52:56Speaker 03

  I mean, technically, it could. That's not the most like interesting use case for it, I would say.

• 53:03Speaker 00

  Okay. And why not? Because when I think about drones today, we think a lot about, you know, drone-based warfare in Ukraine and in and around the Middle East. And having a large drone that can carry more boom, sounds good to me.

• 53:15Speaker 03

  Mm hmm. Yeah, so our vehicles are designed to be quite reliable and low operating cost. And so like if you want to deliver, if you want to do more kinetic stuff, you don't build a vehicle like we did. Our vehicles last too long, basically. They're too nice. If you wanted to make a really low-cost bomber, there would be different decisions that you'd make than what we did with Dropship.

• 53:40Speaker 00

  And probably also like, you know, electronic emissions and noise and despite other factors that would have gone into making this a bomber if you wanted it to be one.

• 53:48Speaker 03

  Mm hmm. Correct. Yeah. And so, yeah, yeah.

• 53:52Speaker 00

  So you guys had a test flight recently, a drop ship, really quick process to get it, you know, from, as you said, CAD to the runway.

• 53:58Speaker 03

  Yeah. Yeah. So we went from like initial CAD renderings to first flight in 180 days, which is pretty awesome.

• 54:08Speaker 00

  Yeah.

• 54:09Speaker 03

  That was, so that wasn't the fastest, I guess, what we called Big Bird, the plane that we built in my parents backyard during Y Combinator. Technically it was faster, but Dropship is way more complicated and useful.

• 54:21Speaker 03

  So, yeah, it's, you know, it's the second generation cargo plane. First generation cargo plane was an all-electric cargo aircraft. Really cool plane. We built eight of them. We're actually gonna probably build quite a few more for commercial customers. That aircraft only has a 200 mile useful range. So that was a sort of key limitation for defense. Anyways, kind of the backstory there is we built eight of those. We sent three of them to the Air Force, got very positive feedback from them. Sent one to the Army, got really positive feedback as well and invitations to some, some big demos that are coming up in the next two and a half months. Air Force, we want to follow on contract as well.

• 54:59Speaker 03

  I think the thing that was so cool about this is we got really, really good insights into what it is that both Air Force, SOCOM and Army want from a attributable contested logistics and multi-mission UAS. So there's no like requirements written for this type of vehicle yet. This is too new. And so the main feedback that we got, willing to share it now because Dropship is out in the market, is basically like Pelican Cargo, the electric thing. Really awesome, super easy to use, practical, easy to maintain, needs much longer range. So ideally 1,000 miles of range with 500 pounds of payload, check, Dropship hits that. It needs to fit in a 20 foot shipping container. So all of our vehicles actually fit in 40 foot containers already.

• 55:44Speaker 03

  We designed them to do that, but they're just like two and a half feet too long to go into 20. So Dropship meets that requirement. The tail is removable, which is really cool.

• 55:51Speaker 00

  Ah, okay, got it.

• 55:53Speaker 03

  Yep. And then the last really big one was, or sorry, two more operation off of heavy fuel. This is true for any defense focused UAS. So JPA, JP5 diesel. And then the most critical one was the ability to airdrop payloads. So the

• 56:10Speaker 00

  Because the first cargo plane opened up in the back like a C-130 versus opening up in the middle like, sorry for the analogy, the Bombay of a B-17.

• 56:17Speaker 03

  Yeah, exactly. So the nose opens on both of them. But on the electric one, the whole floor was a battery. So there's no

• 56:25Speaker 00

  The nose that opened up, not the back. Oh, I totally misread that image. I'm so sorry for funny.

• 56:29Speaker 03

  No worries. Yeah, the nose opens on both. But yeah, we had this giant battery in the way. So we, you know, couldn't make the electric one airdrop.

• 56:37Speaker 00

  Which is, I feel like we're beating around the bush. The new drop ship is hybrid. So it has both a diesel engine, as far as I understand it, for getting somewhere. And then it has batteries for very quiet looping in around the area once it arrives.

• 56:50Speaker 03

  Yeah, yeah, exactly. So it's a really cool architecture. I don't think there's any airplane out there that's a hybrid turbo diesel architecture. So it's a parallel hybrid, like you said. So the diesel engine actually has a propeller attached to it. It can also charge the batteries in flight. And then the electric propulsion system, so the diesel engine's behind the fuselage, just a pusher propeller. Kind of normal UAV configuration in that sense. And then up in front of the wings are two very high power electric motors. And so those are used during takeoff. That's how we get this really ballistic takeoff performance, like takeoff and landing and under 600 feet.

• 57:26Speaker 00

  All three at the same time?

• 57:27Speaker 03

  Oh, yeah, absolutely.

• 57:28Speaker 00

  Oh, so it's just supposed to be going mad like a little beep.

• 57:31Speaker 03

  Yeah, it sounds really cool. It's this like combination of turbo diesel and then the electric. So yeah, the diesel engine is about, it's like just over 30 kilowatts peak. And then the electric motors in the front are each about 25 kilowatts. So when the electric propulsion system is running, you know, our peak power is essentially three times our cruise power.

• 57:57Speaker 00

  Yeah.

• 57:57Speaker 03

  So, yeah, so we use that for this like ballistic takeoff and landing performance. Use it to climb to altitude. And then once that altitude actually shut down the entire electric propulsion system. We have these neat passive folding propellers and then the airplane cruises just on the diesel.

• 58:13Speaker 00

  You know, I was really excited about Dropship, not because of its dual-use capabilities, but more because I was thinking about domestic commercial applications, like getting stuff places, like can I send my mom a couch? You know, but it sounds like our prior, our prior point that we talked about with your agricultural drones is that regulations may not be ready yet to turn Dropship into a domestic FedEx, if you will. Are, are we gonna shoot our own foot here as a nation and not end up with a lot more autonomous flight? Because it seems much more economically viable, environmentally friendly, convenient. Like, it just seems better to me.

• 58:49Speaker 03

  Yeah, 100%. So, I mean, the, the reason that we kind of pivoted away from mass manufacturing Pelican Cargo, the electric one, wasn't because there was a lack of interest from customers. It was really exactly the reason you mentioned. We could not get regulatory approval to fly a scaled beyond visualized site operation in a timeline that was relevant to us. So that, that was the biggest factor. Are we shooting ourselves in the foot? Yes, absolutely. I mean, I think there's this very, just getting back to the original point, like these hardware, software blended products are so important to our society. They're also going to create so much value, but their value is 100% determined by their exposure to the real world.

• 59:34Speaker 03

  You know, like SpaceX is worth how much it is because they blew up rockets, figured out how to stop doing that. And now, you know, the idea of a self-launching recovering rocket is just like taken for granted.

• 59:49Speaker 03

  There are very few companies have, who have figured out a way to actually collect that data. So PICO, we've done it. We had to go to Brazil. ZipLine, they've done it. They had to go to Rwanda.

• 59:58Speaker 00

  Africa, yeah.

• 59:59Speaker 03

  Yep. And then there's a bunch of other companies that have gone to Ukraine. So Ukraine is the, the other place to like learn, essentially. But outside of that, there are very, very few options, which is... Troubling.

• 1:00:12Speaker 00

  I want to make sure we get to supply chains, components and sovereignty. Now, if you were just making agricultural drones for Brazil, I wouldn't really care. But we are talking about the military, we are talking about duties a little bit. So how well are you able to source materials, components, or any sort of input without reaching into Chinese supply chains? And how much better can you do in a couple of years?

• 1:00:36Speaker 03

  Yeah, good question. So I think kind of because of when we started the company, there wasn't like a bunch of stuff we could just buy off the shelf to build these drones out of. And so we actually vertically integrated essentially every critical component on our product. The motors, batteries, motor controllers are our own design. All the avionics is our own design, airframes, et cetera, et cetera, et cetera. So we're already able to create NDA compliant versions of our product. There is some difference in sourcing. For example, the battery management system, we will manufacture here in the United States for an NDA product. For our commercial products, we'll manufacture the battery management system in China.

• 1:01:14Speaker 00

  What's the cost differential there?

• 1:01:16Speaker 03

  It's about 2X.

• 1:01:19Speaker 00

  That's, is that a high-priced component compared to the overall cost of the device?

• 1:01:23Speaker 03

  No, not, I mean, so for drop ship, no, there's two BMSs in the entire vehicle. For Pelican, it's more, there's 15 batteries that ship with each airplane.

• 1:01:32Speaker 00

  Yeah, yeah, yeah, so more complicated. Okay, that makes sense to me. I was just thinking that 2X didn't sound as bad as I was expecting, and I wasn't quite sure why.

• 1:01:38Speaker 03

  Well, so if you own the design, it's not that bad. I mean, so you can, for example, if we don't source from China, we could source from another low-cost region. The 2X is actually doing it in the United States, though.

• 1:01:50Speaker 00

  Oh, no way. Oh, huh, all right, well, that's better than I thought. Okay.

• 1:01:56Speaker 03

  Yeah, I mean, I think if you own the design and manufacturing in the US versus China is order of magnitude 2X difference.

• 1:02:05Speaker 03

  If you don't own the design and you're buying from an OEM and it's like OEM in the US versus OEM in China, then it's probably going to be like 4X difference.

• 1:02:13Speaker 00

  Yeah, quite a lot. Okay, that makes good sense to me. But you mentioned how when you started the company, things weren't available on the shelf, so you managed to do it all yourself. Did that slow you down materially? And I ask that because it doesn't feel like the company is moving slowly, but you've taken on a lot more, what we might call a technical risk by building so much in-house, that I'm kind of shocked that you've gotten this far, given that you really blank sheet of papered this entire thing.

• 1:02:37Speaker 03

  Yeah, yeah, for sure. So I mean, I think we have an extremely strong technical team. It has slowed us down. There are some components that have matured really nicely in the sort of OEM space that, you know, if we knew now how far they would have come, we maybe would have paused on developing our own. So the motor controller, for example, is something we did in-house. They've gotten bigger and bigger and better. And now there's off-the-shelf options that are vaguely similar.

• 1:03:05Speaker 00

  Got it.

• 1:03:06Speaker 03

  But you know, on the flip side, it's really hard to say because it's kind of the grass is greener scenario. Like, we've run into issues with our own designs and our own products, you know, lots of issues. We've resolved them all. Or we haven't resolved all of them.

• 1:03:20Speaker 00

  We're working on the last couple of resolutions, everybody. Give us two or three weeks.

• 1:03:23Speaker 03

  Yeah, yeah. Whereas, you know, we've had issues with our off-the-shelf products and getting to a robust resolution with those often is far more time consuming. So, yeah, it's tough to say. I think the biggest thing for us, though, is like, you know, if you think of the most successful hardware companies out there are a blend of hardware-software.

• 1:03:47Speaker 00

  Apple.

• 1:03:48Speaker 03

  You know, Apple, great example. Yeah, Tesla, DJI, perfect example.

• 1:03:52Speaker 00

  Sure.

• 1:03:53Speaker 03

  And I think like what those companies are able to do is make something incredibly complicated seem just really, really simple to the end user. You know, when you take a picture with your iPhone, it's phenomenally good. It's like better than my dad's super fancy DSLR camera now. And I don't know how or why, but like, there's so much going

• 1:04:15Speaker 00

  What are all these? What are all these little like camera things back there? I don't know what a single one of those does. Not once have I lifted up. And you know what? My pictures of my kids look fantastic. So thank you.

• 1:04:23Speaker 03

  Totally. Yeah. And so like the way you do that is if you you have to own everything. You have to own the hardware and the software, in my opinion. Like, that's how you make that really magical experience. If you try and integrate a whole bunch of disparate systems, you end up with this like customer experience where you can tell.

• 1:04:43Speaker 03

  And so so that's I think that's like the biggest benefit, but it's.

• 1:04:46Speaker 00

  Are we just talking around Boeing's decision to stop making things in-house and just supply everything externally and integrate all the systems? Because I feel like we're slowly circling the Boeing story here. Michael, I have to let you go, but one last question before I do, which is really simple. There is a lot of money flowing around the venture capital world today. However, I talked to a lot of people that are now building enterprise agent orchestration NCP servers. And they tell me that, oh, my God, people are not interested in us. So does the company have enough access to capital today to continue growing? If you still need more money to bring this vision to reality? Because I can see a really cool future in which we have tons of quiet, safe, friendly drones in our skies, doing quite a lot of work for us. And I want to get there quickly.

• 1:05:29Speaker 03

  Yeah, yeah, good question. I would say yes and no. I mean, we we've been able to raise the money that we need. If we had more money, we would move faster. So.

• 1:05:39Speaker 00

  If you're a VC listening to this, hello, come on. Stop backing SaaS companies that are slapping on an AI wrapper. Box something cool. We're building drones, y'all. Michael, a treat. Where can people find the company? And is there a job you are hiring for you want to shout out into the void?

• 1:05:53Speaker 03

  Yeah, great. So you can find us on our website, flypyka.com.

• 1:05:58Speaker 00

  P Y K A, not P I K A. Yeah.

• 1:06:01Speaker 03

  Correct. P Y K A. Very active on LinkedIn, actually very active on Instagram. Most of it's in Portuguese, I'll warn you. But really, really cool shots from from Brazil of our actual customer operations there.

• 1:06:13Speaker 00

  Fantastic. All right. Thank you very much. We'll see you soon.

• 1:06:15Speaker 03

  Great. Thank you.