CNBC Kate Rooney with Apoorv Agrawal & Bob McGrew (Founder & CEO Arda) - October 20th, 2026

0:00 Joining us now, Bob Muglia, CEO and founder of Arda, and Apurv Agrawal, partner at Altimeter. Welcome to both of you. It's great to have you both on set. >> Well, thanks for having us. >> Of course. So, the news here is Arda. >> That's right. >> Altimeter is an investor, so we'll get Apurv's take as well, but you have not spoken about this startup. You were at OpenAI, left the company to found this. Tell us what Arda is and what you're building now in the defense space.

0:24 >> Absolutely. So, Arda is speeding up American manufacturing. So, America is the best country in the world if you want to design an innovative product, but outside of defense, too often we end up building that in China. So, what Arda is doing is we're using AI to help manufacturers design and operate reconfigurable factories so that they can build innovative products in the US. >> You left OpenAI to do this. tell us about that journey and also how that informed what you're building here. I also want to get Apurv's take, but it's just so interesting given, you know, your role at OpenAI early on.

0:58 >> Well, yeah, and and even before OpenAI, I was at Palantir, which is where we met. I was the second engineer at Palantir. I ran engineering and product management there for the first decade of the company. you know, codeveloped what's now called the four deployed engineering strategy. And, you know, after 10 years, ended up leaving, thinking about starting a robotics company, joined OpenAI then, back when it was really only about a dozen people. >> What year was it?

1:23 >> this was 2017, January 2017. And, you know, I thought I was just joining a small nonprofit that had this big dream of building AGI. I had no idea what it was going to become. >> Here we are now. And so, you two overlapped at Palantir, little known fact. Apurv, did you start your career there? You were in banking before. You guys have known each other though for a decade. so, have you watched Bob sort of progress and go out and launch something? I'm sure this is something you had your eye on.

1:48 >> You know, I have to have the good fortune of knowing Bob for over a dozen years when we were both engineers at Palantir. And I'd say Bob has a track record of finding the biggest missions that seem improbable at the time but end up defining the next era. Started his career at PayPal 26 years ago when at a time when people thought moving money on the internet is risky. Palantir second engineer when people thought forward deployed engineering was too expensive or too impractical as a way to deliver technology.

2:15 Open AI bringing modern AI to the world and now speeding up American manufacturing. >> It's such an ambitious project and there's so much to talk about on manufacturing as well. How how are you going to do this? You are launching the company we should say too. You've been building this behind the scenes Anduril is a big partner. How has it gone in the last year and yeah just walk us through what the process is going to look like.

2:38 What are you exactly are you disrupting here? >> Well well actually let me let me even start with what is the problem. >> Yeah. >> So you know let's say 20 years ago the key product to build was a car. So if your company could build a car you were a great manufacturer. If your country could build a car you were a manufacturing superpower and the way that you build a car is you make a design that lasts for 7 years and so maybe it takes a long time to take that design to production. Maybe it takes 2 years and a couple hundred million dollars to build a factory that can actually build that car but that's fine.

3:09 The products that matter today are much more innovative. So if you have something like the iPhone each year you have a complete redesign of the iPhone. In Ukraine for example you can only wait a few weeks between you know a change in the battlefield and when you change your design and actually get that design into production for your new device. And so because of that when the design changes so frequently it's incredibly human intensive. You need a huge number of manufacturing engineers to go and build the assembly lines. You know in America does not train enough manufacturing engineers and you also need a huge number of human workers to go actually assemble the devices.

3:51 You >> can when a design changes you can retrain a human in a day, but it takes 3 to 6 months in order to change the automation to reflect the new design. And And frankly, the fact that we've put the assembly of our most innovative products in China is a huge cornerstone of their industrial strength. So, what are we doing, right? >> Yeah, what How are you fixing this? >> Yeah. So, we The first step is that we've built a software and AI product that basically allows manufacturing engineers to build assembly lines much faster.

4:21 >> So, give me like a an example of if I'm a software engineer or what what does it actually look like? >> Yeah. So, so it's a user interface and an agent that does for manufacturing engineers what Claude or Cursor does for programmers. >> Do you have your own model as part of that? >> So, we use our own models, we use existing models, and we use a lot of specialized algorithms that the models actually have that we use frontier models in order to to operate.

4:48 >> You have also been an investor in the AI space. Maybe tell us about the opportunity. What does it look like for I mean, it seems like a a massive TAM, but what what brought you to this opportunity? I mean, obviously, you know Bob, so there's a founder aspect here, but talk about it maybe from the investment side. >> Yeah, let me let me take this a level level up, which is why is it a good idea to bring back manufacturing to the US? Is it even a good idea? Should we be responding to cheap labor overseas with cheaper labor?

5:15 Well, the answer is we we want to be independent on the most strategic areas, defense and AI. That's where our order is starting. And you know, to give you a data point of why that's important is, you know, in November 2025, the US Army announced, "We want to procure a million drones in the next 2 years." That's about an order of magnitude, about 20 times more than the number of drones procured last year. In AI data centers, you know, our power transformers have a 3-year lead time.

5:43 And the strategic need is real, but the The question is is it economically possible? And this is where AI and robotics come in. Right, how Artav is doing it, you know, act one as Bob described is a Claude or cursor for manufacturing engineers. And kind of what cursor did to software engineers is, you know, speeding up manufacturing engineers. >> Mhm. >> That is a key ingredient to the second act, which is, you know, building flexible auto automation at scale.

6:12 Robotics today are very good at, you know, building the same product again and again, but if you change that a little bit, you know, made a 5% change or 10% change, it gets quite hard. But the vision with Artav is to make that flexible automation at scale. and and that's what that's that's what we're excited about. >> What is the revenue model? This is a subscription product? >> So, it's usage-based. So, similar to Claude, for example. And and actually I really like that because it means that for us, we don't get paid unless people actually use the product.

6:42 >> Seems like there is a sort of a shift, and not to interrupt you, but I feel like that is sort of the way the software industry is going, where you're seeing more usage-based pricing or outcomes-based pricing versus just traditional subscriptions. Is that sort of how you're, I mean, building it, but it also seems like a way to sell it, where customers don't necessarily need to pay you unless it works. >> Well, we actually spent a long time trying to figure out what the right pricing model would be. Maybe we would go with a a flat fee, maybe we would go with per user pricing. And in the end we said, well, what is the thing that is most correlated with the with the models and the software actually doing work for the users, doing work for the company.

7:20 And we realized that that was usage. And so, you know, tokens effectively, but a little bit more complicated for agents. And so, we're really happy with it. I think it works well. >> You mentioned the underlying models. Are you using OpenAI or you leaning on open source more? What is the What is the behind-the-scenes sort of makeup of of the technology and your what you're using? >> Yeah, so it depends. When we're working with a company like Anduril, which is a defense customer, so we're actually deployed inside their systems.

7:48 So, we're able to use whatever models that they have access to and use those to power our agents alongside the custom algorithms that we've written. When we're working with customers that are you know, not as secretive or as as much in need of that as defense, then we're able to use a multi-tenant stack on ours and then we can deploy you know, whatever models that we have access to. >> Are they not allowed to use Claude? I'm thinking of models that they might not be able to go for. Like what would What's the difference between Anduril and I don't know, a software company?

8:20 >> Well, I I you know, one of the differences is they're working with ITAR data. So, this is data that is defense materials. This is data that you know, basically if a non-US citizen actually looks at this data, then you you know, who's holding this data, you can go to jail over that. So, that's an extreme level of sensitivity and security. With our background from Palantir, you know, this is something we're very very flexible with and very familiar with.

8:43 And so, you know, for for some of these customers it makes sense to actually go deploy inside their company. This is an aspect of the forward deployed engineering model. For other companies with less sensitive data, then it's better better and easier to host it on our servers. >> It's really interesting. I feel like there's Palantir, there's Anduril, there's been success stories in defense tech especially lately. Has that sort of opened the door for you to be able to build something like this?

9:08 I mean, this was probably 10 years ago, but it felt like there was almost this taboo in Silicon Valley where defense tech maybe wasn't investable or you just didn't have the examples of really great companies. What does that look like on your end? Is it easier to sell an investment like Arda to the rest of your partners? >> The The saying used to be if it falls and hurts a foot, it's not for venture capital. But, but I think we've come a long way from that.

9:30 You know, I think it's very aligned with some of the technology advances that have happened and obviously how geopolitics has played out. I would say in defense in particular, you know, Anduril, Palantir, and and a series of other great businesses like Castilian and others have demonstrated key strategic needs. They're filling key strategic needs for the country. And I think for ARDA in particular, you know, I don't want to forget the the the second act that that that, you know, Bob is and team are working towards, which is ultimately to deliver finished goods.

10:02 >> Mhm. >> Maybe Bob, tell us about act two that you're planning. >> What is that Yeah, what's act two? >> Well, so, you know, if if act one is is sort of automating manufacturing engineering with AI, the second step in order to actually be able to build factories that can be reconfigured is to automate those assembly lines in a way that you can reconfigure those assembly lines as quickly as you can teach a human worker how to accomplish a new task.

10:25 And so, right now, if your design changes very frequently, the only place you can really build your product is in East Asia because you need an army of human workers who can be retrained quickly. and you can't wait for automation to catch up if your product cycle is only 4 months long and it takes 6 months to build the automation. You you can't use automation to build your product. But, if you can change the automation in a week or in a few days, now, you no longer have to build in Asia because you can build it in an automated way, so you're able to do that at US wages. And so, that I think is the key to allowing fast-paced innovative products like smartphones, consumer drones to actually be assembled in the US.

11:05 >> So, what does that look like? Is that robotics or like how do you actually physically build that? And how much of it is is automated by robotics? >> Yeah, so, you know, we started working with Anduril. and Anduril is, you know, a US defense company. Most of their work is not highly automated. and so, the starting point for the agent is to take the design and be able to build a, you know, assembly line staffed by human workers on the other side and write work instructions for those human workers.

11:34 And the next step is for that same agent to say, "Okay, well, I have the work instructions for a human worker, and I know how the how a robot would need to do this." That's actually how we build the work instructions. Is we use robotic motion planning to figure out how all the parts are supposed to come together. And so now I have the motions the robot is supposed to take, how do I actually make that work in the real world so that you have a robot that's actually taking those motions and successfully assembling the product.

11:57 >> Very interesting. >> And that that's a big challenge for AI. >> There's also So, you talked about some of the geopolitical tailwinds for a company like this. I think when people hear automation, they often think fewer jobs. Have you gotten any pushback I don't know if you spend time in DC at all, but as you talk about automating, is there any fear that this could result in fewer jobs? I mean, there's also been this debate about taking jobs back to the US. I mean, offshoring and manufacturing's one thing, but then there's a labor market discussion.

12:27 How do you approach that? >> Well, no one has ever successfully built a smartphone in the US. Never successfully assembled one. And so if you can build a smartphone in the US, that is not taking away American jobs. That is you know, bringing American jobs. you know, you know, for the iPhone, there are maybe a million people assembling iPhones or components in Asia. >> Really? >> if you were 90% automated, which is a very high level of automation, and you were doing that in the US, that's you know, net 100,000 American jobs.

12:57 >> Wow. >> So, it's a way to bring jobs here where there's a >> Okay. It's a good interesting way of thinking about it. What's your take on it, Deepak? >> I'm going to add to something Bob said. Those jobs, even if the jobs existed, there's the people don't exist in the United States. Right? There's about 4 million open jobs for manufacturing engineers in the US. We forecast only about a million and a half of them will get filled. And so a product like ours does is near mandatory to to to achieve the productive out productivity that we need with the limited human capital we have.

13:29 >> What's the pitch to, you know, the Trump administration maybe who would be looking at this and saying, "Okay, we really want we want to more American jobs. We want to bring manufacturing onshore." What is sort of your bull case and what this could look like? >> Well, I think, you know, what what you have is, you know, you you have to start with defense, right? Where you're building these things in America already. And we can help those companies iterate much more quickly. That allows them to ship more products. That's more American jobs. The next place that we see a huge amount of demand is energy, you know, the data center supply chain.

14:03 We're working at the company there. they can iterate more quickly. That allows them to get to market faster. They can sell more. They can generate more jobs. As we we work more and more with automation as these products mature, then you can start bringing more complicated things. Maybe you bring routers back. Then maybe you can bring consumer drones. You know, really the the queen of all of this is the smartphone market, which is the largest slug of demand that there is. And in many ways it's it's sort of the supply chain for the smartphone market that has built Shenzhen into the powerhouse that it is.

14:34 >> I don't think I fully appreciated that. What would this mean for semiconductors? I mean, there's been so much talk about Intel on trying to bring the chip business back to the US. Is there a world in which you could be in operating in that arena? >> I I think that it's it's different but analogous to what we're doing. So, if you think about a fab, a fab is actually just a giant automated factory. And so, you know, that's already automated. What we're trying to do is take the same level of automation that you see in a chip fab and apply that to assembly, to putting, you know, many kinds of heterogeneous parts together.

15:05 >> Interesting. I want to talk about the financing as well. So, you mentioned you're in stealth mode. I mean, you haven't launched fully in in the public. This is the first time you're talking about it publicly. Valuation of more than a billion dollars. Is it 1.5 billion? >> 1.5 billion. >> Post. what gives you the confidence to invest in this early of a company at that price tag? >> Well, look, it's all about the opportunity and the the team going after it. manufacturing globally is a $16 trillion industry. in the US alone, it's $3 trillion. And you know, we think that Arda and the vision that they've outlined will continue to capture a significant portion of that.

15:44 And so, yes, the price is steep today, but the long-term opportunity is very high. And second, you know, we ultimately it's about the best teams going after the biggest ambitions. And Bob, Augustus, Alex, and team, we think are the best team to go after a big audacious mission. >> What do they need to do in the near term to justify that price tag? >> Yeah, I think I think about it in two ways. One is we got to we got to go deeper with our current set of customers, you know, that's defense, more in defense, that's AI data centers, transformers, as Bob outlined, that might be consumer electronics at some point, that might be energy at some point, and adding those verticals as proof points and expanding the business. That's act one.

16:23 And then act two is the the the assembly line, you know, and so that might be medium-term, a couple of years out, but that's how we think about the business. >> Bob, what was demand like, you know, going out and fundraising? You've had this long career in Silicon Valley. You haven't been a founder before though, right? You've worked at all these companies. What was it like to go out and do this? And what will you spend the money on? I mean, where does $230 billion go?

16:46 >> Yeah, well, I mean, yes, first-time founder, multiple-times early employee, you know, it's harder than it sounds. You know, I will say a year ago when I started raising money for the first time, the hard part was to get venture capitalists interested in the idea of working in manufacturing. You know, it seemed so so so obscure, so, you know, maybe, you know, low revenue, low margin, and so, it was really hard to articulate the vision that actually worked. And I think what what has really helped us over the last year has been the work we've done with Anduril. So, we started with them a year ago. We met with them even before we had incorporated the company and they told us what their challenges were. And in January, we got access to their systems.

17:27 In April, we had our first user who was building an actuator, which is like a you know, the kind of thing that moves a robot's arm or adjust the flap on a missile. It's maybe 100 parts. And he was able to to design it and build it much faster with Arda than he would have otherwise. And then from that we've been expanding. We're now used across all of their product divisions. We've been working with much more complex parts. I think a great example of what we've where we are right now is that we had them they were using us to build the avionics box for missile, which is where all the electronics go. It's 2,000 parts.

18:00 It would have taken them a week to do without Arda. He was able to do it in two days. >> Wow. >> And so, that is the kind of progress, you know, going out and raising again, you know, was challenging. Fortunately, I had the luck to know Apoorva from before and have his confidence in the work we're doing. We've all you know, we're also co-led this round was also co-led by Excel and by Costanoa Ventures. The money is really for us now that we have spent this year working with Anduril. We are now ready to go launch and you know, begin working with many other customers. So, this is for us to expand our customer base, to continue hiring the team, and of course, to buy compute for the work we're doing with robotics and AI.

18:41 >> What are the other pain points you hear from customers? We talk about supply chain and the cost of things like memory, but other other things I might not think about copper, for example, raw materials. It seems like prices are this bottleneck. What are the other bottlenecks or issues you've seen as you're going out and working with some of these manufacturers? >> Well, I think one of the one of the big issues you see in the supply chain it's yes, it's the prices, but worse it's the uncertainty.

19:06 And so, particularly if you're trying to build a supply chain that's in the US, often find you'll be working with one supplier, you know, they'll be, you know, providing you something you're very happy. You get a new order. You know, you need to build 10 times as many missiles, let's say. Well, that supplier can't keep up. And so you now, you need to go find a new supplier. And then what does that do? That feeds back into the design problem. So now you need to redesign the the missile you're building so that you can actually accommodate that new supplier. And so everything just involves this like much faster iteration because everything is changing all the time.

19:38 >> Interesting. Is there I mean we talked a little bit about your time at OpenAI. Is there sort of an OpenAI halo effect? I think of Mira Murati as somebody who kind of left around the same time you did. Does that factor into to how you think about backing some of these founders? It feels like between Mira, Ilya, you're another great example of it. Sort of an OpenAI mafia forming. >> >> How did you approach it?

20:00 >> Well, well, Bob is Bob is the first one of the only one of the three who's not started a new lab. >> >> also the only one we backed, but I'm a huge fan of Mira and huge fan of Ilya. but I think it's it's it's it's a lot more than than that. Obviously that helps to have led the research team at OpenAI so you know all the ins and outs of of of OpenAI. In fact, I think Bob, when you left Bounder, you had joined OpenAI to maybe do some robotics work 10 years ago. And you know, a little known thing called Transformers came along the way. But Bob's been thinking about robotics for a long time, at least 10 years. And you know, what what Bob hasn't mentioned is when he left OpenAI, you know, we spent a lot of time with Bob and he thought about a lot of different things he could have done, you know, in in AI, in AI diffusion, in robotics, and he picked this. And there's as much conviction in all the different paths that he tried and did not pick and and the one that he did eventually pick. And I know he's very thoughtful about it.

20:56 And so what going through that journey with Bob got us the conviction that now is the right time to go after this. Right, the response to cheap manufacturing labor overseas is not cheaper labor. Mhm. You can't compete with five bucks an hour in China with with four bucks an hour. That just does not exist in North America. The way to do that would be to fundamentally rethink the manufacturing process with with the latest advancements in AI and robotics. And and so that's that's what got us there.

21:22 >> Really interesting. Have you talked before about why you left OpenAI or did you leave to start something new? I wonder just the process of leaving a company at a time when it really was this rocket ship. I mean, it's still a rocket ship post chat GPT. What was the decision like to walk away from OpenAI? >> OpenAI is an amazing place. And you know, being there from early in 2017, you know, I remember sitting down in 2018 in a room with Dario and Ilya and Sam talk you know, right after GPT-2 had launched talking about what it would look like when you know, a model could write your email for you.

21:57 Right? And and that that seemed so far away. And you know, the Anthropic people left, you know, I at some point realized I had been holding out, you know, when would AGI actually happen? Was there going to be a weekend when we would all look at this and say, okay, the models, they're intelligent now. Now they're at human level. And over time I realized that that wasn't going to happen. That I think what you're actually seeing is this very smooth capability increase for the models. There's not going to be you know, one moment when the singularity happens. It's actually very smooth. And so during the coup or the blip or as some people like to call it the kerfuffle, >> The weekend, yeah.

22:35 >> Yeah, that one weekend, I remember calling all of my reports and telling them that we had signed this document and if the board didn't tell us why they had fired Sam, we would all have to leave and go to Microsoft. And I remember thinking to myself, you know, I don't know how much longer I can manage being at OpenAI, but if I can just ship reasoning, if I can do that, I think that's one of the key breakthroughs, maybe the last key breakthrough left before AGI, that if I could just ship that, that I'd be happy with my career at OpenAI, and I'd be ready to move on to something else.

23:06 >> Wow. >> And so, the company came back together. It did not fall apart. you know, we continued for another year, and I remember, you know, at some point telling we shipped reasoning, and I told Sam, "I think it's time for me to go. I could stick it out for a little bit longer, but, you know, this is your This is your chance." >> You mentioned it's harder than you thought being a founder. What are the things you took from OpenAI, PayPal, and Palantir that you want to have in the DNA of a company like Arda?

23:36 >> Yeah, I think, you know, one thing I learned from Palantir, you know, Alex Karp is incredible mentor. and I really learned from him that a lot of what you do in building a company is very much about personal relationships. It's about how I interact with you, whether we trust each other, because at some point in any relationship, you're going to ask someone to do something hard for you, something that they don't want to do. And if you have a relationship where you know that I'm asking that from you because I want what's best for you, then we'll be able to move through this, and the relationship will survive. and that is a huge lesson for me. I think from OpenAI, you know, what I what I got from watching Sam was just sort of the freedom to have this incredible vision.

24:19 You know, can you imagine the idea of building AGI back in 2015? It No one believed in this. No one believed in this. Even the people at OpenAI didn't believe in this. >> Elon Musk didn't believe in it, famously. >> Well, that's a complicated situation. >> >> It's a different conversation, but yeah. >> And so, you know, really, I think, you know, it's it's bringing that of, you know, can you have this very amazing vision? Can you have a very talented team to put it together? And can you find a way for everybody to work together?

24:50 >> It's It's probably a cool experience to be able to take all of that those things and kind of start over, and take the best from your mentors. Alex Karp is one of our more interesting CEOs. >> >> I feel like he's always got a really hot take ready to go and he's he's live. does that align for like when you look at a first-time founder as you mentioned, you know, you've known Bob for a long time. You have a lot of trust built up. but you must have a certain muscle memory, too, for what you look for when somebody's going to start a company from scratch. What does Bob have that sort of got you there?

25:22 >> Yeah, look, I think aside from the you know, everything we've already discussed, I would say one of my core beliefs in starting at Palantir is the biggest aligning force for a for for a team is a big mission. Right, you know, kind of how I joined Palantir is a lot a lot of my family's in the in the in in the armed forces and I've noticed that Bob has never made small plans. The team The team has only making the most ambitious plans. I mean, could it sound crazy to bring back manufacturing to the US? Yes. But that That was his track record from from from PayPal and Palantir and OpenAI. And I think the first-time nature of a founder is is one that, you know, having gone through a journey of over a dozen years with him, I I feel very comfortable with and seen him operate. I've seen all the great achievements Bob Bob has had that are no short of any founder would have achieved.

26:16 >> Well, I want to go back to OpenAI for just a second. AI safety's been such a big headline this year. I would think in defense the bar is so high to make sure that nothing goes wrong. There's not a rogue agent or there's not some sort of issue. And how are you thinking about it, you know, now that you're building a defense company? You're relying on other models that have had issues. what's your take on this entire AI safety debate, but also being a client of them and seeing some of these issues, you know, you've been at these companies. talk us through just how you think about it.

26:50 >> Yeah, well, I mean, I think it was it was a little easier to think about AI safety early on because the models couldn't actually do anything. If all the model was doing was, you know, and reading your emails at best, then you didn't really have to worry that it would do something terrible. I think what we're seeing now is that one, people are relying on the models for to solve problems that actually have stakes. Right?

27:13 >> >> In our case, if the assembly line for a critical product has a problem and introduces an error into the system, then that's a huge problem, not just for us, but for our customers, for, you know, their customers. And so, you know, if you're going to use it for something critical, in our case, we see this is as something where the human is always ultimately approving the design. There's always, you know, a skilled professional who's actually operating the model. What the model does is makes them a lot faster. That's how we write our code as well.

27:44 the second thing, you know, that of course we've seen is that the models are now much more capable. And so, for example, you know, maybe you ask it to go design an assembly line and it goes and hacks some third party. Well, you know, that's a problem. This is I think the the central problem that the the model companies need to handle. >> It seems like a liability question, too. I mean, you've been to these companies, but is it the startup's fault? I mean, if you did that and then disrupted something for Anduril, you're kind of on the hook there. But also, you're a customer of a technology that might not always work. I mean, they haven't had any examples, we should say in enterprise, but I I also just worry or wonder about liability in all of this.

28:25 >> I I think the right way to do it is to make I one, I I'm not sure anybody knows how liability works in these situations now. I think we could go back and forth on who's the right person to be liable. I think the right technical solution though is a multi-layered approach of really having good, you know, can we take lessons from cybersecurity here? In how even if the models are doing something they're not supposed to be doing, that we can monitor them, that we can stop them from doing the wrong thing.

28:49 So, we can apply some of the same lessons I talked about in terms of, you know, there's always a human in our situation, there's always a human, you know, guiding it and judging it. Maybe, you know, for some of these other situations, there's always an AI watching the AI. So, you know, if if AI are best guardians, then, you know, let's have the AI's guard the guardians. >> Well said. It's a it's a good way of putting it. I wonder about part two, as you look beyond OpenAI and Anthropic, as these companies now get to the trillion-dollar plus level in private markets, where are you seeing opportunity? And one founder put it to me this way, that they were looking for things that were not disruptable from Anthropic, at least.

29:25 Where is sort of the next green space? Where are you guys looking to invest outside of the AI labs? >> Absolutely. Well, I'd say, you know, the ChatGPT moment for for large language models, we're starting to see cusps of that in a couple of different industries, right? I would put robotics probably next on the line. Obviously, some of the work that Arda's doing, and you know, we're seeing green shoots of great opportunities being formed in robotics.

29:54 Now that robotics in of itself is applied to, in the case of Arda, for defense and AI and in in in life sciences and in in wet labs, there is early shoots of the same work in life sciences. Labs like Anthropic, labs like Isomorphic, using all the great work done in AI to drug discovery, the development process, and ultimately building drug candidates. and then a long tail of other ideas where, you know, we're still still monitoring them, like material science.

30:25 but, you know, we're we're hopeful for the ChatGPT moment in many many industries. >> I think everybody wants to have their ChatGPT moment. Last question, we'll wrap up in a second, but what else needs to go right for manufacturing to come back to the US? I would think Arda is a small part of that, but are there other macro things, geopolitical things, or even these policies that would encourage manufacturing to come back, anything bigger that could actually help you guys accomplish this bigger goal?

30:51 >> Well, I think, you know, our focus is on assembly, and you can only assemble things if you can actually get the parts. And so, you know, beyond what we're working on, I think there's a huge amount of work that's required in actually making sure that the supply chain, you know, where appropriate is an American supply chain or an allied supply chain, or even that other companies like China do not restrict the availab- the availability of these parts or the underlying raw materials to American companies. And so, I think that is the complement to what we're doing.

31:22 That's required. >> a trade war complicate this? The tariff conversations calmed down a little bit, but have you, in building this, were tariffs at all a headwind or just the changes in prices and some of the uncertainty around that? >> Well, I think I mean, I think a a trade war would probably not be great for the US, but I think it'd probably be good for our business. >> Interesting. >> in that it will force more people to build more things in the US. And our focus is on helping American manufacturers do that. You know, the more things change, the more work that we have for manufacturers to help them change things. So, I don't necessarily want to see a trade war happen, but I can't say it wouldn't be good for us.

32:01 >> Guys, it's a fascinating conversation, and thank you so much for being here and explaining all of this. It's sort of a new topic. I'm doing AI coverage every day, but it's an interesting space, so thank you both. >> Thanks for having us, Caroline. >> Yeah, I appreciate it.