← This Week in Startups
This Week in Startups
$2.5B Chip Heist, The Future of American AI, and Purpose-Built Robots | This Week in AI Ep 6
01:15:24
Loading summary
Transcript174 lines
- [00:00]
A
Hey, it's Oliver from this week in AI, the brand new podcast from the team at Twist. We're dropping a sneak peek right here in your feed to show you what we've been building. If you enjoy it, join the community at ThisWeekInAI AI or find us on
- [00:12]
B
Spotify, Apple Podcasts or YouTube.
- [00:14]
C
AI, I think will transform and we'll continue to push the world forward and it will affect a lot of jobs and people. Upskill. The question is, what are they upskilling into?
- [00:22]
A
Being a cpa, an accountant or a lawyer was considered professional services, not commodity. And here we are in 2026 and we're like the bottom 50% of those. Jobs are chores that machines can do easily.
- [00:37]
B
When you see technology companies saying they're taking on manufacturing in the kind of ways that you see, you have to understand these sectors have not changed for the most part in the past 40, 50, 60 years. Get robots everywhere. Get AI everywhere.
- [00:49]
A
Thanks to our friends at PayPal, the exclusive sponsor for this Week in AI try the payment and growth platform that's trusted by millions of customers worldwide. PayPal open start growing today@paypalopen.com. all right, everybody, welcome back. It's episode six of this Week in AI. We decided to start a dedicated show just for AI. It's how I meet the smartest people in the world. You can subscribe to this podcast at thisweekinai AI thisweekinai AI There's a substack as the number one AI podcast in the world. We've got two all stars here this week. Jake Lucerarian, we've had on the program before. He's a friend of this week in startups and all in. He's the CEO and co founder of Gecko Robotics. They deploy robots for very specific mission critical infrastructure products. They're not humanoid robotics, which have become very in vogue all of a sudden. Jake. No. You've been working on purpose built AI robots for, is it seven or eight years now?
- [01:49]
B
If you count the college years, it's been 13 years.
- [01:52]
A
But Gecko Robotics is more of a seven, eight year story, if I'm remembering correct.
- [01:55]
B
Yeah, that's about right.
- [01:57]
A
And so what's the state of the art now? Give people an example. Or maybe my team can pull up a video from your YouTube channel of how these robots work. People can visually see them here, what they're doing. So here we see cantilever. That's your B2B product. Walk us through it.
- [02:15]
B
Yeah, that's a software. So basically actually, it's actually been 13 years. I started building this in the college dorm season, which it's been a long time.
- [02:22]
A
You and Michael Dell.
- [02:23]
B
I know, I guess.
- [02:24]
A
So it's hopefully the same outcome.
- [02:26]
B
Yeah, hopefully. Yeah. Yeah. I like to like to put, put some money into the, the funds as well for the kids. But yeah, what you're seeing here is cantilever. So basically, you know, 13 years ago when I was in college, I thought to myself, man, there has been so many deaths of robotics companies in the world. There's so many important jobs to, to for robots to be, to help out with and solve for. And so, you know, what's this Delta? And the base of the Delta that I figured out back in college was this idea of gathering information data robotics to help drive better outcomes. It's this whole idea, if you're building robots just to build robots and scaling those, that leads to a commoditized future. And in reality you're not really delivering on the value that the robots are able to actually gather and collect. So gather the information about the health of the built world was the original idea. Basically build Minority Report but for physical structures and be able to predict a catastrophe before it happens. And then that's begun to build now into this software now that as all these robots are feeding information data into to help to optimize how infrastructure performs, how healthy is it and help drive things like how do you make a kilowatt, you know, for less BTUs, how do you produce more barrels per day with less cost? How do you get a ship at a dry dock faster or manufacture a new vessel in quicker times with higher, higher value, with higher quality and, and speeds? These are the kinds of things that we build robots for. Very specific, mission critical.
- [03:43]
A
You started this long before the chat GPT moment when you were doing it, it was machine learning, you know, on the margins for this robotics company. So you were predating large language models, hitting the fidelity they have now. How has that impacted the business now to have the claw at 4.6 is out there, you know, and just all this new Gemini stuff, et cetera.
- [04:07]
B
Yeah. Back in the day, I mean, we were just using hobby as motors and gearboxes, planetary gearboxes, and we were trying to ensure that we could build robust systems and building them in the environment. When Silicon Valley actually was like, you know, put it in a lab, make it autonomous and then launch it. And we just didn't believe, agree with that fundamentally, you know, back in the YC days in 2016. But now what's happening is you're Beginning to get this super, super high focus on what's the pragmatic impact of artificial intelligence? Especially for the companies like, you know, the energy, oil and gas companies, the power companies, the manufacturing companies, the defense, and now Department of War are completely looking at how impactful can robotics and information data sets be to actually drive better decisions, outcomes, et cetera. And so what you're, you know what, what we focused on is, is building robots that actually affect things today. Not a painting, a vision of five or 10 years. And the models are putting a, basically a huge, a huge freaking spotlight on the importance of really important, valuable data sets that don't hallucinate. Especially with, you know, things that if they do hallucinate, could cause an explosion and kill people.
- [05:06]
C
Yeah.
- [05:07]
A
And maintenance is one of those crazy things that people ignore but have massive impact. All right, and our second guest is Chris Lattner. He is the CEO and co founder of Modular that you guys build Chris a layer that will let you deploy models on different types of hardware. And that's important why? Maybe explain to the audience. I'm sure you have questions for Jake too.
- [05:30]
C
Yeah, I was totally going to explain
- [05:31]
A
to the audience what you're building.
- [05:32]
C
I was totally going to grill Jake, but I'll happily jump in.
- [05:35]
A
No, no, in about one minute. After the audience understands what you do.
- [05:38]
C
Totally cool. Yeah. So we're building a software layer that enables people access to lots of hardware. So the obvious problem that we all face is that AI is everywhere, should be running in massive data centers, also on your wrist and everywhere in between. A lot of the world is really consolidated around the Nvidia platform, which is really amazing and it's very powerful. But there's a lot of chips from other players too. And what we want is we want more people involved in the ecosystem. We want to make it easier to adopt this technology. We want more hardware vendors in the space. We want people to have choice. And so that's a deep tech problem. But it really comes between, you know, like Enable developers that, you know, are typically thinking about CPUs to get into the GPU era, get into AI that's more customized, build more application specific optimizations and use cases and things like this. I think Jake's probably a leader in the space. And what I was going to ask Jake is, I was going to ask him, you know, so as you know, the whole world is discovering robotics. You've been at this for 13 years. What does it feel like to have a 13 year head start?
- [06:37]
B
Well, yeah, the party. The party is jam. Packed now with so much funding and folks who are, who are, who are jumping into it. Listen, I think, I think it's, it's, I am so thrilled. I mean it's just incredible. I mean I am very excited and optimistic about, you know, what, what the future will be with robotics and how, how in particular it makes us all focus on the first principles here, the first principles of artificial intelligence, the first principles of how do you actually build an economy and create this incredible growth and prosperity, you know, for the world. I mean there's a very optimistic future. The key is being deterministic though. And I think that's maybe where we're like lacking a bit determinism as it relates to, you know, making sure that if you cross the bridge and won't collapse or making sure that if you have, you know, you know, right now we have like two, you know, 200 every five ships are stuck in some dry dock or pier side somewhere. You know, and that's, that really affects like the deterrence and, and the geopolitics around the world. Or you have, you know, refineries that are shut down for you know, months at a time and, and you just have increased costs of energy. I mean these are all things that robotics deterministically could be focused on improving. And that's why we built, you know, Gecko itself being very mission focused robotics to help us understand, hey, not all data is important to go get and gather. Hey, not every action is important to solve for. And then more importantly when you have that paradigm being able to put humanoids to work in high ROI use cases, this is the key. And I think that that's underlying infrastructure on the software side to help guide where is, how do you employ robots. And I think that you see this with, you know, Travis, Travis put coming out with Adams co. I mean, you know, he's, he's basically the manifesto he wrote is just like hey, that was my manifesto 13 years ago. And he's absolutely right.
- [08:19]
A
Chris, when we think about your product, what are the hardware and language model combinations that are emerging? This year we bought a Mac studio and we've been testing Kimi. We've also been testing it in the cloud, but having Kimik2.5 I guess is the latest version. We're running that on Apple Silicon with I don't know if we have 128 gig or 256 gig and it's reasonably good. It's not Claude, it's six months, 12 months behind it. Um, but I, it is free essentially. So that's kind of the right price. Claude is expensive at scale with lots of open claw agents. And so what are the combinations that you see most often? Is it people trying to run on Mac Silicon, on Intel, you know, on commodity stuff? Where are the combinations showing up?
- [09:14]
C
Yeah, well, So I think GPUs have captured the world. And so since the ChatGPT moment, GPUs have really taken off. The inference side of it is huge. We currently support Nvidia, AMD and Apple Silicon. And so those are the three that we currently support and can scale across a number of different variants of those. It's super funny if you look at the consumer side that you're double clicking on because Nvidia has the DGX Spark, AMD has like the Strix, Halo and other crazy kind of hobbyist pro systems that are, that are cool boxes you can buy, but the chips inside of them are all slightly different. And so I don't know your experience, Jason, but getting the stuff set up and actually getting the latest models can actually be a real pain. And have you ever double clicked into why that is?
- [09:58]
A
No. Tell us. Yeah, this is a great opportunity, I think, to educate the audience on what the issues here are in terms of compatibility and making this easier for consumers or even prosumers if you lean in
- [10:10]
C
and you really get deep into it. There's both structural business reasons, but there's also just legacy accents of history reasons. And so the structural business reasons are that turns out hardware companies don't get along with each other. They all care about their products. Apple for example, they make great systems. I'm a veteran, I'm a huge fan of what they do. They don't really get along with Nvidia or AMD or whoever, Grok or lots of other people that go out there and build systems. So of course they build software for their chips. What that does is that then fragments the entire world on top of it. But developers all want choice. Like you want to be able to run on like if you get a Nvidia box or an AMD box or you get somebody else's chip, you want to be able to run it, but you have to switch to different software stack. And the problem is that there's never been that unifying layer. Everybody just builds on top and builds on top and been building up layers and layers and layers of cool stuff and very interesting capabilities. But it's all kind of duct tape and bailing wire. And so if you go change anything, it breaks. And so our approach on that is go Burn it all down. Go build software that goes and replaces the software that the hardware vendor uses. And so we love Cuda. For example, we use, we can interoperate with Cuda if you want to, but you know, our native stack replaces Cuda and that's actually a pretty big deal because that means that that entire stack can now move over and is consistent across different hardware.
- [11:32]
A
And this is important for the audience to understand. If you know Cuda is a language created by Nvidia many years ago and it's kind of creates lock in. Right? I don't know if it's stored as open source or not, but there is a bunch of lock into the Nvidia chipset if you use Cuda.
- [11:51]
C
Yeah, it's definitely lock in, but it's not specific to Nvidia. AMD has their open source software stack called rocm. We can debate whether it's very good or not compared to Nvidia's, but it's completely proprietary to AMD chips and so it's open source. But that doesn't actually help you because the vendor doesn't want their stuff to run on somebody else's chips. And the other big problem with all this stuff is that it's not actually that good. I mean, if you go look at this, Cuda for example, is the shining star of system software for GPUs, but it's 20 years old. Like the entire world has changed five times in that time. Right. And so really this stuff isn't really designed for the modern systems, not designed for Genai. It's all like C and it's not Python. And so it's like it's coming from a different world. And so what we're doing is we're investing in really rebooting that and then we bring a lot of benefits, both from technology, but also big open source community and bringing people together and catalyzing new use cases and all this kind of stuff. That's, that's why it's fun when you're
- [12:48]
A
building on modular as opposed to using CUDA or the AMD stack, which is Rock M. Yeah. Can you abstract multiple hardwares on modular? In other words, could I have like an AMD and Nvidia and an Apple hardware in some sort of cluster and
- [13:03]
C
all talking to each other.
- [13:04]
A
Yeah, all talking to. And modular being the layer above it, or is it just, hey, use modular and then you could swap it out one for one type situation.
- [13:12]
C
Yeah, so you can totally run modular on all three of those and so you can build heterogeneous systems. And so you get soft hardware with different kinds that are all talking to each other. This is actually super powerful if you look at what Nvidia announced last week. So they announced their Vera Rubin Grok platform that they're coming out with this fall. Vera is a cpu. Groq is a specialized asic, a custom accelerator for AI. And then of course, they have their
- [13:38]
A
GPU for doing inference, the Groq pieces for doing inference. Your query goes in.
- [13:43]
C
Yeah, yeah, exactly. And so the key thing about today's compute, but also very much more, the future of COMPUTE is you get these heterogeneous systems where, where you have different architectures that are all talking to each other and you don't want to have to rewrite all your code or your model or your ecosystem every time you want to try something. Right. So being able to scale across that is a huge benefit. It's also to your point about lock in, it gives enterprises choice. Right. And so a lot of people are running on Nvidia, which is fantastic. They have great flops, but they also want choice to be able to adopt other systems as well. And so even if you're staying on Nvidia, having an amazing experience and good performance and all the things that you want, good usability, reliability, all the good stuff you want on Nvidia is fantastic. But then you can scale off and you can go where, wherever your business takes you.
- [14:25]
A
Jake, how do you think about this at Gecko? Because certainly you're going to be facing this increasingly of where you put your investment. Now you're not building a large language model, I don't think, but you're probably building a lot of proprietary, you know, info. You certainly have a lot of proprietary important information. So you probably have to be careful where you put it.
- [14:45]
B
Yeah, 100%, yes. And I think that like the. There is an interesting data set that Gecko has amassed, like being now like, you know, we have like five or six hundred thousand assets inside of Cantilever, critical assets that we have gathered information on using robots and like information around those environments as well. And so there is an interesting data set that no one else does have. But one thing that, you know, you mentioned something, Chris, that was really astute and interesting. I like to double click on. And that's around, you know, hardware companies don't like to play with each other or like they have a hard time playing in the sandbox. Now what's interesting is in the, you know, in the kind of like an industrial or like, you know, let's talk about energy for a second. On the energy sector you have, you know, typically would have a lot of like point solutions on the hardware side, IoT sensor companies, you'd have maybe like robotics, drone companies or you know, you have a. And then you have kind of like a different kinds of infrastructure and software that also to feed into. You have, you know, digital officers, you have, you have you know, very like software focused, like.
- [15:49]
C
And I imagine you have a lot of standards that connect these components.
- [15:51]
B
That's correct. And one of the things that's hard for these companies is to understand like what's out there. And innovation teams do a really bad job. You know, whether it's like a, you know, nothing name like them specifically, but like all the top 10 oil and gas companies, right? And they will pick a solution, they will maybe invest into a company, know through ARM and then like they'll try to leverage and use that. But you, you'd end up with a robotics or a hardware company in particular is trying to figure out how do I, you know, make the big, you know, return for, for the investments that I'm getting. And okay, you have to like have some sort of software platform and then there. And you end up having like a dozen like software platforms that I all have to be logged into and a dozen different like point solutions. And that's not actually what the customer wants. They just want to be able to make, make a barrel, you know, for less amount or make more of it per refinery. And so like it ends up being like, hey, is there like something that can just bring solutions and help me be able to evaluate like what's actually valuable and not. And so, you know, you kind of have like this like almost, you know, this, this environment has been the reason why robotics hardware companies in particular have had a hard time and that you just don't get like the power law returns if you have 10 companies all trying to approach this. So you have to kind of like play this like Android game of like we'll either acquire everybody or like we're going to put this all under lattice and you know what I mean? And you can only have that happen a few times, you know.
- [17:11]
C
Well but so I mean when you're building your robots, I assume you have a bunch of AI running locally on the robot.
- [17:16]
B
Yeah.
- [17:16]
C
Where do you source your chips from? What does that look like? What are the pain points that you hit?
- [17:20]
B
We're processing at the edge, but we're also then just like a massive source of information data. And because that information data is relevant in terms of the localization in real time. But there's also like a post processing of the data sets that we're okay doing in the cloud. It doesn't have to be instantaneous. That's going to become more and more edge forward, especially as we do more of the, not just finding it, but fixing it side of this. And that's where obviously it's all headed.
- [17:44]
A
So the future is you have chips right now, you have some chips on the edge, I would assume like giving the robot, letting it make decisions in the field, but you can do the review later. Like if you're doing a giant navy ship, some aircraft car or something, you don't, they don't need that information in real time. But, but if they were going for maintenance, where hey, there's a problem and we want the robot to actually fix it, which I've never heard you talk about, but is the plan for you to have robots identify a crack or something and then go weld it as well, like a maintenance droid in Star wars or something?
- [18:21]
B
Yeah, you're exactly right. I mean the, the, I mean it's always kind of been for me there's a, you know, atoms to bits. We hear that talked about a lot. Yeah, but then there's actually a back to atoms and that's really where the impact comes from. And the, the large unlock too. I mean it's, you know, you want to be able to be the best in the world at understanding material science and the physics behind that. You know, as, as a, as a data collector of all this, you know, this critical infrastructure data and material, material data, physics data. And then it's just like, okay, now that I understand all these different use cases, materials, environments, metadata around like what kind of conditions, you know, how much, how much, you know, how humans, the air, like all that kind of information is very useful as, as we begins to lead towards taking action on how to fix and replace repair. I mean we want to live in a world where infrastructure that we rely on every single day is not down unless you absolutely need it or have and, and, and want it to be. We shouldn't have, you know, we shouldn't have maintenance cycles. That should not be a thing. Like there should just be like I'm down like every once in a while I fix the stuff I need to fix. But that's like you don't need like a outage twice a year. If you're like a thermal facility. You don't need like, you know, these turnarounds that occur multiple times at energy facility. I mean these are incredible amounts of money and environmentally it's horrible. It's horrible to stop stuff and then turn it back on. Just like when you accelerate in a car it's like bad. That's bad environmentally and just like, just so many reasons bad for the, you know, that's when you typically have like things breaks when you stop and start it. So that's, that's what we're, what we're building and in order to do that you also need to fix stuff, you know, after you find it. And so yeah, absolutely. And that's, that's even what we're doing on the shipbuilding side right now.
- [19:59]
A
Let's talk about the shortage in chips. Chris. Is there a shortage like we're hearing about? Is that marketing? Obviously you have a bunch of players doing custom Asics. You're having, you know, Elon talking about his fab. Amazon making their own. Google making their own. Facebook is going to make their own with Broadcom I believe so there's like Broadcom as this provider to help people build their own. But those people are obviously, I interviewed Jensen last week on All In. They're obviously trying to keep up. They're making this, you know, complete solution with multiple types of compute on one product. But tell me about the shortage, how real is it? And then about all these new competitors to Nvidia coming to market, when they'll land and what impact they'll have.
- [20:50]
C
Yeah, I mean if you snapshot today, just you go out and try to buy a hundred black hole nodes, it's very difficult. There's no supply out there. It's very stocked out for a long time. You have to get many of your commits. They're installing a tremendous amount of compute. But unless you're one of the biggest players building your own data centers, it's very hard to actually get access to the number of flops that a lot of people want. Meanwhile the AI is exploding the agents, all the different, the clause like all the different use cases are happening and so everybody wants choice. Now to your point. There's lots of chips out there. There's the Trainium, there's the Google TPUs. There's lots and lots and lots of chips from lots of different vendors, but people generally don't use them. If you're anthropic or something like that, you have one workload, you can put one workload and dedicated team to it, but other people generally haven't done that. Even amd it's pretty close architecturally to an Nvidia but you know, it doesn't have the adoption and the penetration that Nvidia is seeing. Um, lots of people are building, installing that. But again, it comes back to the software problem and people want, they want choice, but they also don't want to have two different software stacks. And so if you end up running with a big chunk on Nvidia like Jake, you guys are, you may want choice, but now are you struggling with like, okay, well I get choice, but I get two different stacks, I get two different sets of bugs and I, my software team has to be twice as big and how am I going to go manage that? And what if my models change? Right, so this is what we're trying to help with.
- [22:13]
A
Who's going to make the biggest dent in Nvidia's dominance? Is it going to be amd? Is it going to be Google or Amazon making these chips? It does seem like Amazon is making quite an investment in chips. Are they going to push the industry towards alternatives? Is it going to be this massive open air with AMD deal that seemed to have come the week after Sam announced the big open AI Nvidia deal and then like whatever, 10 days later does an AMD one? I think Jensen was a little perturbed, maybe be the right word by that. And so like, yeah, we have the opportunity to invest in OpenAI. We'll see if we take it. So there's a little bit of back and forth there, but if you had to rank 1, 2, 3. Chris, knowing what, you know, who's going to compete with Nvidia at scale in 2027, 2028, 2029?
- [23:02]
C
I think the biggest, the biggest player that most people are still not paying enough attention to is Google. Google's not an AI startup chip company. They have been building TPUs for seven generations. They're really good at it. They have better scale out than Nvidia does. And so in some dimensions they're way better already. And they're used at scale. Tremendous amount of flops that are available out there. And so they just need to decide what they're doing with their business. And I think Google has an opportunity to add a couple trillion dollars more to its market cap.
- [23:34]
A
So that's Google and the idea would be they will offer it through GCP as cloud computing resources or do you think they're actually getting TPUs?
- [23:43]
C
They're getting more ambitious and so they're selling it, I think through Fluid Stack. And so there's other, there's other vendors are now starting to get access to TPU's. I don't know their business strategy obviously, but if I were at Google I'd be advocating very strongly for, yeah, let's lean in and let's make a gigantic business. This is a huge opportunity for them and they've earned the right by being years ahead building the transformer and all these other things to really lean into that and make that happen. Now, I don't know how that factors into their cloud business. I mean they sell, rent them out through gcp.
- [24:13]
A
But I guess the question is like, why don't we hear about them? Yeah, you know, like why is that not. You're saying it's kind of like the sleeper. Why is it the sleeper?
- [24:21]
C
Yeah, I think they struggle with two things. One is GCP. So for a long time TPUs were GCP only and so that just kind of segmented out off the market share that they had into gcp. Now GCP is doing better. That's, so that's, that's a good thing. And now they're breaking past gcp. And so I think that's a big bold move for Google and I think that's huge. The other is that basically nobody can use them. If you're a big lab and you've got a team of 50 people to throw at it, you can, you can use it. But if you're not, then you can't run open source model, you can't run Kimik 2, you can't run standard models on these devices because two reasons, one of which is that Google has no community and so there's no developer community out there that's like using them and a hobbyist community, things like this. But also because Google itself is not investing and building into that. And so they like many of the different massive players in the space, they're building amazing things. I think Gemini is amazing, amazing, amazing. But it's all proprietary. They're not incentivized to actually share anything. They're going to open source their, their high performance GPU code or TPU code. And so like they're not actually set up to actually catalyze their own platform in that way. And this is one where CUDA is pretty amazing. This is the Nvidia software. They've leaned into the community, they've leaned into teaching people, they've leaned into universities, they've really leaned into getting the technology out there. And that's a huge advantage that Nvidia has going forward. So this is a huge opportunity I think for the whole industry to figure this out. But you have to Change the playbook a little bit. You can't just run the, the standard, the standard mode of operation they've been doing before and they have to decide is allowing the hardware to sing actually worth doing new things. And I think they'll probably say yes.
- [26:01]
A
The thing that's interesting about all this, Jake, is Anthropic, which is a major competitor to Gemini. Right. They are the big first, my understanding and Chris, you would know better but Anthropic is, I don't want to say all in on these tensors from these CPUs from Google, but they have a significant footprint. And then Google I believe is like a quiet shareholder in Anthropic as well. So in typical Silicon Valley fashion, no conflict, no interest. Like they're competing with Claude code with Claude cowork with the Google suite with Gemini's products. And Gemini suppose is going to have a really good, you know, they're going to, they're really focusing on writing code or code code copilots. But yeah, this like TPU thing is super interesting, I think. Who's number two? Craig, it's super funny comment on that.
- [26:55]
C
Yeah, yeah, it's super funny and super confusing to understand like how these companies are related to each other. But Google does everything. Like it makes LLMs, it makes autonomous cars, it makes search and ads, it makes obviously TPUs. It's crazy.
- [27:08]
A
And who's number two?
- [27:09]
C
Yeah, so number two, that's a big question, right? So I would go. It's either between amd, who is the. It should be number two in many ways because they've been working in the space for a long time, they've been competing or but it may actually be Amazon. And so Amazon AWS is obviously massive. They've been leaning into this. Anthropic, for example, also uses the AWS chip, it's called Trainium. And so they're on their third generation, incredible chip, very big scale. Again, they're not, it's not the first chip they're building and so they've done some iteration, they've optimized and gotten progress on it. I think again, just nobody's aware of it, nobody uses it. Software is just a completely different weird world. And they do have software, it does actually work if you configure it out. But it's just such a different universe. And again the customers, they're not going to open source anything. They're the biggest companies in the world, they're all competing with each other. And so the ecosystem is just tiny. And this is I think a Huge thing holding people back.
- [28:09]
A
So Google TPU number one, AMD number two, Amazon Trainium. And is it infer? I mean the branding's so like on
- [28:19]
C
the nose but it's, it's overly clever. Inferentia.
- [28:25]
A
Yeah, it's just terrible branding. We get it.
- [28:28]
B
It's.
- [28:30]
A
I mean did they just ask ChatGPT to name these two products? Like I feel like they just asked ChatGPT. What's a corny name for an inference thing? Inferentiania. It sounds like a spell from Harry Potter or something. So what do we think of the ban? I don't know if you saw this, Jake. The ban on chips in China and then the MicroStrategy CEO was using. I mean it's from Super Micro's co founder. Rather not MicroStrategy. Super Micro's co founder smuggling $2.5 billion worth of Nvidia chips to China through a middleman. And then they were doing like fake paperwork and using a hairdryer to take off the serial numbers and replace them or the model numbers. I mean, and then talking about it on social media. There was a video going around where he was talking about it. Any take on the brazen insanity of this, Jake? And people are talking about it in the group chats.
- [29:31]
B
Obviously I was actually going to bring bring this up as it relates to the Department of War. And, and like I actually think there's, it's, it's astounding to me that our budget on the, on the defense bill or the amount of budget going from trying to go from a billion, a trillion a year to 1.5. I'm actually surprised it's not way higher that we're not trying to like make it way higher. And why, why I'm saying this is, you know, it's like little insights like you just like you're talking about with, you know, with this story that indicate very clearly that this is an all out sprint. It's an all out. I don't know if you want to call it a cold Cold war. It is a cold war. That's what it is. And it's a war that's not exactly that cold either. It's being fought in a lot of different ways than we have never seen before. But this is an example of look, there is an unleashing and this stuff doesn't get, get the light of day unless the, you know, unless the, the government in, in China is also allowing it, you know, to be exposed there. I think it's kind of just like a look we can break the rules, we can go around things. And also like, like you interviewed with, with Jensen. I mean you know it's opening back up and there's a lot of folks that like are looking to buy this. I think there's like a look, there's. You cannot, you cannot regulate this stuff. I mean it is going to be you know just like, just like life will find a way and like just it's you know Nvidia chips will find a way is basically like the idea here. And, and the infrastructure is an all out race. It is a matter of national security. Energy is a matter of national security. These like sectors and industries, wars are being fought like around these assets in particular. So I think what I'm.
- [31:06]
A
We've fought over oil for.
- [31:07]
B
Oh my gosh. Yeah. I mean like now we're fighting over
- [31:09]
A
the chips and the oil. Now we have two things to.
- [31:12]
B
We got a. In the Middle east right now. Like we, we, we in UAE in particular. We've like I can't comment on the sites that been hit but just to let you know like there is, this is all very real for us at Gecko.
- [31:24]
A
Well if you, yeah, if you do share it, the UAE government's like please don't share by getting hit. A little sensitive to it which I
- [31:32]
B
understand but what I'm saying from is basically just like there needs to be. Government needs to be getting way more involved both in the amount of, both in the amount of spend that they have on the on on ships themselves, the amount of independence they have on the compute. This is why I think Chris is probably right as it relates to Amazon over amd by the way. I just think there's going to be, you know there, there's, there just needs to be a much more aggressive amount of spend both from a deterrence but also as it relates to if you want to actually supercharge these companies and not stifle growth. We cannot over regulate here that we have in the past. We take too much of a historical perspective on regulating these sectors and industries and in reality all of this is very different. There are not a bunch of historical precedents for what we're seeing right now. And we need to be unleashing and unlocking and having more primes emerge, having more integrations between old primes and new ones that are trying to emerge and also just becoming way more, way more self determined on the government side as it relates to leveraging these tools. And I'd be interested as well. Jason, you should have someone on this podcast talking about how is AI being used in the government, how its code is being written or being evaluated using these tools. I mean there's test procedures. These are great applications. I'd love to hear from the horse's mouth as it relates to, you know, how, how much impact the tools that we're talking about here are being used with our most important, you know, apparently
- [33:05]
A
like when we had Emil Michael with that whole kerfluffle. Chris, on all in. I don't know if you saw that interview. Claude was, you know, like a leading provider and Claude is a leading provider to Palantir. So it's integrated. It's probably the top model right now or the one that's kind of taken a lead all of a sudden. Leapfrog ChatGPT so it's definitely being used. I didn't believe the reports that they would use it to pick the targets without a human in the loop. I mean it could evaluate targets, I guess. I think that would be a fine use of it, probably a good use. But you would definitely want a human in the loop there. But, but Chris, what are your thoughts on just broadly export controls and do we want to really. Are you on the SAC side? Hey, we want to have our standard be the global standard. So we have to let China, other people use it so W doesn't beat our standard globally like they did for 5G. Where do you, or should we just, you know, not give them the top ones, you know, and let them suffer with some previous ones?
- [34:07]
C
I'm not the expert on geopolitics here, but the thing I'll point out is that China's building their own chips. We know this, they're quite good. Yeah, yeah. And so there's a number of different groups there, they're increasing in sophistication. I kind of agree with Sachs that if you hold yourself back then it feels like a short term win, but long term you end up kind of losing and so I can see definite advantages for short term tactics.
- [34:32]
B
And Chris, how many customers do you have internationally and what does your customer base look like?
- [34:37]
C
Yeah, we're global and so we both have a big commercial base, but we also have a big open source community. And so with open source you don't know where any of your bits go. And so I'm sure there's folks in China that are playing with things and doing things. But honestly what I love is I love people coming together. I love the developers building new things. I love people and the ideas getting out there. And this is where again with AI, there's a, I think fair argument to make that AI moving faster is just good for everybody and so it does not know any walls. The Chinese models, the open source Chinese models are really good. The Kimi K2s and all this kind of stuff are actually quite good. And we're benefiting a tremendous amount. We as America are benefiting from their work on this stuff. And yeah, they also benefit from our work. And so we could either be defensive and try to hold things back or we could be aggressive and lean forward. And I think if you look at OpenAI and Anthropic for example, both American companies, they're leaning forward. You look at Nvidia, they're leaning forward, you look at Google leaning forward and I think this is the way that you make good progress. Now we could be that, you know, try to put up walls around everything and, and play defense again. Maybe if there's a short term strategy that you're trying to like make work that can work. But I think that leaning forward and driving, driving the status quo, driving the standards, being the platform, being the leaders is really what, what, what winning looks like.
- [35:58]
A
Yeah, and they're making bets. There's this concept term, I heard four little dragons. I don't know if Chris, you've heard that one before, but this is for more threads. Metax, Byron and Inflame who are doing GPUs locally. This is not Huawei. These are these little ones that are startups here. And that's typically how China does it. They create, underwrite, you know, pick, they don't pick the champion. They kind of let 20 champions bloom and then narrow it down to the two or three big winners and have this like more constrained version of capitalism with the goal of of course flooding the market with things that are, you know, under, under price. Right. Like we're seeing with cars right now. I think, yeah, I think that's the
- [36:46]
C
thing I respect about the Chinese system is it's very competitive, right? And so there's like the, the people are very competitive and they're, they're pushing hard and they're trying to make themselves better. They're learning, they're studying hard, they're working hard. But then the provinces within China are all competing against each other and so they're all trying to build and one up each other. A lot of this, this nature, a lot of this mindset is really what's driving the open source AI movement coming out of China. Right. And they're trying to one up each other, trying to do better things. And when I look back on what made America great. It was really about that capitalistic competition of companies trying to outdo the other and capture value and deliver value. And so I think that again, this is where progress comes from. Like we shouldn't be playing defense. We should be leaning into building a bigger, better, faster moving flywheel and be able to capitalize that and be able to drive that up into our products. And I think we're seeing that. And if you look at the American economy over the last year, plus it's been just on fire. And a lot of that's been because of the transformations people have been able to roll out because of the increased value being delivered. And yes, all the fabs getting built too. That doesn't hurt.
- [37:52]
A
So let's pivot a little bit here and talk about self driving. Chris, you were involved a little bit in the self driving space. What's your assessment of it today? I asked Jensen last week about the Nvidia stack. There's been a lot of talk about this. There's a really good podcast, the Road to Autonomy with the couple of gents every weekend. Just go through all the announcements. Obviously Uber and Nvidia have a big relationship. Nvidia's got a giant relationship now with I think 18 different OEM. Yeah, original, I guess they're the OEM to the car companies. Then you have, you know, people making their own software like Neuro who use the Nvidia stack, but then I guess are going head to head against the Nvidia driver, which would be like the Neuro or fsd. So your assessment of this now, Chris, and your history in it.
- [38:41]
C
So, I mean, I haven't worked in the space for nine years, but I led the Autopilot team at Tesla and helped convert them from hardware 1 to hardware 2. And a bunch of technology ecosystem improvements back then. Are you curious about who's going to win? Are you curious about what it looks like in this new future?
- [39:00]
A
Are you curious about, I guess your assessment?
- [39:02]
C
Because Waymo's obviously in the lead.
- [39:04]
A
Yeah, Waymo has the most miles, most rides. It works in a constrained environment. Then I guess you have Tesla going for this. I wouldn't say incredible vision.
- [39:14]
C
No, it's not a realistic player. They're not filing the permits to have actually autonomous cars and they still have humans in their car. But I think that in China.
- [39:24]
A
So that's the thing to watch is when they file for. Not a. Because right now they filed for ride sharing. Yeah. So they're doing. Yeah, like Uber, but they're Using the software under supervision. So you think that's the tell when they start filing to be autonomous, that's when you know it's going to get real.
- [39:46]
C
Yeah. So my, my understanding, which again I'm far from being an expert in any of this stuff, but my understanding is they have single digit number of cars that are doing anything and they're in one Geo area and Austin or something like that. And so it's a very small, small, small program. I don't know the time frame, but it does take months to get approval in a place like California. And so when they start applying for those permits and going, then you can see that as a lean forward of when they're going to scale. Yeah. Again, if you look at China, China has amazing humanoid robotics. Jake, I'd love your take on this. They have amazing autonomous cars that are pretty widely available as well. And so this is one where for a company like Tesla or for, for these other players, like the question is kind of where do they sit on the world stage? And I think this is a huge question. Jake, I'd love to know what you're seeing from the China robots world. And I see these videos on YouTube of the humanoid acrobatics almost. It's pretty crazy. Is it real?
- [40:44]
B
Yeah, yeah. I mean like it's demos can be written and can be executed and the kind of way that you saw, I guess I got the, you priced all the kung fu moves and the Olympics and those kinds of things.
- [40:56]
C
But what entire like on stage shows with like 30 robots and humans and robots all dancing together doing crazy stuff.
- [41:03]
B
What it shows off is like the, the actuation, it shows off the, the perception. It shows off, it shows off a lot of those things. The real tricky stuff is the dexterity. I think this is stuff that probably Elon talks about with the Optimus 3. I mean Jason, you've probably seen the Optimus 3.
- [41:19]
A
I got to see it early. It's super, super incredible.
- [41:22]
B
Yeah, I think like, you know, the, the real, the real questions are going to be, you know, who's going to buy it? What sorts of like ROI is there going to be? We've tried out from Europe, from China and from the US the best mobile platforms. So the walking dog types, these sorts of mobile platforms. What I have been able to determine is that there's very little value that the data sets and the actions that can be taken from the mobile robots actually have. And I've looked and tried my hardest because it's so cool, right? It's like so cool to be an implementer of that on the canilever stack. But what I've determined is that there is just too little ROI for the amount of effort that it takes to actually get these systems to be valuable and useful.
- [42:11]
C
I'm curious on that because it's also, it's not just about having a general robot, which you might want for a consumer product. It's also about being efficient, good reliability. What does the repair look like? More complexity is actually just worse for operating at scale in an industrial setting, I would think.
- [42:26]
A
Jake, here's the figure robot with, I believe this is Marc Benioff messing with it. It's sorting packages, just putting the barcode face down, I guess, or the shipping address face down. And then it's got Benioff throwing packages back at it and it's sorting it. You see something like this, Jake, I mean, what is this? You know, we were talking about demos, you know, being faked or you know, perfect conditions or even self driving demos. Chris, in your day, you know, five years ago, a lot of times the self driving demos were very canned, I would say, I wouldn't say, you know, they, they weren't fake, but they were produced in a way for a limited, you know, space, you know, so they, you know, didn't have the, the edge cases, let's say. But that video right there, that looks like a general purpose robot doing a general purpose task that humans are doing right now.
- [43:24]
B
Yeah.
- [43:24]
A
For 20 bucks an hour, 24 bucks an hour. And that robot, you know, if it doesn't break down and it has decent fidelity, could be sorting packages 24 hours a day at a factory. That's hundreds of thousands of jobs. Yeah, it's probably millions of jobs.
- [43:40]
B
Yeah. Regardless of if that robot was teleoperated, if the more complex task of flipping over a more complex, more, more nervous need for a more dexterous robot is a fundamental shift in the way that economies can be built and economies can run and unfairly advantage that companies that adopt this sort of technology early and natively will have over the next three to four to five years. Like I, I, I've talked about this publicly. I, I think like there is no better decade for private equity than this decade that we're in right now. And the reason was because your ability to take and buy especially capital intensive and commoditized infrastructure assets, you know, like, you know, waste, waste, turn energy, you know, from waste into energy or you know, our water treatment facilities or power plants that are old or these sorts of, these sorts of investments over the Next decade are going to be incredibly high return because you can be able to, you can be able to turn these assets into more and more autonomous assets. You can be able to self insure more and more. So reducing the amount of risk taking on more risk itself with tools and tech technologies where you can understand the health of these assets because the management
- [44:59]
A
team's not going to do that. But this might be what we're seeing with these open AI and, and you see this happen, private equity deals they're doing and they're going to legacy businesses and then AI them.
- [45:10]
B
You see like Thrive holdings, right, like with Josh Kushner. So he's buying accounting companies and, and injecting AI models and stuff. Like we're still seeing like how much return there is with that strategy. But you can see the strategy being implemented. I think it should be way more money. But basically he's doing that for info, for manufacturing because he's seen the impact that millions of robots have for his distribution centers and helping two day shipping. You know, it's, it's, it's a robotics defined operating, operating platforms that gives us such an advantage. And so I think it's like, look, it could be humanoids, they'll be involved. But you're going to see extremely high growth of specialized robotics that are very specifically focused on these critical tasks, these critical data sets that help to unfairly run large pieces of infrastructure over the next decade. And then I'm going to think you're going to see a consolidation of which energy, which manufacturing, which, you know, mining companies are the dominant ones because you just have unfair, you know, unfair P L as it relates to how you run and manage these things.
- [46:08]
C
Well, and J.C. also knows and has talked about quite a lot as the cost of software goes to zero. It's not just the hardware side. Right. So the transformations are going to accelerate.
- [46:19]
A
Here's the story. Hold on, let me just share this. Here's your New York Times story. So Thrive, and this is Josh Kushner, Jared Kushner's brother. Jared's making peace in the Middle east right now, or attempting to his brother Josh, I guess. Sam Altman, OpenAI took a stake in this company at Thrive holdings, which then has this roll up where they've bought I think 20 or 50 accounting firms. And so it's called Crete C R E T E Professional alliance and the IT service provider Shield Technology Partners. And so those are the two that they've put together. They've committed 500 million to CREIT, which the trade publication Accounting Today described this year as one of the fastest growing accounting firms in the United States. So the idea here, buy up all the accounting firms, a lot of those already have offshore. And that's always my little tell to figure this out. The step before AI automation was move the work to the Philippines, move the work to India. You know, the top 1 to 5% of those markets, knowledge workers, are better, more consistent than the average worker in American America, which is to say the averages are not the same. But the elite in India, the elite in Philippines are going to be better than the average in America. And they're going to cost, in my experience, 4, 5, 6, 7, 8 bucks an hour versus 40, 50, 60 bucks an hour in the United States. So that's an interesting kind of wrinkle here.
- [48:01]
C
Yeah, well if you look at that analogy, I think that analogy may also be comparable to the AI replacement cycle, right? Where the lower value jobs get replaced, the higher value jobs don't for the same reason that people have been out offshoring and getting lower cost labor and other GEOs. But there's still an incredible amount of engineering and incredible amount of talent in the US and so a lot of people are betting big on American talent I think will transform and will continue to push the world forward and it will affect a lot of jobs and people upskill. The question is, what are they upskilling into? And where does the value accrue? Where, where does that get captured?
- [48:38]
A
Well, yeah, and you were sort of talking about like commodity work and being a cpa, an accountant or a lawyer was considered professional services, not commodity. And here we are in 2026 and we're like, well obviously that's a commodity job. Like, you know, yeah, the bottom 50% of those jobs are chores that machines can do easily.
- [49:02]
B
Maybe the top half, like we're still gonna, we're gonna have to see like, is that strategy pan out like this year, next year? It's still like one of those things. Also as it relates to manufacturing, there is just like, listen, like there are single points of failure in so many supply chains, whether it's turbines and generators and you know, nuclear reactor parts and like all these like different like parts and forges in the US ecosystem, the re, the re industrialization that's happening right now, one thing that's important to understand is that there's also, when you, when you see technology companies, you know, saying they're taking on manufacturing in the kind of ways that you see or you see empty materials, you know, taking on a mine and making it More effective and efficient. You have to understand these sectors have not changed for the most part in the past 40, 50, 60 years. And so there's actually not a lot. Look, technology is, there's low hanging fruit, there's low hanging fruit. Like there is just like, hey look, we're still using clipboards guys. Or like we just like don't know how to hire because like no one wants to come work in these sectors and industries anymore. Like there are 200 folks who graduate with mining degrees every single year in the US like look, let's just make this sexy and just like first and foremost just try to get, you know, the low hanging fruit to be wins. And if you want to masquerade them as technology advancements and improvements, great. I mean I guess that's like fine. I don't love it from a, I don't love it as a, as a trying to tell the truth in the matter and understanding the truth, but I also don't, I also really want there to be a resurgence into the sector and technology advancements are going to come. You're going to see contracts being won by companies that are able to show like, look, we're using automated welding, automated inspecting, we're using humanoid here, here and here. But that's, that's like it's going to happen. You need to be forward looking and you need to be as native as you can possibly be. But I think it's just important to understand it's like there is like the same sort of principles exist with accounting. There's the same sort of principles exist, you know, in these, in these like sectors that you're seeing this innovation like flooding into. But you know, in just reality understand there is a lot of low hanging fruit. There's a lot of just like get smart people into the sector, make it cool. And then I think, you know, this, these, the leaders and the CEOs. I, it just boggles my mind. And Jason and Chris, like the best recruiting tool in the world is just like get robots everywhere, get AI everywhere. Like there is like that's so exciting. If you're like in energy, you're manufacturing or mining, whatever it is, it's like make it cool and like the folks in these communities will run towards these jobs, otherwise they should run somewhere else. Right? And so you want to re, you want to re, you know, you want to like reinvigorate, you know, the, the working class. You want to reinvigorate these towns and these, these, you know, these places around the US where you just have so Much dissipation, people moving away from these cities and towns. Like, that's beautiful. That's America. We want there to be research.
- [51:45]
C
It's super, super hard to do that. I mean, Jake, I'll give you one example from my, my little world is I'm trying to get software engineers who are one of the biggest groups of people that are under threat from this AI tooling explosion that's going on. And all these folks, depending on who they are, have like these, this existential dread of what does it mean for my job? Or if you're a college kid, you're entering the workforce, am I going to get a job? What are people hiring? What does it all mean? But there's this amazing platform out there of GPUs and there's this amazing thing happening. AI right, just to your point, it should be obvious. Everybody, it's happening, right? But what I see is that some people do adapt and adopt and jump in and upskill and then differentiate themselves. And some people do that. But a lot of people are just kind of following the university playbook. The university playbook is a few years ago.
- [52:34]
B
You know, what is happening like the most, by the way, Chris, it's hap. It's happening in the countries that are most at threat economically and the hungriest. Yeah, yeah, right, exactly. It's in the Middle East, Jason, you know this well, whether it's in Saudi or it's in the UAE or it's in Israel, there is threat to the economy that is causing these leaders to be big, bold as it relates to how they are adopting technology, trying to be 100%, trying to figure out like we have to be, we have to be technology first. You have to be robotics native, AI native to drive and create a technology company out of these energy companies. And, and they are doing so, you know, because they understand that survival is the most important thing. Whereas in the US we've been very accustomed to peacetime, very accustomed to, you know, we all, we all get this percentage of the energy market and you know, we're all, you know, we have utilities that have 30 year power purchase agreements. I don't, I could just pass through my losses down to, you know, to the, the, the single mom is just trying to make it every, you know, every week, week to week, paycheck to paycheck. And you know, this is, this is why like we're spending time out in these regions. But it's my goodness, like the US needs to wake up as it relates to these, these sectors. We need to have a, you know, a re. Industrialization and an excitement in these, in these like fields because it is going to be, is going to hit us hard and it's going to hit us fast as it relates to how vulnerable we are. Because there's just not enough people in these sectors and using and trying to adopt big, bold visions from these CEOs and, and these CEOs, most of our finance, you know, were CFOs before that. And so their minds are not focused on vision and ambition. They're focused on how am I going to get an extra couple cents, you know, up tick in my, my, my ticker today. And my goodness, that is not what America needs. We need a complete change in that approach. And so that's the CEOs that I look for. You know, we work more looking to
- [54:21]
A
do big deals and yeah, very clear we're going to need a lot more tradespeople. We have, even before the AI boom, needed more plumbers, electricians, teachers, nurses, doctors. Like, there's a whole group of jobs that have been desperate and literally bringing in people, you know, essentially importing talent, you know, whether it's Jamaica, you know, or India or the Philippines, to find nurses, to find doctors and underwriting.
- [54:51]
C
This is the funny thing. Like, so 10 years ago, people in AI were saying that all radiologists were going to be obsolete.
- [54:57]
B
Yeah.
- [54:58]
A
How'd that go?
- [54:58]
C
Definitely don't go into radiology because you'll never get a job. Right. Here we are 10 years later and there's not enough radiologists. AI is definitely being used, but we have so much demand. And I think this is the thing that everybody kind of forgets about is that the market, the economy, the capabilities are expanding. As you get more creation, you get more products, you get more customer experiences, you get more opportunity. Like this virtuous flywheel is what has made our ecosystem in our community, in our world great. And so AI is an accelerant.
- [55:28]
B
Yeah. And I think that CEOs like me and you, Chris, like, we gotta be judged more on how quickly can net new people have no idea about the sector use, use the technology that we're building for them. I mean, this is what it's all about. I mean, 100%, how fast? You know, for me, it's just like, can I get a McDonald's employee or home Depot employee to be, you know, an expert in and weld inspections and evaluations within three months?
- [55:49]
A
The answer is yes. Maybe not all of them have the motivation.
- [55:52]
B
Yeah.
- [55:53]
A
But when faced, hate to be, you know, based here but when faced without a job and faced with having to put food on the table, humans, not necessarily the entitled humans we've had in America, who have had massive abundance and an incredibly low unemployment rate and incredible wages when compared to the globe.
- [56:16]
B
Yeah.
- [56:16]
A
Now they. People might take offense to my statement and cut out the. When compared to the rest of the planet and the rest of humanity, we are massively entitled with the job opportunities. If an Uber or, you know, being a greeter at Walmart existed in the Philippines or in India or other places with high unemployment, the Middle East, China, if those jobs actually exist, people would be racing to them.
- [56:39]
B
Yeah.
- [56:39]
A
And be thrilled with them.
- [56:40]
B
Yeah.
- [56:41]
A
Let alone a trades craft where. Well, you know, if you need to put food on the table, you're making $18 an hour in fast food or $25 an hour being a gig worker.
- [56:50]
B
Yeah.
- [56:51]
A
I think actually if you presented the path, here's the training, here's what it costs to get trained. And on the other side is $45 an hour, you're going to make twice as much money and benefits, Which I'm guessing McDonald's doesn't pay benefits. And obviously you get a token amount of benefits as a contribution.
- [57:07]
C
Are there enough electricians in the us?
- [57:09]
A
Not even close.
- [57:10]
C
Right. Plumbers are the big one. Yeah.
- [57:12]
A
It's an unbelievable amount of opportunism. Yet we are incredibly pessimistic. And I also find it, I don't know if you guys find it offensive, but when like rich people or powerful people, successful people are like, yeah, but poor people, they're just not capable of learning a new skill. And you're like, are you not paying attention to humanity? Like, humanity learns new skills as a group. Like, nobody knew how to use a laptop, a phone. Nobody knew how to use any of these tools until they might be laggards in terms of the technology adoption cycle. It's not an insult.
- [57:47]
C
This is one of the reasons I see everything accelerating, because it feels like things are accelerating. I think we can all feel that right now. And it used to be, you go back five, 10 years ago, it was YouTube and you could learn anything off YouTube. And before there was, you could not learn how to fix your car or whatever it is. But now with YouTube, you could do that. Today we have AI and now everybody has assistance. You have infinite knowledge, you have problem solving, you have reasoning, you have all kinds of different capabilities that people haven't had before. And so we talk about the open clause. People get personal assistance. And JCal, you've had a few assistants I suspect. And so you know what that's like. But for somebody that's never had one, having to deal with the inconsequential details of everyday life is a huge drag that prevents you from investing and upskilling and doing whatever it is that you want to do with your. Your actual life.
- [58:38]
B
I think, too, like, we, we also, like, look like you. And you guys all know, like, you've talked to so many people. There is such patriotism in this country. Like, real patriotism about, like, really, really proud. I mean, you might not agree with every decision that's being made by the administration, whichever administration it is. There's patriotic Americans. And for the most part, I grew up in a small little Maryland town. Like, people don't typically leave and they want the ability to be able to, you know, to, you know, to be able to, you know, not have to do the same job that, you know, their, their, like, father or their mother, like, did. And if they did, they. That, that, that's great too. But we, we have to understand that, like, this, this, this hillbilly allergy that J.D. vance, like, ran on and that was very successful, that is hitting on a very important part of America, especially in the Midwest. This is where Pittsburgh is very interesting vantage point. People don't want to leave. They want to be able to be. They want to be able to be in technology companies to invest in them, to be able to bring them along for people who are building products, to build products for them that are oriented towards how. How much, like, how easy for me to use, how, like, foreign is this.
- [59:45]
A
It's your. So this is such an opportunity, Jake, is, I think, what you're framing here. And if you were to just think for a moment, like from first principles, as all these jobs are getting cut, you know, square, obviously, and people are saying, hey, you know, maybe they're using AI to make a bloated job cut or whatever. Okay, fair enough. But there are definitely AI cuts happening in customer support, in product management, and, you know, maybe just not hiring more developers because the current developers are getting 20% faster a year. Okay. Even if it's just 10, 20% faster a year. If you had 100 developers, you needed 10 more.
- [60:18]
B
Yeah.
- [60:19]
A
Now you don't need 10 more. You just need to keep paying more and making sure your hundred developers are happy.
- [60:24]
B
That's right.
- [60:25]
A
But what an amazing opportunity for professional development. And if these big companies don't see that opportunity, well, then they're missing what I think would be a massive strategic advantage.
- [60:35]
B
It's not Just as soon as companies. Yeah, I'm sorry to cut you off, but yeah, you're hitting it. I'm getting very passionate about this because like the same, like if we have just a few companies that are all benefiting from the big wave we have right now, which by the way is the natural progression of freak of like these like free markets, this capitalism, it's like it's going to, it's going to make it, it's going to make these like larger companies larger and larger and larger. I'm telling you there is going to be a degrading of democracy that'll happen when that as that continues to happen. Just like with the farming industry, you had small local farms that would create food for the local villages, the local cities. And then as free markets and capitalism went on, you began to get these, these large like Tyson facilities that were making you know, large things and they were putting a bunch of chemicals into our food and all this kind of stuff happened. Let me humor. And so we, I really, really what we need to do is have. There needs to be these like robots, this art AI. There needs to be a, you know, the governments and, and capitalism. There needs to be a. How can we create these small energy facilities that are creating small data centers instead of these big reaction.
- [61:48]
A
Chris.
- [61:49]
C
Yeah, so Jake, Jake, Jake, I hear you, I see what you're talking about. And yes, the small farmers having a bad day. How do you square that with the power that every person has in AI? People can build things like they've never been able to build before. They could become a software developer. They could build apps for the App store. They can do anything. Right. They just have to, they could be
- [62:10]
A
a great writer when they were a terrible writer. They could be a great handy man or whatever. The non gender version of that is like 90%. What is it called? Handy person. It sounds so weird. That's, I'm sorry, I'm just gonna say handyman and I mean human when I say it. Handy human. I'm just taking out the HU for efficiency. It's a non gender term.
- [62:33]
C
Don't cancel that.
- [62:34]
B
Yeah, this is where like there is a shift that you're all seeing right now as it relates to Adams. And I think that is where you know there is, there is a, there's incredible natural resources around the US There is, there is like these, you know, these like US steel towns. These like the US Manufacturing. These, these, these are the sorts of, like these are sorts of. And you know, when I, when I, when I talk, when I talk about like, you know, why we're in Pittsburgh? Like I want to be in Pittsburgh because you're not supposed to build a billion dollar, you know, multi billion dollar technology company in Pittsburgh. You're supposed to build in California. And, and I think that's like, that's what I'm trying to, you know, that that's a, that's a side quest, I guess, of like this mission of Gecko. It's like I want to be able to show that you can build these kinds of companies, you know, in, in, you know, strategically, like strategically and statistically bad area decisions of like where to
- [63:30]
A
look, places that are not growing. And I think you can find value in those areas. Like it's happening in Austin where a group of us saw in Austin, hey, wait a second, you can own a home for 500k, 250k, 750k. You could save 14% on your taxes from California. Your thousand employees could be happy. They could be happy and be able to hire a nanny. Like a white collar couple with dual income in San Francisco can't hire a nanny because the nannies in San Francisco, I hate to blow people's brains who are not from the Bay Area. They're getting paid $100,000 a year to be a nanny, $125,000 a year to be a nanny, to be a housekeeper. I know this sounds insane to people, but 40, $50 an hour for a domestic staff is the standard in the Bay Area because the cost of living is so high, it's no dig. But then when you could hire a nanny for 50k for $25 an hour or something like that in the in and make the same salary in Austin or in Philly or Detroit. Like this is why these cities, I think I agree, are going to have a huge comeback. I'd love to invest in a company. I just had a company idea. So if anybody's listening, who wants to start a company, a company that was a bridge between gig workers, you know, and the lowest end of knowledge workers, you know, fast food workers and then doing handyman services. And by the way, handyman is handy human. The hu is silent. I just learned the hu is silent. Imagine just that, handyman, people who can fix things around the house. Then handyman to junior plumber, electrician, H Vac. Just a company that just does that from home for but $500 a month in tuition, $6,000 a year in tuition. You work at home, you do zooms. And then maybe you could come and do an intensive for a week or Two, if you wanted to, and pay an extra fee. Just that job, you could go from 20 bucks an hour to 50 overnight. Overnight. Make somebody build that company. I will incubate it.
- [65:41]
C
As the resident software guy, sometimes I joke. I just turn zeros into ones and ones into zeros. That's all I do. I don't get to build cool things like Jake does. But in this world, it turns out there's a huge amount of opportunity. And it turns out that building software is faster, cheaper, more accessible than it ever has been. Anybody can build AI and GPUs these days, and GPU software. We're seeing people that are taking Python and a lot of people program Python. Of course, they're using our stuff to move it to this Mojo language that we're helping build. It's open source. And you can just take AI tools and say, hey, move this stuff to Mojo, and it goes thousands of times faster. Right. And so what this means is that there's a huge amount of opportunity out there in the world. And what we need is we need people with imaginations. We need people that are willing to, like, put themselves out there. We need people that are willing to try. And I think this is where, again, it doesn't matter if you're in Pittsburgh or if you're in the Bay Area or wherever you are. It's really about. It's really about your mental mindset. It's about what you want to achieve, like your ambition and drive. And I think that if we can catalyze that, then everything else flows from that.
- [66:45]
B
Yeah.
- [66:45]
A
Here's something interesting. Breaking news. I'll just get your final comments here as we wrap. Brett Adcock from Figure Robotics, which we just talked about earlier, making Humanoid Robotics the ones sorting the packages. Today I'm excited to introduce Hark, a new artificial intelligence lab building the most advanced personal intelligence in the world. We've been in stealth for eight months assembling one of the greatest AI and hardware teams on the planet. I want to explain why I started Hank, blah, blah, blah, spent the last three years working on the hardest AI challenge. AI humanoid body digital side. I've been using the existing LLM chatbots and have to say they feel incredibly dumb to me. Whoa. Dis shot across the bow and AGI quote in the Limit should feel like a sci fi movie. It should be able to listen and talk. It should have persistent memory and be highly personalized. It should see and touch the world. But we're far from this. Today we're crafting a new interface to AGI intelligence that lets you offload your mental workload into a system that begins to think like you and sometimes ahead of you build the world's most advanced personal intelligence paired with next generation hardware. Hark.com this is breaking news folks. This just happened. I guess this is a new entrance into Jake, the space going from hardware to LLM as opposed to LLM and then adding hardware. Your reaction? Did you know about this project? Have you, has it been back channel about it, Jake? No, it's just breaking news to you as well.
- [68:15]
B
I don't, I mean like I, I don't really follow figure all that closely.
- [68:20]
A
Yeah.
- [68:20]
B
But yeah, I think, I think it's yeah there's a rising amount of attention like in this particular tech stacks. I don't have too many comments about it. I'm actually kind of interested in Chris's thoughts on it.
- [68:33]
A
Yeah, Chris, any thoughts here of.
- [68:35]
C
I know nothing. Right. So this is just me as a tech bro commenting. The curious thing to me is not just the technology but the product. Like how do you actually bring something full pro, full featured, full product into the market? Something people weren't willing to pay for.
- [68:52]
B
Right.
- [68:52]
C
And I think Jake, like you've shown, I mean you've learned it's really hard to do that.
- [68:56]
B
Right?
- [68:57]
C
Getting some of the technology components, there is one thing but getting it to scale, shipping the thing, having people actually use it and then breaking it, breaking it in the field and then trying to fix it, like all this stuff's really hard.
- [69:08]
B
Yeah, I think like the reason I don't pay attention to figure is like figures philosophy is the exact opposite of our philosophy. And I just, yeah, I just, I just don't, I don't.
- [69:17]
A
They're. Yeah. General purpose robot versus specific ones to
- [69:20]
B
the task built in lab, then like try to use in the environment. So I think like, like I think I'm excited for, for the attention to the sector. I'm excited for the potential for the products. I'm really excited for you know, the, the, the amounts of people that are hitting us up in terms of just like applying for jobs because robotics is like cool and it's a sector doesn't create. So it's like that's great. And I think like one, one thing I was going to say on the last like part Jason, just to finish up my thoughts is like democratizing the amount of ability to create prosperity across the country and, and also like you know, just like provinces are, are competing against each other in China, states should be competing against each other kind of like they were like a hundred years ago, fifty years ago when, you know, when we first created the United States.
- [70:05]
C
They should be competing with talent pools, not with tax rates.
- [70:08]
B
Exactly. And. Exactly. And, and then like we, we, we like the whole game of politics, like, just doesn't reinforce force this like, fierce amount of like competition amongst the districts and states. But I think like the, the vision of the future we have to like be really excited about and striving for helping to create as well is, is like, I want my 9 year old to be able to create a lawn service business and use robots or for
- [70:31]
A
sure, you know, we call that job shepherd, shepherd for robots. I have a cynical take on it. I'm guessing figure. No, no, just on the figure, figure stuff. Yeah, I mean, I think the most generous take is what he said. Hey, we tried to use these alarms. It doesn't work for our robots. Okay, fair enough. You make your own cynical take, which would be, hey, the valuations of OpenAI, the valuation of Claude, you know, getting to 800 billion, 400 billion. I'm sure Claude's next valuation will go from 350 billion to 800 and be neck and neck with OpenAI's. I believe they'll be probably be similarly valued. So if they're going to be worth 900 billion and XAI as part of SpaceX is worth 2 trillion, I need to be in that group. Not in the robotics group, you know, competing with just Optimus Head to head and the Chinese companies. I want to be part of the, the bigger sack. And part of that is, you know, OpenAI and Claude will get into robotics. So we need to get into their business now and just, you know, kind of put us in that competitive set. A lot of times founders and CEOs and boards will be like, hey, wait a second, let's rewrite the rules here. Let's be in that set. Actually, Jake, for you, with your robotics company, if you said, hey, we have our own proprietary LLM just for maintenance, it would increase your valuation. It would be another asset. I would look at it as an investor or a board member or just a market participant as, oh, wow, they're going to power other robots with their underlying technology. Might be something for you to think about.
- [72:06]
B
Yeah. And also, Jason, there's also other kinds of data sets too, as relates to just like we are in all these sectors collecting so much information about and we're basically in some ways like you can think about like the data sets of like the YouTube for these industrial sectors is a very valuable amount of data sets we would train on. I think it's also a sign that manufacturing is hard. I think it's also a sign there too.
- [72:29]
C
Yeah. My take on this is that, you know, when you're building a business, when you're building anything of merit, you have to decide what your contribution is. Trying to play other people's playbooks, trying to be the same thing that everybody has already done, trying to, trying to just like, you know, appeal to the investor isn't really actually a great strategy because you get too diffused, you don't achieve things, you can't differentiate. You become mediocre at a bunch of different things instead of exceptional at one.
- [72:52]
B
Right.
- [72:53]
C
And so I think this is the challenge.
- [72:55]
A
The North Star should be the customers and the value providing to them. And then what happens in a hot market and in an escalating market is people will do unnatural acts.
- [73:03]
B
And I think.
- [73:04]
C
And don't get lost playing games. Right. I mean, I think that the creating core value that people are willing to pay for is the, is the recipe. Right. I think games, you can get caught up in it.
- [73:13]
B
Yeah. And then Chris. But I think the thing that's like, for the, for the listeners as well, Jason, is like, well, we are, we are a. We like to mimic things like, from a very Gerardian perspective. Like, we like to mimic things we see. We want to mimic as founders. Like, I want to get as much validation as I possibly can. Look, this person is achieving like this valuation. I need to achieve this valuation. I think the lessons that Chris is talking about are only lessons that come from scars from failures, from understanding the fallacy of pursuing the wrong things and the potential wisdom or lack thereof of folks that seem to have.
- [73:49]
C
And what a success.
- [73:50]
A
Really important.
- [73:52]
B
There's no, there's no replacement for the grit and hard work that, you know, Chris, you know, the people like Chris has, has, you know, his wisdom right now is like, it's so important to just really rock and like success will come if you adopt those principles.
- [74:04]
C
Yeah. Success is not the press release. Success is the product. The press release needs to compound and propel the product. But if you get confused and you just want to have announcement, announcement, announcement, and there's nothing of substance behind it, then you're not winning, even though it feels like it in the moment.
- [74:20]
A
Bingo. Short term gain versus long term gain. And the long term gain takes a lot of pain. Another this Week in AI is in the books. ThisWeekinAI AI to sign up for the newsletter and all the links. Thank you to Jake. Thank you to Chris. They're both hiring go to their websites and hardest position to fill. Jake right now for you who do you need?
- [74:43]
B
Robotics engineering.
- [74:44]
A
Perfect robotics engineers if you want to get in on the ground floor of a company that's going to go 100x from here. I said that not him. That's just my professional estimation and same thing with Chris. What are you hiring for Chris?
- [74:54]
C
Cloud platform, GPU programmers, AI software professionals.
- [75:00]
A
Little tip here. Sometimes the CEO or founder gets their first name@company domain.com or AI. So just you know and a lot of times if you build something and share what you built as opposed to a resume and begging for a job and promising things sometimes you build something dope, you send it to a CEO sometimes they'll click the link and see what you built. Just a little professional tip there from your boy jcal. We'll see you all next time. Bye bye.