This Week in Startups - Uber, Lucid & Nuro team up on robotaxis by 2026 | E2173

Episode Date: September 3, 2025

Today’s show:Jason and Alex are joined by Nuro’s Dave Ferguson to unpack the Lucid–Uber–Nuro deal aiming to launch fully driverless Gravity SUVs on Uber by 2026. They dig into why off-the-shelf sensors and Nvidia’s Thor chip drive costs down, how mapping 150 cities improves safety, and why lidar still matters for night and edge cases. Plus: lessons from Cruise’s collapse, how regulators think about safety multiples, and why autonomy could finally make rides cheaper than owning a car.Timestamps:(00:00) Show intro…(06:23) Uber x Lucid x Nuro: why this 2026 robotaxi deal matters(09:57) Lemon.io - Get 15% off your first 4 weeks of developer time at https://Lemon.io/twist(11:13) What the self-driving “cost stack” really looks like in 2025(14:08) Road-mapping at scale: Nuro’s 150-city data collection push(15:38) Can AVs handle snow & rain? Weather rollout and Uber Black positioning(20:05) Northwest Registered Agent - Form your entire business identity in just 10 clicks and 10 minutes. Get more privacy, more options, and more done—visit https://www.northwestregisteredagent.com/twist today!(21:07) Show Continues…(24:24) Probabilistic vs. deterministic driving models — and the DARPA roots(30:08) Stripe Startups - Stripe Startups offers early-stage, venture-backed startups access to Stripe fee credits and more. Apply today on stripe.com/startups.(31:08) Show Continues…(35:38) Post-Cruise reality check: transparency, regulators & trustSubscribe to the TWiST500 newsletter: https://ticker.thisweekinstartups.comCheck out the TWIST500: https://www.twist500.comSubscribe to This Week in Startups on Apple: https://rb.gy/v19fcpFollow Lon:X: https://x.com/lonsFollow Alex:X: https://x.com/alexLinkedIn: ⁠https://www.linkedin.com/in/alexwilhelmFollow Jason:X: https://twitter.com/JasonLinkedIn: https://www.linkedin.com/in/jasoncalacanisThank you to our partners:(09:57) Lemon.io - Get 15% off your first 4 weeks of developer time at https://Lemon.io/twist(20:05) Northwest Registered Agent - Form your entire business identity in just 10 clicks and 10 minutes. Get more privacy, more options, and more done—visit https://www.northwestregisteredagent.com/twist today!(30:08) Stripe Startups - Stripe Startups offers early-stage, venture-backed startups access to Stripe fee credits and more. Apply today on stripe.com/startups.Great TWIST interviews: Will Guidara, Eoghan McCabe, Steve Huffman, Brian Chesky, Bob Moesta, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarlandCheck out Jason’s suite of newsletters: https://substack.com/@calacanisFollow TWiST:Twitter: https://twitter.com/TWiStartupsYouTube: https://www.youtube.com/thisweekinInstagram: https://www.instagram.com/thisweekinstartupsTikTok: https://www.tiktok.com/@thisweekinstartupsSubstack: https://twistartups.substack.comSubscribe to the Founder University Podcast: https://www.youtube.com/@founderuniversity1916

You know, there is an ethical argument, as much as, you know, it may be a little bit uncomfortable. There's a pretty strong ethical argument that as soon as you are epsilon safer than an average human driver, you are actually costing lives if you delay deploying this technology. This weekend startups is brought to you by Stripe Startups. Stripe Startups offers early stage, venture-backed startups, access to Stripe Feed Credits, expert insights, and a focused community of builders. Apply today on stripe.com slash startups. Northwest Registered Agent. Starting your business should be simple. With Northwest Registered Agent, you can form your entire business identity in just 10 clicks and 10 minutes. From LLCs to trademarks,

domains to custom websites, they've got you covered. Get more privacy, more options, and more done. Visit Northwestregisteragent.com slash twist today. and lemon.io. Hire pre-vetted remote developers and get 15% off your first four weeks of developer time at lemon.io slash twist. Hey, everybody, welcome back to this week in startups. I'm your host Jason Calcanus,

and I talk really fast because I'm from Brooklyn, born in 1970, was a magazine publisher, then a blog publisher, podcaster, angel investor, founder, all those great things. Man, people, I forget, like, half the different careers I've had, Alex. And we reminisced yesterday a little bit about my publishing days as a print publisher.

And I decided in 26, it's a big announcement. Okay. I'm going to do a print magazine. Okay. Oh, just one. Okay. One issue. But I want to get back to because, you know, I've been reading, I told you about this

Monocle magazine that I'm obsessed with, Monocle.com. and I mentioned them on a couple of the pods previously, but the power of a nice print magazine is underrated. And I got Wired magazine. And it was like the flimsyest print, you know, like very low-grade paper, very low-grade printing process, like the weight of the paper.

But my monocle was really nice. And I was just like, my God, Wired has fallen so, deeply. And it's just paradoxical because they used to be like the most in, I guess WIRE was a little before your time, but Wired was the most punk rock, enthusiastic about change kind of publication. And then I started looking through all the stories in the new Wired. And it's literally like the bitter Karraswisher, like anti-attack, everything terrible attack, just one thing after the next. I get it.

Like, you know, people want to hold truth to power or whatever. It's like a thing. Yeah. I mean, technology used to be this big. And now it's every single large. I mean, I kind of get it, man. I will say, though, why are just something really cool recently? They sent one of their reporters to embed over a notion for a couple of days.

Oh, that's interesting. And they did a great piece about that. So I think that. That's nice. I got to give a little credit for that because every other thing is just like, here's another terrible aspect I'm pregnant. I'm just like, who is editing this publication that like four out of five stories above the fold is just tech is burning? I mean, I don't think it's true, but okay.

I don't think it's true, but I do think it points to a broader discomfort with certain elements of technology. And I was reading coverage from the, there's a big conservative conference going on right now, Jason. I think it's in D.C. And there were several panels about AI and its impact on the job market and what that might do to families. So criticism of kind of technology trends right now is not a, I think that you're implying kind of a left of center thing. I think it's a bit more broad than you and I here in our circles because we're in pretty rarefied spaces, but something to kind of keep in mind, I think, for both of us.

Yeah. I mean, it does, technology is impacting everything that was kind of Wired's original premise is that technology is going to, you know, make contact with every aspect of our lives. And man, that was an understatement. So here we are. But yeah, you know, maybe I call it wired on a bad day or whatever. Or maybe it's just, you know, if it bleeds, it leads. If it's negative, people can get into it, you know, whatever.

But today I was super excited because you and I are obsessed with saving lives because of self-driving. And, you know, we got Waymo. Everybody knows Waymo. We got Robotaxi. Everyone knows Robotaxy. Then there's Zooks, I guess, would be the next one. everybody knows Cruz, that plan got kneecapped, and it's kind of over. And then there's a long tail of other players that people may or may not know.

And this next company, I was really excited that producer Oliver, who has now raised above tolerable to me because of this get today, I am like he's more than tolerable, which is like a big compliment. Ask for a big raise. So he's just gone above the tolerable level because he booked this guest, because he heard us talking about it over and over again. And the long tale of self-driving companies is fascinating to me because I've always had the premise that just, you know, feels like a dozen people could hit autonomy all in the same three-year period. That's not a dig to Waymo or Robotaxe or Zooks or anybody. It just feels like there's so many competitors in the space that like large language. models. We're just going to see, man, maybe two dozen people kind of join the race in the next

year. And I, you know, I'm an investor in a taxi cab company. You may have heard of Uber and still hold a decent size position. And they did a massive partnership with a car maker and this self-driving company. So it's a kind of a three-way. And so yeah, why don't you T us up here? All right. So we are very excited to hear all. at Twist to talk about the neuro Uber and Lucid deal. We've brought it up several times on the show. If you know Uber, you know they used to have a self-driving arm that eventually got sold off. But now they are back partnering with Lucid and it's gravity SUV and EV and Neuro's self-driving

technology. The neuro driver is going to keep this whole thing running. And Jason, this is supposed to hit the roads next year, 2006 in at least one American city. So everybody, please put your hands together and welcome to the show. It's Dave Ferguson, co-founder and president of Neuro. Dave, how you doing? Great. Thanks for having me.

Great to see you. You're Australian? Is that what I'm hearing there? I'm from New Zealand. New Zealand, oh, okay. Kiwi, very nice, very nice. Welcome to the program. Thank you.

You started, and we saw this in the video, having not like the serve robotics, street, sidewalk robot, delivering burritos, but a little bit of a larger delivery footprint in the streets. and maybe we could show that. Here it is.

Tell me a little bit about this product and the testing of it and where it's at. And then we'll get into the self-driving partnership. Yeah, for sure. So we started the company nine years ago. And our original vision was to create delivery as a service. So Jason, as you mentioned, so custom vehicles that drive on roads and do all manner of deliveries from FedEx style packages, groceries, Uber Eats and so on. And we're pretty excited about that. We're still pretty excited about that, frankly, as a business model, but it is very,

very capital intensive. And so we were sort of going gangbusters and then in 2022 or so, the cost of capital skyrocketed. And so we figured we needed to find a new go-to-market that was much, much more capital efficient, much less capital-intensive. And at the same time, because we had been operating these vehicles on public roads in suburban and urban areas, because that's where people in business. businesses reside. We had built very general purpose self-driving technology. So we said, look, we need to find a different path that's less capital intensive, and we have this general purpose

technology. So perhaps we should expand beyond just goods transportation, beyond delivery, and add passenger transportation as well. And so we pivoted to license the tech across mobility for passengers, mobility for logistics, and even personally owned vehicles. And this Uber partnership that you mentioned is is the first pretty massive partnership that we've announced that is along this new vein, this new direction of us around licensing the tech. So what I'm hearing from that is in order to put a physical fleet on the road, you need to build a factory, you need to make, you know, tens of thousands of these. Eventually, you're probably going to have to get to hundreds of thousands and then millions. Easier to work with people

who are building cars anyway, put your software and then your sensors, or is there an equivalent if you're a software company of a sensor company doing what Neuro is doing? So that's a great point. The neuro driver is both software and the hardware. However, the hardware, which is effectively all of the sensing compute that you'd stick on a regular vehicle to turn it into a self-driving vehicle, that is all off-the-shelf automotive grade hardware. So we have the full reference design of which sensors we stick and where do we put them,

but we basically hand that to our automotive partners. So in this case, with the Uber deal, that's lucid, and they integrate that into the vehicle on the production line. When you're a busy founder, finding a new developer, my God, that can become a full-time job, and you've got enough on your plate. I mean, you're running a startup. But lemon.io has done the hard part for you. already. They've got a crop of pre-vetted developers that they've ensured are experienced,

results-oriented, and prepared to make an impact at your startup. And they can work right now at competitive rates. These are skilled, hand-picked devs with a minimum of three years of on-the-job experience. And just 1% of applicants are accepted into their program. Lemon.io isn't just recruiting you the top talent that's out there. They're helping you integrate them into your team. If anything goes wrong, Lemon will find you a replacement developer, ASAP. And many of our launch founders and Founder University companies have staffed up with Lemon.I.O and we always get the best feedback. So go to lemon.com slash twist and find the perfect developer or even a tech team in less than 48 hours.

And Twist listeners get 15% off their first four weeks. Stop burning money. Hire developers smarter. Visit lemon. dot i.o slash twist so how much is that package approximately today because there's a large number of people online having this debate some of them are like the tesla q people who want a short tessel some of them are the tesla bulls and they're all day trading on this stuff so that it's hard to parse reality but they keep saying a wamo cost 175 000 waymo cost 250 000 now i know those numbers were from

like 10 years ago because i remember reporting on them alex members reporting on them. What's the reality today? If you were to guess, you know, when Waymo makes a car, what does a sensor package cost? And when you're going to be making these with Lucid, which is, yeah, just a fantastic luxury car maker, what do you think the package, just ballpark, will cost in terms of buying it? And then maybe what does it cost to bolt it to the car, as it were? Yeah, well, so part of the goal and part of what we loved collectively about Lucid as a partner is that the sensors and the compute will be integrated at the production line, Jason. So there won't need to be any bolting, which is fantastic.

So it'll be a frank in Robotaxi. It'll just be a Robo taxi. Exactly, which, which, you know, saves an extra hop and a whole bunch of additional labor and redoing work and so on and so forth. So that part, I think, is awesome. And then it's purely sort of the cost of what that sense and compute configuration is. Part of why Uber was so excited to work with Nuro is because all of our sense and compute are off-the-shelf automotive grade, we're able to benefit from the economies of scale that already exists for those sensors in the standard consumer car market. And our compute is Nvidia's Thor, SOC. So we're also benefiting from, you know, the incredible scale that Nvidia has.

And so overall, the cost per vehicle of our system, we believe, is roughly an order of magnitude less than what other full L4 players have. Now, that still makes it more expensive today than, say, Tesla's outfitting, but it is low enough. 10K, 5K? Yeah, I'd say, you know, you can think sort of 10K for very low volume, right? And so that already is low enough that we think there's a very clear path to integrating that onto personally owned vehicles because that's very small scale as the scale grows further. that cost is going to come down further.

And so it's... You could see it maybe in a year or two going down to 5K or something at scale? I think with scale, yeah, it's more, it's more what's the scale plan as opposed to sort of time, Jason, because they are already components that have come down a fair bit in cost.

So it's sort of riding a little bit more of the general cost curve, but mostly our own volume cost curve. So Dave, one thing I was impressed by when the Uber Lucid Neuro deal was announced was the timeframe. You guys are talking about having your system live

in at least one American city by 2006. And one thing we've also seen from Tesla and Waymo is they go to cities beforehand, they do a lot of mapping, and they get prepared. I think Waymo calls these road trips. So I'm curious, have you guys begun that process to prep for human transportation? How far along are you? And can you give us a hint about which region of the U.S. you're targeting for your first city?

Yeah, so on the latter part where we're keeping fairly mum on that for now. but on the former questions, Alex, we've actually been operating fully driverlessly for five years now. So Jason mentioned we're proud to be, you know, we hope at the top of the long tail, Jason. Yeah. You've got Waymo, Deslin. Which city were you operating in? So we've been operating in the Bay Area. So our headquarters is in Mountain View.

So the Mountain View, Palo Alto area. we also been in Arizona as well as we have a facility, a close-class facility in Vegas. And then this year, over the last nine months or so, we've done a road trip similar to what you're talking about Alex, where we've actually collected data in 150 cities across the US. So we have a lot of diversity of data that we use for both training our system as well as validating. But it's really those five years of driverless operation that has given us, confidence and gave Uber confidence that we really do have a maturity in the technology that

allows us to deploy in this very, very aggressive timeline that you mentioned, which is late 2026. Talk to me about inclement weather, because one thing we're seeing Waymo just do now is go to D.C., go to Denver, go to places like Seattle where it rains and where it snows, and I live in the northeast. This is pretty pertinent to me. You guys mentioned testing in Arizona, California, drier areas. How does your technology perform when there's snow, ice, rain, sleet, hail, and fog?

We follow a pretty similar playbook, I would say, to Waymo in that regard, in that we've designed the sensors to be effectively future-proofed for coping with very heavy rain and snow. And so we feel very confident that the hardware of our vehicles is going to be able to do it. We have delayed solving all of the software-alonomy-related challenges of operating in those areas to time it with the roadmap that we're going to have of deployment. It's basically, look, we want to deploy as quickly as possible and get to really pretty significant scale as quickly as possible. And that means that on day one, we don't want to take on all of like the heavy snow of the northeast because that would basically delay day one. And so we're sort of putting it in the roadmap.

But we do feel very confident that the hardware that is on these vehicles on the lucid platform will be sufficient to handle those weather conditions. Okay, so on the lucid front, first of all, I'm going to pull up a picture right now, the gravity. the SUV, which I believe is your chosen car. It's stunning. Like, Jason, I don't know what you think about this, but I think that's cool. It's three rows and it goes 450 miles. I mean, these are two very important things when you're coming from the airport or you're an executive. This is what I look for with my family when I order an Uber black SUV. So knowing what I know about Uber, this partnership seems to be not the two-door robotaxie that Elon's building, et cetera, or UberX. This is an Uber-Black offering.

I would think, Dave, you'd agree with that? Yeah, I think it hits, well, obviously any vehicle can do UberX, but I think to your point, Jason, this does, this hits the Uber Black and the Uber XL. And honestly, as you said, it's an incredible vehicle. So it not only hits those bars for today's current service with Uber, but I think it's going to far exceed them and it's going to be a product experience that people are going to absolutely love. Alex, what are the starting prices of those? And there's the lucid gravity.

Are you guys going to do the sedan as well? Or is it just the SUV for the 20,000 that Uber's buying? Yeah, so that's a great point. The gravity is the first vehicle we're doing. Obviously, it probably makes sense for there to be an even lower cost option over time, given that we've already done all the work of integrating the tech and the three-way partnership. So stay tuned for more on that. Sure.

Jason, answer your question. The lowest trim level MSRP for the gravity SUV is 80,000, and it goes up from there. The grand tour reinstals at 95. And according to my researcher, you can actually spend up about 140 on one of these with every single option trim and so forth. That's the luxury, a chauffeur-driven one, I believe. Yeah, Dave, like that 140,000 one is what, like, if you were, you know, a CEO. you might, and you had a driver type situation you would do.

That one is very, very high end. It's kind of the, the, there's, I looked into this because I was thinking about having a driver so I could recapture like two hours a day, I mean, maybe an hour a day, whatever. And I came to the conclusion, like the BMW I7 has a chauffeur driven. Have you seen that satan, Dave? I mean, that's a gorgeous car. And I know somebody who has it, you can like lean all the way back. It's got a big LCD screen or LED screen.

so you can have a wide screen. Alex, you should show that on the screen. I'm pulling it up right now. It is the most gorgeous self-driving car outside of the amazing Alfred and those offerings, Dave, that you probably know about in Japan, Asia, Malaysia, Singapore.

Have you seen those cars or driven in them? Yeah, we've spent a lot of time on the market and you guys have talked about in the past, but China is, it's incredible how quickly things have moved there in general and the quality of some of the vehicles coming out of Yeah, mostly staying in there, which some of our automotive partners in the West are thankful for, but also the tech that they're building just generally is very impressive. Will you have safety drivers in the first year, or are you ready to go without a safety driver?

Hey, listen, we meet a lot of early stage founders here at launch, my investment company, and some, they don't have a lot of traction yet. They just have an idea. Maybe they haven't even finished their product. They've just got an MVP. But they still need investors and accelerators like ours to take them seriously. And you know what, we can't just wire money to your Gmail or your PayPal. That's not how it works, folks.

We need to know that you're a legit and official business. We need to know your company is incorporated. That's why you need Northwest registered agent. It's the service that will help you run your business the right way from day one. In 10 clicks and in under 10 minutes, you're going to file for your LLC or a C Corp if you're a startup, get a domain name, launch your official website, claim your business email, and even fast-track your trademark application, which some people forget to do. We're talking about more than just company formation.

This is your entire identity as a business. Go to northwest registered agent.com slash twist and show the world you're in business. And make sure you use that URL slash twist so they know that we sent you. So over the next year, year and a half, as we're doing all the testing and development and final validation, for sure, we'll have safety drivers. the intent is to launch a commercial service by the end of 2026 that does not have safety drivers inside. Amazing. Price point, Dave, I'm curious.

People talk about Uber, sorry, Waymo being a little bit pricier than Uber's in and around the Bay Area, my favorite place in the world. But not super more money. So I'm curious, is this going to be comparable to Waymo, more expensive? I'm just going to figure out your market position for launch. Where is that Uber's decision, I wonder? Yeah, so that's on Uber. And so they're going to figure it out. I mean, what I can share, which is what we're most excited about,

is that this partnership is over a long enough period that together, we are trying to hit the future that you guys have talked about a bunch, which is drop the price points of mobility services so that they're competitive with personally owned vehicles, which then dramatically expands the overall market. And we intend to get there with this partnership over time. So on day one, obviously, I mean, Uber will decide the pricing throughout, but you could probably get to see the pricing evolve, but that's where that's where the goal is for

us collectively. You know, the interesting thing about this partnership, I understand, is you're getting a base fee for your technology from Uber. Uber also invested, NVIDIA invested, so congratulations on that as well. So that's great that you got a war chest now. But you're getting a per mile fee, is that correct? So we're not divulging the exact economics, but the way that the general relationship goes is that lucid is selling vehicles and they're getting paid for for those vehicles uber is owning and operating them and then neuro is getting compensated in both a fixed and variable way so that we are fully vested in the success and the utilization and and the amount of miles that we're able to drive together with uber and we think where we've landed is actually a really

good place we think long term as i mentioned will be able to hit an incredible compelling price point for the end consumer, and Nuro will be able to continue to play a very asset light position and still capture a pretty strong revenue over time. Jason, I think that was the most artful, yes, I've ever heard from a founder. I don't want to say I have inside information, but I'll just say word on the street. Word on the street, yes. There's a licensing fee, but then as it scales to certain milestones based on miles, not trips, neuro would get a kicker. That makes a lot of sense to me.

Pay for performance. I mean, they want to be incentivized to do well. I think it makes good sense. And also then Dave has a lot of money. I like Dave. So, you know. Thanks, guys. Appreciate it.

Let's go right at the probabilistic versus deterministic debate. If you could, Dave, since you've been in it for nine years, maybe explain to the folks listening, especially the Tesla bulls and the Tesla bears who have been, for some reason, thinking I'm anti-Tesla when I'm one of, you know, an early investor when it was public, I had IPO shares literally, and I own the first Model S and I own the 16th Roads. I'm as all in as you could get in Tesla. I have great exposure to it. I want to see everybody win.

but there is a real big debate about probabilistic versus deterministic models. And I would love for you to explain it to the audience just concisely. And then tell us the reality of this, if this is actually a debate that's worth having or does everybody have the option to use both depending on, you know, the challenges in the short, mid and long term. Yeah, I think the debate tends to be a bit more around whether you have an end-to-end model, Jason, that basically just take sensor input. There is a big model, foundational model that perhaps is a big black box, and then it spits out a single trajectory for the vehicle to execute,

or maybe multiple that then get determined. That is an approach that has come about in probably the last five years or so in the self-driving space, whereas earlier, you know, I was at Google beforehand. I did my PhD, so I've been in this space longer than nine years. Jason, but, you know, as they've seen it all, you know, there used to be approaches that were much more segmented where you'd have one module that takes the camera data and it looks for objects and tries to detect pedestrians and then it hands that to another module that tries to predict what they're going to do and then hands that to another module. I think that all of the leaders in the space have largely converged to leveraging as much

as possible the power of all of the breakthroughs that we've had collectively as an industry. in terms of foundational models, the same core technology that is within LLMs. You know, the transformer architecture was invented by Google back in 2017. It's been around a long time, but that really led to almost all of the breakthroughs that we've seen since, including LLMs. We've been using all of that. As, you know, I'm pretty confident Waymo has as well, Jason. So I don't think that on the technical side, if you spent time within Waymo, within

neuro within Tesla, I don't think you're going to see a huge amount of daylight in terms of the machine learning techniques and tools that are being used. I think there's a little bit of difference perhaps in how we add in validation and additional sort of real-time checks to really enforce safety. And I think there's still a little bit of a little bit of uniqueness that each company may have in that regard. But I think the larger, the larger delta, particularly on the Tesla side, is really, which sensors are you using? You know, how complete is your sensor stack? Do you use LIDAR or are you solely using camera?

And perhaps how much do you leverage maps versus not using any maps when you drive? Those are probably the two biggest. So let's unpack both of those. Let's start with maps and then go to sensors based on everything you've learned. And you start in this industry when you got your PhD, I'm assuming 2010 time period or thereabouts? 2006. Oh, wow. So were you were you part of the DARPA challenge that? I was. I was on the Carnegie Mellon DARPA team. Yeah. Got it. So explain now we got three things we got ahead on here. I want you to explain to the audience the DARPA and your experience there. Take a minute or two to sort of explain that.

Then we'll go to maps and then we will go to sensors. Okay. So DARPA is basically the defense agency in the U.S. that has staged a bunch of really transformative research projects, you know, Arpa created the internet, a lot of the technology led to that. They had a series of grand challenges around self-driving vehicles. The first one was in 2004. It was basically a robot race across the desert at speeds, at speeds that were an order of magnitude faster than any robots had basically driven before in unknown terrain. And so this was one, 2004, no-one won. They redid it in 2005, Stanford one, CMU came second.

Then they had the third one in 2007, and they shifted it to driving within a road network. So basically trying to emulate urban, suburban driving. So you had road rules, you had traffic signs, you had to yield and deal with other vehicles. That was the first one that I participated in. I was at CMU. We won that one, thankfully, and restored some honor to Carnegie Mellon after the second one. And that was what really kicked off a lot of, I'd say the resurgence in the US around self-driving. You know, that was the grand challenge that Larry and Sergey were at,

and they were then inspired to start the Google self-driving car project. That project pulled a lot of the folks that were on various teams at the DARPA Urban Challenge into Google, to work on the Google self-driving car project. It was a phenomenal group of people. And that's where my co-founder and I met each other. Actually, we worked together there for five years. And then we started Neuro in 2016. Incredible.

Okay, so let's get to Maps and then Censors. What have we learned? Because Map Fidelity back then was close to non-existent. GPS was not high fidelity. But we had 20 years of Google Maps, Apple Maps, and GPS evolution and different rules around that that the government set in terms of fidelity. So what have we learned about maps over the last two decades and the usage of them in self-driving? Everything is moving so fast at early stage startups today.

It's amazing what I'm seeing people get done. Founders need simple solutions that can live in the background and solve their problems, right? Well, now they're Stripe startups. Their startup program now offers a ton of business. benefits for your business. And this is when you know when a business like Stripe specifically creates something for startups, you know they really care about founders. So you get Stripe-free credits, you get a unified dashboard where you can track your performance, and you can even start issuing your brand's own payment and credit cards in just weeks, not months. For AI companies

specifically, that market is shifting and changing every day, requiring an unprecedented level of flexibility from payment providers and partners. That's why 78% of the Forbes AI 50 are already using Stripe is their financial infrastructure. Here's your call to action. Just learn more about this incredible program and apply today by visiting Stripe.com slash startups. Very simple. Stripe.com slash startups. Yeah, so I think that my personal opinion is, and you'll hear this on the LiDar side as well, Jason, is not to be binary about these things. Like, oh, maps are good or maps are bad. I think the reality is if you use human driving as an analogy, we drive better when we have an idea of where we're driving. You know, when we have a map either that we're following or it's an area that we've driven before and we have an internal map.

So I think maps are always going to provide some value. I think the argument is then, well, how much value and can you just drive where you don't have a map? And so I think that the proper debate or the more valuable debate is really, what's the cost of, of building maps and maintaining maps, and does that cost outweigh the incremental benefit that you get from them? I think where we are going and where the industry generally is going is, there will be prior maps, we will try to take advantage of them. We probably will be able to drive when we don't have them,

but we're always going to prefer having them to not having them. And once we've driven an area, there's no reason we shouldn't then be able to have a map so that the next time we drive, we can drive more confidently. Dave, with modern technology, with stuff you have today, how big of an impact does having a map have on reducing interventions and small collisions and such? I'm trying to figure out how impactful that is today. I presume it'll go down over time as tech improves, but where we are, how much does it matter? Yeah, I think

that's a hard question to answer. We haven't done too many of the very detailed studies of, hey, if we remove the map. We do have systems that can operate without maps, Alex, but we haven't dug very deeply around like how much does it actually improve. What I can, what I can, what I can give you as a couple examples that are pretty clear cases where having a map can give you better performance, even if your onboard system is perfect. And one of those is imagine you're driving down a road that has a median and there are like trees on the median and there's a speed bump in the oncoming direction. And you have an intersection, maybe it's unprotected intersection and you're looking to turn left, right? And so you know you have to yield to oncoming cars. If you have a map and you know

that there's a speed bump in that oncoming lane, which, by the way, you couldn't see from your sensors. So even if you're on board sensing is perfect, you can't see it because it's occluded. If you know there's a speed bump, then you know, even though there's a speeding car coming, that they're going to have to slow down for that speed bump, so you have time to take a left in front of them. You're never going to be able to replicate that quality of behavior unless you have prior information in the form of some kind of map that tells you that. So there's always going to be some value, even when we're 100% perfect with onboard sensing. And so in our minds, it's really how do we ensure that we can make it an absolute slam dunk ROI obvious decision that the cost

of us developing maps is low enough and the value that we're getting from them is high enough. And we feel like we are there. Like, we feel like there's no reason not to be using maps as long as you can make them low cost enough and easy enough to generate. Let's shift the discussion. here. We'll get to sensors in a moment, but the disclosure of data around self-driving is a question I can't figure out a definitive answer for. I know Waymo discloses their aggregate stats today, and that's required by California and Arizona or not? Yeah, there's a little bit of nuance there. So in different states have different requirements over disclosure. It's mostly a state thing, Jason. So like the California DMV regulates all self-driving vehicles, and then there's additional regulations if they're in a commercial operation.

And so the DMV has some requirements on reporting. One such requirement is that they want you to report all disengages. Now, the exact definition of what it disengages is fairly ambiguous, which has led to some critique of how useful this is. But certainly things like accidents have to be reported. And that is true, I believe that is true nationwide, but certainly that's true on the state level. But beyond that, it differs based on states. California is the most onerous or the most thorough in terms of what level of reporting it requires. So if we talk about California, I'm sure you studied, as did the entire industry, what happened with crews.

Now, to just catch the audience up, maybe, Alex, you could explain top level what we know about the cruise. shut down of their unit and how that came to be. And then Dave, we'll kind of unpack it a bit. All right, Jason, correct me if I get any part of this wrong. But there was a accident involving a person and a car. That person, the human, was pushed into the way of a cruise vehicle. The cruise vehicle struck the person and pulled them for a little bit. Now, when someone else gets pushed in front of your car and it's not your fault, I don't think we'd all point a finger here at the self-driving company. But what Cruz did, that was lethal to its business operations Jason was to not show the full clip, if I recall, to regulators.

And so essentially, the old phrase in politics, it's not the crime, it's the cover-up, came into play. They got into quite a lot of trouble. And GM kind of pulled the plug on the project, moving a lot of the cruise people into... I think you explained it perfectly. Yeah, they dragged the person for some number of feet, and they lied about it, allegedly, or at least that's what California said. So, and then obviously Uber had a tragedy in Arizona where they had a safety driver who decided

instead of doing their job, they would play Candy Crush. And they just overest it. And the technology that led to Uber pausing their program. And eventually Dara decided he would go with partnerships and sold that unit. And that's what we're discussing here today. What did we learn specifically from the cruise accident, Dave? The industry yourself. Look, I think if anyone thought that you can operate a business in this space

without investing very heavily in establishing the trust of regulators at every level and, frankly, the community that is a huge misunderstanding, right? And I think, you know, when you mentioned the cruise and the Uber incidents, you know, both of those were tragic. One of them, Alex, as you mentioned on the cruise incident, the initial accident was not actually cruises fault. You know, it was another car actually hit a person and flung them into the in front of the cruise vehicle, which stopped, you know, from what I could tell, it reacted fairly well, at least initially. Whereas the Uber accident, you know, arguably that was, you know,

some fault of the combined operations that Uber was running. And yet, Uber would not have been shut down for that, right? Like Uber shut it down exactly as you said, Jason, because at some point, Dara made a strategic decision that it didn't make sense for ATG to. to be an Uber-owned effort and that Uber should partner, right? And so I think... ATG was the Advanced Technology Group, which was the subsidiary of Uber. They eventually spun out, but that was their self-driving group. Just keep me alone.

Yeah, sorry, ATG was their self-driving group. So I think that even just between those two examples, there's a pretty good lesson around how much of an impact being very forthcoming and really like standing up when you make a mistake and owning up to it and trying to make sure that you're all. always prioritizing that trust in your communities. For us, we've invested incredibly heavily in relationships with regulators at the federal level, even the state level. When we started the company, Jay-Z, my co-founder and I, we didn't even tell employees what we were working on when they joined the company for the first year or so. We were that, I think, overly anal about it,

but within that time, we were open with NHTSA. So we went and talked to NHTSA, you know, the federal regulator of vehicles. We said, hey, we're building a custom occupantless self-drawn vehicle. We want to have a really good relationship. We want to be forthcoming about how we're doing this and the safety and so on. And so we've, I think from an early day, we've, we've recognized that regulations are a critical part of us being able to scale and deploy this and actually realize the end goal, which is not a few pilots or getting a vehicle out onto the road. It's really getting hundreds of thousands of these out. And you're not going to get there without. frankly a few accidents along the way. And so you need to make sure that if and rather when they happen,

you have a really good track record and that you do have trust and you certainly are operating from that position of trust throughout the sort of the crisis of what's going on. Have you had accidents during your pilots and maybe you could describe some of them and what you learned? So we've been operating autonomous vehicles for nine years. A lot of those have had humans in the driver seat, a little bit like the Uber ATG case, where we're running it in autonomy mode, but we have a safety driver. And then we've also had five years of driverless operation with our custom vehicles. So those ones have no one on board.

We, today, and I knock on wood, we have had zero accidents that were caused by Neuro's autonomy system. So we have had some accidents, you know, some of them are sort of fender benders, but in every case, it was sort of judged that it was the other vehicles fault. What was the worst one? because you've had to disclose these, right? If we did a deep dive, we could find them. So what would be, um, describe like the most,

uh, concerning or, you know, maybe it's not even concerning accident. Yeah, we, honestly, we've been very lucky.

We have not had any significant injury causing accidents. So, so today. Now remember for the first six years or so, we were operating in very much urban, suburban areas, not freeways. We've expanded to freeways in the last couple years.

So the speeds were lower, but certainly the speeds are similar to what cruise, you know, was dealing with in San Francisco. We'll call them 35 miles per hour. You get an accident between 15 and 35 miles per hour with another vehicle. Zook's had somebody bump into them. The fender benders, we're just talking about property damage. Maybe somebody gets a little whiplash in the car.

So no harm, no foul. Let's take a look at the TeslaFSD tracker.com. This has been, and I'm sure you guys look at this on the regular, I guess what they do here is, is my understanding is fans of Tesla or superfans or just passionate people, upload their data on some regular basis. And we see this tremendous improvement on critical disengagement. If you were to look at the 2001, 2020, you know, 82% of drives according to this.

And this is a human superfan who has the car saying, hey, 82% of the rides had no critical disengagement, meaning 18% did. So one in five had a critical disengagement defined as like, hey, this could have caused property or human damage. And now boom, all of a sudden the spike here, they're up to 98% basically. Is this data accurate in your mind? And certainly the trend line is, but what do you take away from this kind of data? Well, I think in general, the biggest challenge in launching a fully driverless product is not actually driving nominally on roads, right? And arguably, you could go back and say in the early days of Google, you know, they had this internal competition called the Larry 1K, where we had to do a bunch of routes through the

Bay Area with no disengages. And that was back in like 2009, 2010. And that got done. The challenge is really dealing with every possible edge. case and weird scenario and the people lying down in wetsuits on skateboard, shooting down hills in front of the cars. You know, that's, that's, that's what the bar is to actually launch something where you don't have any option of a human jumping in and taking control. And so I think that that is where the real challenges. And, you know, I personally feel like FSD is an incredible driver assistant product.

I think what Tesla has is, is phenomenal as an L2 plus system. But there is a massive, massive gap between a system that operates at 98 or 99% of the time without hitting something and something that the three of us are going to put our kids in and trust that it is going to be safer than us driving behind the wheel. And I think that that is what Waymo has demonstrated very clearly to be able to provide that. And that is what we will be launching when we launch with our partners. It's like, I think that's, that's the delta. So, Waymo is what, 99.3 nines, 99.2n's. Where are they at with critical disengagement, you know, in terms of industry stats and reporting? Yeah, I mean, I don't want to speak too much about, I'm not deep in the internal weeds there,

but they did, they did release some stuff recently that suggested, you know, with enough, with enough statistical significance behind them, that there's something like five times as safe as a human driver. Got it. sort of an average human driver. And you know, you can add all sorts of asterisk to what that means, but very clearly safer than a human driver. And if we jump in a Waymo, I think we have every reason to believe that it's a safer option than us jumping in an Uber X or a human-driven equivalent. What do regulators need to see in what do consumers need to see here in order to feel comfortable with these systems explicitly saying as good as humans, twice as good as humans,

five times as good as humans, 10 times as good of humans. What does your knowledge as somebody who's been a technologist in the space for 20 years or so, what do you think the bogey is? What has to be achieved? I think when folks, right, I think Waymo has done a good job sharing some of the numbers to give people confidence. And frankly, I think the numbers that they already have, people, again, when the option is getting in a car driven by a human, like a mobility service vehicle versus getting in a Waymo, it's pretty clear, which is a safer option. And so I think that they've already done what needs to be done technically. In the US from a regulatory perspective, there aren't clear bars. And I think it'll be hard for us to go down that path.

You know, exactly to your question, Jason, how would you set it? And so I don't think we're going to see clear bars there. So it really is more on the consumer acceptance piece. You know, there is an ethical argument as much as, you know, it may be a little bit uncomfortable. There's a pretty strong ethical argument that as soon as you are epsilon safer than an average human driver, you are actually costing lives if you delay deploying this technology. Now, you know, I think we would all love to have it be 10 times safer and so on. And for sure, that's where we're going to get 10 times, 100 times, whatever. But again, the more we try to establish a very, very arduous initial bar, you know, frankly, the more lives we lose it. We lose 40,000 lives a

year on US roads. As soon as we can start reducing that, it's a moral imperative for us to do it, right? So I think we shouldn't, we shouldn't let that bigger picture be lost in our desire to make sure that we're on a trajectory to dramatically improve the situation, not just incrementally improve it. Yeah. I think it's a really good point that there's a moral imperative to get the stuff on the road because it is safer. And I think that that is so lost amongst the discussion of things like the Cruz situation. Like I hadn't reprepped that day. I just had that burn to the bags. We've talked about it so much because that's what the public wants to discuss. But the thing that I'm concerned about isn't the fact that Cruz had a problem once.

It's that there's a 16-year-old driving suburban down my little residential street at 50 miles an hour because they have the IQ of lettuce. And that is what I'm concerned about. I want to go back to the top, though. While texting, by the way. Yeah, well, no, no, they're on T-T-T-T-T-T-T-T-T-T-T-T-T-T-T-T. Dave, at the top, Jason said that he thinks there's going to be a number of companies that get to the point in which they have a commercially ready

and publicly safe self-driving situation. He said 12. I was thinking more like four of maybe you guys, Waymo, maybe Tesla. How many people do you think are going to get this right at school? scale maybe in the next five years. Yeah, I think I would lean much more towards yours, Alex, like a handful. It's just, guys, it is technically incredibly difficult to do this. The last few years, and you guys talked about this on an earlier episode, the last few years, we've already seen a ton of companies fail or give up. And this is not like just little bitty hero startups.

Like it's also Apple spent billions on Titan and gave up. Uber shelved their own effort. You know, Lyft had an effort. Like these are really, really capable, very strong companies that have recognized this is very difficult to do. So I don't think we're going to be seeing a ton. It's also very capital intensive, even if you try to optimize for capital efficiency like we have. You know, we've raised $2.3 billion to build hard tech. Like it's not, it's not easy to either build the tech, nor fund the tech, nor get the partnerships, nor, you know, make it through the regulatory and consumer acceptance hurdles that you can have to do. So I, I don't think we're going to see more than a handful.

I do think it's going to be a few, though, because this is just way too big an opportunity, both in terms of TAM and in terms of societal impact, for it to be just one or two companies. So if we were to envision this world where five or six people get there, we also have the OEMs and Nvidia playing a big role. But open source, you know, there have been some open source projects. I think there's been some open data projects. Obviously, there's open maps and stuff like that. Is there an argument, and I think Bill Gurley's made this publicly many times, for there to be open data, open source, open weights, etc., to accelerate this and maybe reduce the proprietoriness of it because we now have 20 people going after this prize. Everybody seems to be doing it in a vacuum. Have you considered? You must have.

open sourcing some of this to accelerate? And is there anybody doing an open source project that you find notable or that you can learn from? There's not, I mean, in general, I would say the research community around AI has been hugely valuable. You know, we have a very strong internal research team. It takes the state of the out that's being published and we build upon it and we expand it and we have some proprietary stuff internally. But there's obviously value. So generally, the research community is providing effectively open source accelerants to all of us. I do think for self-driving, because it is so safety-critical guys, like the bar and the importance of getting it right makes it fairly challenging. Like we spend a lot of time and effort on getting really clean data,

making sure that it is in really, really good quality, that we are collecting a ton of real-world data. So I think everyone in Waymo is for sure the same. The key players have their own data that they've invested a ton in. And so there are sources that wouldn't be available in an open source capacity. And then there's also the architectures and how you validate a safety critical application. Because while we're still leveraging similar AI foundational technology as, say, open AI in chat GPT, the consequences of us getting something wrong are wildly different. And that shapes how you develop the architecture, what safeguards you put in, how rigorous you need to be, even how careful you need to be about the quality of your inputs.

Because for us, we can't, you know, Chachy-B-T phenomenal. I use it all the time. Still, not that rarely I will get a pretty wacky answer that's totally off-base. Like we can't have a wacky trajectory that's off-base when you're driving, you know, my kids or your kids around. Yeah, yeah, yeah. And so there are some differences and some real investment that needs to be made specifically for this problem, which I think makes it harder for it to just be open source, use the internet of data, and you'll end up having something that is going to pass muster with an OEM or a mobility service from their safety requirements. What has NVIDIA's data and sensor efforts had on the industry to date? because it does seem like Jensen is saying, hey, we can really help here by making the hardware just commoditized, standardized. We're going to have some data in the cloud for you to work with.

I understand. So talk a little bit about Nvidia and why you selected them as an investor and their impact on your company, specifically in the overall industry. Yeah, so Nvidia. Nvidia has done a lot generally. I mean, by far the biggest value that. that they've provided us, which is tremendous, is that their latest system on a chip, the Thor-based architecture, in our minds is the first SOC that they have developed, or frankly,

that anyone has developed that has the compute required to do full L4, full driverless operation on vehicles. And so what that has meant for us is, you know, we've been doing this a long time, and previously we had to have something like five different compute elements. So we had to have FPGAs that connected directly to the sensors that were. would process the raw sensor data and then we had CPUs, we had GPUs, we had safety computers. We've been able to largely collapse all of that into just this Thor SOC, which means that it's much cheaper. It's also much simpler architecturally and that also has potential safety benefits because it's been designed to be automotive grade and have all of the sort of ISO-26262 compliance

that you need. And so really, Nvidia has made a massive, contribution through their SOC chip sets to autonomy generally. They also have a number of products that you mentioned, Jason, on more of the cloud side and simulation and providing tools. We're partnering with them on a bunch of stuff, but I will say by far the biggest value is being able to have Nvidia hardware brains in the car that do a great job and are much, much more power and cost efficient than what we were dealing with previously. An SOC for folks listening, a system on a chip, which you basically, if you were to describe it, means what?

It's basically an entire computer on a chip. So instead of having like one CPU and a GPU and different elements, all of it is on one chip. So everything that you need to process data and control the vehicle. And FGPA is the field programmable gate array, if you were curious. Yeah, I'm sorry, guys. Sorry, guys. Too much of a dork on the show today. You talk and we will translate for folks who are catching up in some cases.

In some cases we're catching up. Yeah, this Thor product seems to be a game changer. Waymo did a deal with Toyota. They announced it this year to provide their product to Toyota. I think Volkswagen, that group has their own internal group. But it seems like Salantis and Ford and everybody else is up for grabs in terms of partnership. So have you announced any other partnerships beside the lucid one? And do you think they're going to just pick one partner or this will be, you know, the kind of thing where I might pick one day

to have Waymo or Noro as a package added to my car? How's that going to play out in the next couple years? Yeah, I think in the next couple of years it's probably going to be each OEM having one partner. I think long term it may be a possibility, Jason, but I think things are a little bit too bespoke at the moment with specific sensors and also frankly these engagements are very very heavy in terms of the effort required. We have not announced any but I mentioned this new go-to-market for us is to license our tech across mobility and personally owned vehicles so it's definitely an area that we are investing heavily in. We see a future where Uber is going to have fleets of AVs that are managed by them or more likely a third party and there will be a

TVs that are on the Uber network that are personally owned. And so we want to make sure the neurodriver is powering as many of those as possible. And so we do have a number of pretty advanced partnership discussions going on, but nothing to share concretely today. I think this is the really interesting future, Alex, is being able to buy, because really the only car you can buy with FSD right now is obviously Tesla, right? There's no other car you can buy with greater than a level two. So am I correct, Dave?

There are a couple L3 vehicles that are out there. So I think Mercedes has one. I think BMW has one. So we are starting to see. It's usually fairly limited, Jason. So it might be L3 on highways, certainly not L3 everywhere. But I do think we're going to see some pretty interesting products that will progress from sort of L2 plus everywhere.

to L3 then in expanding set of areas and eventually we'll transition to L3 or L4 everywhere. Now, before you guys have to correct me, L3 is basically you're in the driver's seat, but you don't have to pay attention. So this is one step above what Tesla has today. You can be on a call. You can be sleeping. You can be eating, whatever. L4 is you don't even need to be in the driver's seat.

In fact, you don't even need to be in the car. It's doing everything on its own and it doesn't need any human fallback whatsoever. So Tesla's L2, Waymo's L4, and Tesla, we believe, is on the brink of having L3, where you might be able to check your email while you're driving, yeah? I think they're hoping. They're hoping to get there. I think time will tell. Should there be a federal standard in the United States for self-driving, or should states have that right? You know, it's the United States of America.

I know you're from down under. I'm sorry, you're from New Zealand. Also down under, but not the right moniker. Is that being discussed? Have you been involved with discussions about a national standard? Is the industry lobbying for that? Or is it best for today, you know, states to make these decisions and have, you know, a couple of different, let's call them, experiments running concurrently.

California being incredibly rigid, maybe ours. Arizona being pretty permissive because they have a grid-based system and most of the roads there were designed in the last 50 years as opposed to New York and Boston where they might have put these roads in 150 years ago. Yeah, I think long term a national standard makes the most sense, right? I think similar for vehicles. We don't have different FMVSS, which is the federal motor vehicle safety standard that regulates whether a vehicle can go on roads. We don't have different standards for that to be on California versus New York. And I think it's going to get confusing. And it is very challenging for states to do that. I mean, for anyone to do this, any body to,

sorry, any regulatory body to figure out what should the standards be and how do we regulate this is a lot to take on. And so expecting that every state is going to build up that expertise and do it really, really well, I think is tough. So I do think the industry is heading towards a federal standard. I think that's what makes the most sense. I think the players in the industry on the AV side would frankly love to see that. We're not anti-regulation. We just love consistency and clarity. And frankly, it's kind of good for us. I think it's good for the very responsible players for there to be clarity and help build public trust, knowing that there are clear driver's tests that these vehicles have to go through. And it ensures that no sort of bad actors will

have significant negative impact on the entire industry. All right, let's wrap on LIDAR, the great LIDAR debate. It used to be very expensive, $10,000, $10,000 to put LIDA on a car. Even more, yeah. Yeah, back in the day, I guess probably the first prototypes when you were at Google, $100,000, right? With the sensors, the DARPA ones were maybe a quarter million dollars in sensors, maybe a ballpark?

Yeah, the original LIDARs were around $75,000 for a sense. single spinning, you know, we call it the KFC bucket on the roof. Yeah. I don't mind them. Some people don't like them aesthetically, but I like safety above all else. The computer vision versus, you know, cameras versus LIDAR debate, obviously you've chosen LIDAR and to have those sensors. Is there any reason not to have them if they're sub $5,000 now?

No, honestly, I don't think so. And I think that the computer vision versus LIDAR is how it's commonly framed, but it's not actually accurate because no one is doing LIDAR without cameras. I think we all recognize cameras are incredibly valuable. I think the question is just, is a LIDAR providing positive RUI or not? Again, it's similar to the maps. So is the incremental value you're getting from a LIDAR worth the incremental cost and complexity of having the LIDAR on your system? I think today, it's an absolute no-brainer, yes. From a physics perspective, LIDAR provides different information from a camera. It's an active sensor. It's giving you direct range measurement. And so if you have a mature enough AI-based autonomy system, it should be able to extract some value

from this complementary sensing source that is alongside the camera. And so really, I feel like it's kind of an easy one. I think for sort of historical reasons it's become this sort of big debate. Religious debate, it's very strange, yeah. Totally grand. Ten years ago, ten years ago, I think it was not feasible, Jason.

Like when there were $50,000 that had come down, like, I totally agree. You can't put that on a personally own vehicle. Absolutely. But today, that's very, very far from the case. When you did your testing and you've tested driving with just the cameras and with LIDAR, what today is the difference

in terms of disengagement and safety? Is it a 10% difference, a 2% difference? I mean, each percentage matters, but you've obviously tested and said, hey, do this ride without the LiDR or simulate this ride without the LiDR. What did you learn in those tests? Yeah, so when we run,

so we have a version of our system that operates without LiD, and you can think of that as sort of providing an L2 plus FSD-like system to our potential OEM partners. So there's definitely value in that. It performs incredibly well. I mean, the reality,

And this shouldn't be a surprise because Tesla also performs incredibly well. But that one is one that we're able to evaluate against our validation requirements for doing driverless. And we could see places where it very clearly did not perform as well as we needed for the bar for driverless. And there are a lot of videos posted online of what these could look like. At night, it's dark. There's people in black clothing that are passing the roads. I mean, frankly, some of the videos that we still see from Vision Only systems are pretty trivial cases that with a LIDAR you would never have any problems, right? Like black cars that are across the road on the freeway at night.

And so there is very, very real value in having the LIDAR, but it is very much at the edge cases, Jason. It's not like for the 98 or 99% driving. It's really for, hey, these weird things that can happen very frequently, one or two percent. is still very frequent that require or at least benefit significantly from having this additional sensing modality. There was a YouTuber Alex that did a series of tests, including the Wilde. Coyote painting of a fake wall. Maybe you could cue that up.

You could tell us, Dave, when you saw Mark Rober's test of this, what your takeaway was, fair test, unfair tests, and obviously people have created multiple tests, but does that accurately depict the impact of LIDAR? You know, I think that the tests, I think it's interesting in that, hey, this looks like something that a human would not fail at. However, if you had a distracted driver, they may well fail at this. I mean, it's specifically designed to make it challenging for a vision-only system. So I don't love this test as much as just the very real examples of where not having LIDAR fails, like the sort of the black cars at night or the dark pedestrians, because this is not really, unless someone is specifically trolling you, this is not a situation that you're actually

going to realistically encounter. And so, you know, it's sort of in some ways interesting, but I think actually much less relevant and less interesting than the very real cases that we see all the time. that are challenging. I think it's more interesting. You're absolutely correct. When I saw this, I thought this is very entertaining, but not informative. If you were, you know, playing with the radio or you were kind of dozing off in your cross-country trip and you had a two-hour drive

from Austin to Houston or whatever, yeah, you would fail this. If you were paying sharp attention, yeah, the human would kind of see the edges of this and say, what's going on over here? I'm going to slow down. Whereas maybe a camera might not today, but the camera obviously could be tweaked to kind of figure this out over time. In some ways, the fact that you can even run this test today speaks to how far along we are, yeah, Dave? Yeah, I think so.

And like I said, the FSD and the L2 Plus products generally that we have today are incredible for what they are. I think that they're great products. I think that it's just when we start, you know, trying to confuse or conflate them with higher levels of autonomy me that it gets more challenging. How can you compete with a partner or a competitor like Tesla that if they do figure out computer vision and don't have to put LiDaron and Yulang can make, you know, I think he's made 1.9 million cars in the best year. How can anybody compete with them? I think you do it by partnering with the best of the best, right? We're partnering with

the best mobility service in the world, Jason. You know, Uber is phenomenal at what they do. We need to make sure we have an OEM that can build cars. that are competitive with Tesla in terms of price and performance, and we need to make sure that our autonomy is also competitive or better. I think the LIDAR part is not going to be the game changer there. Like the LIDAR is incremental cost. It's not even $5,000, Jason. It's more like $500.

So that's not going to be the difference. Is that correct? Because I've heard a debate about that with the LIDAR that like people say, oh, it's $500, but that might be one that you wouldn't actually use you know, and they're not quite there yet. Like the LIDARs you're using today are more than 500, I would assume? Well, we use several solid state LIDARs. And, you know, I won't give away our suppliers negotiated pricing,

but it is, you know, it is definitely sub $1,000 per lighter. Amazing. Amazing. You've been to China, I assume, and tried some of the self-driving cars over there. And we've also seen in Wuhan, of all places, some protests from drivers there. And there is some discussion in China of limiting the number of licenses to protect jobs because young men in China, they have these driving jobs. And when you get to 20% unemployment amongst young men, that's not a great thing for

societies. You can see Spain, Greece, Italy, and Egypt and some other places where that's occurred. It's not pretty, folks. what did you learn about some of those Chinese players? Are they the real deal? And how do you look at them as competitors? Yeah, from an AV perspective, it's honestly a little bit hard to tell. I think that we do have respect in general for the AV industry in China. I think that moved very, very fast. I think they're, so they definitely have something. the bar for performance for launching driverless in China versus the U.S., I suspect, is dramatically different. We don't know exactly what that is, but I've heard that there's pretty strong support from regulatory bodies over there to help, even in the event of accidents and so on, to really help

accelerate that industry. I'd say broadly, though, the automotive market, what we've seen in China's phenomenal. So BYD was our partner for our R3 custom vehicle, the most recent one that we were intending on scaling and then decided to pull back on. BYD is an example of one of the absolutely phenomenal automotive companies from China, the speed with which they move and the quality of the product. I think that's the biggest change over the last 10 years. The quality of the vehicles that they're putting out, both fit and finish, but also the tech that's inside them of being able to just speak to them in natural language and all of the tooling and the ADAS and so on. Like it's it's absolutely phenomenal and i think that you know there's very very good reason for

european and u.s automakers to be pretty nervous and frankly i think that's where a lot of the protectionism pushes is coming i think in the u.s it's going to be successful right i don't think we're going to see a lot of chinese evis and certainly not chinese avies uh penetrate the market but but But I think there's no question that the underlying vehicle technology is very impressive. And here I have a picture of the B.YD Dolphin, Jason, that you requested. Everyone took a look. Yeah, I took a couple of rides in these. They're all over Mexico City.

They are dirt cheap. I think they are $20,000 US and don't know what they charge for them. It is obviously a Tesla Model 3, Model Y knockoff down to the interface of their touch panel is literally like stenciled on top of the Tesla one. It's, you know, they're not known for their respect for, you know, IP in China, obviously you can check Disney in the music industry in there and the software industry for that. But you sort of alluded to the protectionism.

I mean, if they allow these for sale in Germany and the UK and France, it's the end of BMW. It's the end of Mercedes. It's the end of Volkswagen in those countries. I mean, just old people might have an affinity for those brands, but young folks are not going to pay 50% or 2x, right, Dave? Yeah, I think it'll be, I think it'll be tough. I think the high end, you know, there's probably still going to be some competition, but these are very, and you show the dolphin that is one of the absolute cheapest low end ones. But I think where they've also made huge strides is their high end vehicles are incredible.

And in China, I think the big shift is that, you know, people, you can look at the vehicle sales for Mercedes and others. like they're starting to get real competition at the high end because the products are really, really good. Listen, Dave, I really appreciate you being so honest and educating us so much. We wish you continued success with it. I think the work is important. Alex and I, I can speak for Alex in this regard. The two of us just want to see all of you succeed.

We'd love to see a half dozen of you get these things on the road because it's going to take, you know, by my estimate, you're going to have to make, I think there's 50, million cars made a year, you would need to take up all the capacity for a half decade to move everybody to these systems. And then that doesn't even account for the fact that we're going to induce more rides. So the concept that the number of rides we see on the road today is the prize is not correct, because lower prices will induce more utilization, as everybody has seen, when we first were investing in Uber, I was the third investor, a fourth investor,

I can't remember, it was, you know, everybody was trying to figure out, well, will we have more Ubers than taxis in a city? And we very quickly had five times as many. And the question was, well, what's happening here? And it was car ownership was going away. And then people said, you know what, I can go out on a Friday night and have a glass of wine or three.

And so how soon do you think we hit, I don't know, 20% of rides globally are part of this new ride-sharing FSD world. Do you have a vision for this? And then what do you think the TAM is? I think you've laid out all the reasons why it's going to take a little bit longer than we'd want. And I think we're going to be in a world where it's prevalent and ubiquitous. And you can use it.

And the three of us are probably using it all the time. And yet that number is still going to be stubbornly low because you mentioned cars today last 15 to 20 years. Like they're going to stay in circulation. We're going to have to transition not just a few OEMs that have fully autonomous products that they can sell, but basically all of them, right? And so I think we're going to still have the total relative penetration of AV being a bit lower.

There's going to be some holdout of still human driven vehicles. But I think the impact and in terms of the opportunity that we're going to have on a daily life basis to completely shift how we use mobility. I think that's going to happen in the sort of the five to 10 year timeframe. All right. Listen, it's amazing talking to you, Dave. We wish you continued success. Thank you, guys.

Great being here. I swear to you guys, the best part of the show is getting to drag founders on here and just pepper them with an endless number of questions. I prepped for this neurochap, but I learned so much during it. Just a real treat. Anyways, this has been this week in startups. We're back here. We do it Mondays, Wednesdays, and Fridays, about 1 p.m. Eastern, about noon central.

If you're still here on YouTube, throw us to subscribe. We'd appreciate that. In the meantime, I'm Alex. He's Jason. We're out of here. Bye.