Only One of These Cars Can Make Money

I'm 50-50 on whether or not we'll be able to do all three of these stops without having some type of critical disengagement. Netflix versus Blockbuster, MySpace versus Facebook, BlackBerry versus iPhone, Tesla versus Waymo is tracking to be on that list.

You can't go one block in San Francisco without seeing someone taking a ride in a self-driving car. They're everywhere. Self-driving taxis are not just coming; they're here.

Waymo has just surpassed 10 million fully driverless rides.

Zero rides in self-driving cars five years ago? 100,000 per day today. And now, the long-awaited entry by Tesla into the competition has happened. Tesla has rolled out its robotaxis. And what makes this so interesting: the battle's shaping up between Waymo and Tesla is that each of their strategies is almost completely a bet.

The other company's strategy is dead wrong, so completely opposite that if one works, the other one will likely be counted as obsolete. Like how the hardware keyboard by BlackBerry was ended by the full touchscreen from Apple. But is Tesla even close?

They launched 5 years behind Waymo. So I had to run a test. Let's do this. That guy ran a stop sign. We have just merged onto a four-lane highway. Somebody honked at us. It doesn't know what to do.

The streets of San Francisco: three destinations, three rides in a Waymo, three rides in a Tesla. Can both or either of them do it without any problems? No human intervention. And importantly, as you'll see, can one of them complete the challenge significantly faster than the other?

I was genuinely surprised by the results that I got. Interesting. There's a coneright in the middle of the road. Oh, this is the same road. Oh, it's still there. Allright, let's see. Let's see. But first, the bet. The bet that Tesla is making is that I, with just two eyes, can drive a car.

If me, no way to see out of the back of my head. Certainly no way to see front and back at the same time can do it. Shouldn't a car with 8 high-definition, wide-angle cameras that can all be looking all the time be able to do it?

Everything is designed for human drivers. The stretch words on pavement, to the yellow chosen for critical warning signs. Humans can see yellow the easiest, especially over long distances. This is a very specific pattern you've seen before. It is exactly the same everywhere you go in the U.S.

The first letter on this sign will be 6 inches tall. Add spacing and arrow. The whole thing should be 52 to 55 inches wide. It says soright here in the manual on uniform traffic control devices. 1,113 pages long.

This is just chapter two. 92 pages. An absolutely comprehensive rule book for how to mark and design roads. So yes, if you measure this sign, sure enough, that's what it is. This stuff isn't just ballparked or like, yeah, that's pretty good.

It is all, if not exactly the same, very similar. The tendency to group all signing in the immediate vicinity of rural interchanges should be avoided by considering the entire route in the development of signing plans. Extra effort should be given to the placement of signs at natural target locations to command the attention of road users, particularly when the message requires an action by the road user.

In other words, put signs where and when drivers need to know about something they may need to do soon, like exit. Anticipate their needs, like they'll be required to turn left orright at the top of this exit. It's programming.

It is code, but for humans in a 3D environment. If you can get human or better vision attached to a vehicle and then put a brain in the vehicle, should it not stand a reason that a car will be able to drive better than a human?

We got a honk. Hey, cutting me off. That is Tesla's entire bet. The bet that Waymo is making is that no, you can't do that. A car doesn't have the intuition of a human. What it lacks for in feel, it needs to make up for with raw input.

Input that, for instance, can't be hindered

by the weather. The weather is the reason why every single company that's trying to do this starts in Phoenix or Vegas or Austin. Bright and sunny 300 days of the year. Little rain. Wide modern roads. Recently built for booming populations.

Nobody is operating fully autonomous vehicles anywhere outside of the Sun Belt. The magical line that divides the country into the areas with good weather and the areas with bad weather. And by bad weather, we mostly mean snow andice.

The Sun Belt is easy. Waymo has been there for over 5 years. So Waymo has all the cameras. It has microphones that hear emergency vehicles. But it also has this thing, this massive bet on every car. Lidar, which is just a kind of radar.

It shoots signals out as it rotates, listens for the signal to come back, and based on how fast the signal returns, it can get a very accurate idea of what objects are where, how fast they're moving, and in what direction.

It doesn't care if it's sunny. It doesn't use light to see, making it great in the dark. And in theory, when paired to cameras, able to see through all the stuff that might jam up cameras or human eyesight.

These are two obviously different technologies with major trade-offs. 29 different sensors for Waymo, 8 high-definition cameras for Tesla. And this is one of the major reasons why this whole Tesla versus Waymo thing is so interesting. Expensive versus cheap.

More data versus less data. Whoever loses is going to have to figure out how to adopt the other system because whichever of these technologies wins will most likely end up as the winner in a winner-takes-most market. VHS beat out Betamax.

Blu-ray beat out HD DVD. Thomas Edison played this exact game with electricity and lost.

Electricity was just starting to become a household word in the 1890s. Everything was stacked in Edison's favor. Money. He was already famous. The backing of J.P. Morgan. Edison was convinced the way to send electricity from one place to another, like from a power plant to a house, was direct current.

Direct current, or DC, flows in one direction. Despite it requiring the plant generating the electricity to be within one mile of where the electricity needed to get to, Nikola Tesla took the other side of the bet, alternating current.

AC could be generated in one place and pushed hundreds of miles down electric lines. It got dirty. Edison publicly electrocuted a horse in the streets of New York to scare people away from AC. The very first death sentence carried out by an electric chair in the United States, Edison finagled it to be done with AC current for the negative publicity it would create.

Despite that, your world is powered by alternating current electricity. It was obviously better. Edison was marginalized within his own company. The whole DC system that Edison built was thrown out and rebuilt with AC technology. What is today hundreds of billions of dollars.

That little saga and the competition between Waymo and Tesla. A lot of game theory embedded in it. You want the nerdy technical name for this type of game? It is stochastic sequential game with delayed feedback. Tesla is used to have a front-facing radar in the bumper.

Now they don't, which says they actually feel comfortable with fewer sensors. Waymo now has four LiDAR units in each car, but the cost to outfit their cars has gone down, which says Waymo can lower what was a tremendous expense, the cost for all its sensors.

The game evolves to us as observers, one event at a time. Sequential. But time has to pass before we know if a move was correct. Delayed feedback. And it all happens in the noisy, unsterile world of humans. Stochastic.

So where are we now in this game? Starting at the Golden Gate Bridge, each vehicle would drive a short 1.3 miles to the first destination. From there, they would drive to the second destination, 3.4 miles across town. Once complete, turn around and take the longest third leg of the test all the way back to the start.

Could each car simply complete each leg without a human intervening? How long would it take each car to complete each leg and the whole circuit in total? We start with the Tesla. Destination number one starting now. Let's do this.

So we're basically testing two things in these cars. The first is, are there any problems? Right? Can Tesla get all the way there or can it not? Is it going to fail? Waymo's do this every single day. The other thing is, how long does it take?

Is one significantly shorter than the other? Is one a more pleasant ride? This is not the easiest trip in the world. The first one from the Golden Gate Bridge downtown. We have just merged onto a four-lane highway. Somebody honked at us.

I don't think she needed to honk. She was in the wrong. She didn't have her turn signal on. So far, so good. We are halfway there and not a single problem with the Tesla. Granted, it has been all highway driving.

We'll see. Technically, you're supposed to keep your hand on the steering wheel or at least paying attention. I mean, I am paying attention for all those keyboard warriors at home. I got this. 0.3 miles to go. Time check.

7 minutes, 10 seconds. Unless you're actually writing. Go ahead. Go ahead. Nice job, Tesla. That was pretty good. That guy ran a stop sign and Tesla stopped. So that's pretty. That's actually fairly impressive. And we're here. Time check.

08:48. Onto destination number two. Through the thick of the city. Traffic, construction, people. This is where the meat of San Francisco is. This is a 23-minute cross-town trip goingright down through the center of the city. I would suspect that this will be the hardest part of all of this because it's dense.

It's crowded. We're going to see what happens. I wouldn't have done this little challenge if I didn't think there was at least a small chance that Tesla could match Waymo and could get to all these spots without having what we call a critical disengagement.

A critical disengagement. You'll know it if it happens. Sounds go off. There's blinking. There's flashing. The whole screen goes red and it makes the human that's behind the wheel take over the car. I'm 50-50 on whether or not we'll be able to do all three of these stops without having some type of critical disengagement.

Just doing San Francisco stuff. Drinking lattes. Eating chia pudding. Riding in a Tesla. This is what people in San Francisco do all day. We got a honk. I don't know if that was at us. We're more than halfway there.

Mile and a half left to go. Nothing to report so far. One honk. And I don't know if it was at us. Oh, look. I don't know if you can see that over there. That is a Zoox, which I believe is owned by Amazon at this point in time.

It's a self-driving, fully autonomous car. It's really the second brand after Waymo that's trying to do this. You don't see these very often. That's the first one I've ever seen on the streets. And I've driven in San Francisco a decent amount.

Time check. There's a cone.

Uh-oh.

It doesn't know what to do. It doesn't know what to do. Okay. We had one. There. There's a coneright in the middle of the road. Not for anything. It was just a cone knocked over and the Tesla stopped.

Obviously, it's trained to not go over safety cones. That is the exact kind of situation where you would kind of expect to run into a problem. One critical disengagement. The thing with that one is I'm not sure if Waymo would have gotten that either.

That was... It's almost like somebody did that on purpose. I guess we'll see. Waymo's going to take the... We're going to do the exact same thing with Waymo. We'll see if that cone is there and if it decides the same route.

It's hard to say. Time check. I mean, Waymo's are everywhere. You see them constantly. We're here. One critical disengagement. 23 minutes, 54 seconds. Wait, wait. Pause itright here. In the moment, I called this a problem. I said it's a critical disengagement.

When that happens, though, the screen flashes red, beeps. You got to take over. It's like, take the wheelright now. It didn't actually do that. I stepped in before it asked me to. So it's really not fair to call it a critical disengagement.

It might have figured it out. You can see this car coming on the left. It's hard to say.

And we're off. Ride number three in Tesla. All the way back to where we started at Batteries Car Park by the Golden Gate Bridge. 7 miles, 31 minutes. We are in straight-up downtown San Franciscoright now. Oh, hey, cutting me off.

City drivers, am Iright? Waymo. Time check. 18 on the dot. Oh, we have a new style Waymo pulling upright next to us. We've never seen that one before. I think that's one of the Zeekr. Yeah, Zeekr Gheely. Waymo is moving forward.

There's no way this doesn't become a thing in every city. It absolutely works in San Francisco. No reason to think that it won't work in other cities. And because they keep iterating on different car versions. It was the van a while ago, then it was the Jaguar, and now it's the Zeekr.

That just says they're getting better and better at it to me. I would prepare yourself for a world, for cities at least, of autonomous vehicles. But I'll tell you, the Tesla looks better. It drives pretty well. So, allright, just 10 minutes left.

One critical disengagement so far. Get out of my lane. Allright, we're getting pretty close. Okay, we are here. 30 minutes, 30 seconds for this last leg. We had one critical disengagement on trip number two. Pretty much everything else was very smooth.

This actually went better than I thought it was going to. Total time for Tesla, about 63 minutes. One potential but unclear issue. Much smoother than I had expected. What I am really curious about is if we're going to run into that road cone in the Waymo.

When I put in the destinations, I don't select the route. Part of the reason we are monitoring time so closely is to see which vehicle routes more efficiently. So we'll find out. On to the Waymo. I'm a Twitter guy.

That's my social media of choice by far. But man, is it hard to poke through the algorithm? Like, what is the other side saying about a story? I don't even like using the term other side, but you know what I mean.

There's another side of this story and I want to know it. The title, the framing, it matters. So what is the other side of this story? Left,right, center, there it is. All the coverage that exists for this story, for any story, and you can just see how things are framed.

So I've been spending more and more of my time browsing my news through this site, Ground News, that pulls coverage from hundreds of different sources so that I can view each story through a wider lens. Essentially, the Spotify of news.

Homepage to see what's trending. Blind spot section that shows stories that are being reported on dominantly by one side or the other. Stuff you would never see in an algo chamber. And then if you click on one, there's a summary for different viewpoints.

You can often see what podcasts are covering it. Maybe you want to know what a podcast that talks about this kind of stuff. It even tells you whether each source is corporate-controlled, government-controlled, or independently funded. This QR code, lovingly created for viewers of Maxonomics, gets you an exclusive 40% discount off their Vantage plan.

The same one I'm using. Scan it and you're in. Or just go to groundnews.com/maxonomics. $5 a month for unlimited access to everything I've mentioned. Thank you to Ground News for doing that and for helping us to create videos like this.

Now let's get back to these cars that drive themselves. Heading to Lulu Lemon. Here we go. Leg one of the Waymo going from Battery Park by Golden Gate Bridge down to Lulu Lemon. Now this is taking a completely different route than the Tesla took.

Time check. 4 minutes, 22 seconds. When we took the Tesla to Lulu Lemon, it was a much shorter route and it wentright through the middle of the city. Kind of like all the hairy spots way up down the hills.

A little bit more traffic. This is pretty chill where Waymo's taking us. I'm wondering if there's a reason for that, if they do that on purpose. Allright, we're here. So we're about to go on to the next spot.

Let's see what happens. So the first leg took 12 minutes, 23 seconds. That's about 4 minutes slower than the Tesla. Right now, this says it's going to take 23 minutes to get to the next stop at Plow. We're already about 5 minutes into this trip.

Tesla did it in about 24 minutes. So we might be seeing a pattern that this is much slower than a Tesla, which little by little doesn't matter that much, but in aggregate, that would end up mattering quite a bit.

It's more time getting to destinations, less passengers that you can pick up. There's an inefficiency to it. Takes more electricity to run the car. That would be very interesting if Waymo is consistently slower than Tesla, especially by like 20, 25%.

That's significant. I expect exactly zero critical disengagements from Waymo. This is what they do. It would be bad if every third or fourth trip there was one of those problems. Though if there was, you can just hit the support button.

You can talk to somebody and they'll figure it out from some control center somewhere that they take remote control over the car. Get out of the situation. Oh, this is the same road. Oh, it's still there. Allright, let's see.

Let's see. That cone that we ran into in the middle of the road with the Tesla, Waymo's going to run into. It's in the exact same spot. Somebody actually picked it up and turned itright side up. Oh. Oh, it just zippedright around and it just totally ignored the cone.

Interesting. Okay. Okay. Now, to be fair, I think the cone was a little bit more in theright lane, but still, the Tesla should have gone around the left lane once nobody was coming. So, point to Waymo. Time check.

We are here. 24:51. It's not significantly different, but it is a little different.

Third and final ride in the Waymo. We're going back to the Golden Gate Bridge. The thing says it's going to take 40 minutes, which would be 8 minutes longer than the Tesla took. I don't think that this is a traffic thing.

It's the middle of the afternoon. It's about 2:15. So it's not a high traffic time. We'll find out. But if it does take 40 minutes, that's about 15 extra minutes of drive time in a Waymo versus a Tesla.

Not that that's the most important metric, but it is interesting to note. That would be unexpected. Huh? A little hesitation. Zoox, Zoox, Zoox, Waymo. Time check. 15:08. We're back at the Golden Gate Bridge. End of ride number three.

It took an additional 10 minutes. Total time for Waymo, 78 minutes, 59 seconds. This turned into an interesting situation. I was not, A, expecting Tesla to be able to do it so easily and, B, definitely not expecting Waymo to be so far behind in time.

Now I know that this was not the most scientific of tests. If you wanted to make this bulletproof, you would need to run multiple circuits for multiple days at many different times of day. This was daytime. That has to be noted.

Also, the Tesla I used, obviously, it isn't actually a Tesla robotaxi. It's a regular Tesla that has the full self-driving package. Anyone could buy it. And the version of the software is one full version behind what's in the latest full self-driving package and almost two versions behind what Tesla was using in its robotaxi pilot in Austin.

But it's that time difference. A whopping 16 minutes. A 16-minute difference is a big problem for Waymo. If you took this route and ran it 24 hours per day, a Waymo would complete the loop six fewer times than a Tesla.

Tesla would get the riders faster, dropped off quicker. Using less battery per ride would keep Tesla cars driving productive routes instead of wasting time charging. Tesla would be able to cover the same amount of riders with 25% fewer cars, roughly.

I'm sure the math wouldn't be that clean-cut, but Waymo has to make sure Tesla doesn't maintain that kind of time gap over a large sample because let's just assume it costs Tesla and Waymo the same to build and operate one of their cars.

It doesn't, which we'll get into in a second. But if it did, Waymo would take 25% more time than Tesla to pay off each car, all while likely losing riders because riders would choose a quick pickup from Tesla over waiting for a Waymo.

A very difficult game of playing catch-up. Okay. Okay. But let's just assume Waymo figures this out or the test results are wrong. If we're betting on who wins, what's the one thing? The one thing that ultimately decides who will win, who can do the whole thing cheaper, but more narrow.

Who can build their vehicle cheaper? It cost Andrew Carnegie $100 to produce one ton of steel in 1870. Iron ore, coal, transportation. Those are the three main costs of making steel. Get iron ore, burn coal to boil it, cast it into the shape you want, get it to your customers.

The man with his name on universities in the Met, Carnegie, a Scottish immigrant, started with one steel mill in Pittsburgh. The challenge he realized was very obvious. At $100 per ton, he could make steel at the same price as dozens of other steel companies.

He had exactly zero advantage. Buy iron ore from a mine in Michigan. The mine in Michigan takes a cut. Their profit margin. Put the iron ore on a train. The railroad takes a cut. Buy coal. Put it on a train.

Coal mine takes a cut. Railroad takes a cut. Ship steel to customers on a train. Another cut. By 1900, Carnegie owned huge tracks of the Mesabi Range iron deposits. Thousands of coal ovens and massive coal deposits in Pennsylvania.

600 miles of railroad and seven cutting-edge steel mills. No more cuts. His new price for a ton of steel, $11.50. Everyone else's price was $25.

That is the magic of vertical integration. Assuming you have enough money to build it all out because it is not cheap to do. But if you can, not only do you stop paying the profit margin for every supplier you buy from, you have control.

Tweaks, major changes, whatever you need to do isn't a negotiation with the supplier. Only 35% of the money spent building a Tesla goes to outside suppliers. It's about 60% for a company like Ford. A new mid-tier Ford Bronco without all the bells and whistles costs as much as a Tesla Model Y, a true luxury vehicle.

The Bronco, to just get kind of close to the same in features, is $8,000 more than a Model Y. Tesla is about the purest form of vertical integration the US has seen in manufacturing since maybe the 1950s. Waymo is going the completely opposite route, making cars is way out of their core competency.

Business schools speak for what are you best at? No, no. They will buy cars from car companies and outfit them with sensors. The cars are not purpose-built for Waymo. You've all heard of Android,right? Write the Android operating system, make it care as little about the hardware it's running on, and then let those that make hardware figure out how to get the software to work on it.

That's how Android works. Same playbook for Windows. Any manufacturer like Dell can buy a license for Windows and then cobble together a bunch of computer parts and voilà, laptop. No one can build a phone that runs iOS except Apple.

Now, we can nitpick about whether we prefer iOS or Android if you want to start that fight. However, the glaring difference between phones and driverless vehicles, the cars can't be buggy. There just can be no failures, no blue screen of death.

Android phones and Windows laptops were, and some would argue still are, notoriously unstable, especially compared to their Apple counterparts. And that is simply because the hardware is not purpose-built for the software. Androids run on literally thousands of different phones.

iOS runs on one. The Jaguar I-Pace that Waymo is usingright now was not designed to be autonomous. The thing mounted on the roof was not on the Jaguar's design board. So anytime Waymo wants to make a change, they have to base it around what their suppliers will agree to and then pay that supplier a profit margin.

Waymo's current cost to build a vehicle, the lowest estimates are $100,000 per car. Tesla can build a Model Y for $38,000. Waymo knows this is a problem. They've shaved the cost of one car down from about $120,000 three years ago.

But that gap is huge, especially when coupled with the next leg of building cars. How many can you build in a year? In a previous video on Waymo, we found that 300,000 vehicles could probably cover most demand for ride-sharing in the 30 largest US cities.

Across the four factories it owns, Tesla can produce 900,000 cars per year in the US and 2.35 million globally. So if Tesla really does find that it can get to regulation-satisfying full self-driving, it could direct just 20% of its factory output to making robotaxis flood the market, shorten the time passengers have to wait to get picked up, offer more longer routes.

Waymo's solution, it's the Android way. Partner with a bunch of companies: Chrysler, Jaguar, Hyundai, and the latest and most controversial is Zeekr. Together, all of these companies can compete with Tesla's factories. Individually, they all have to be negotiated, designed, tested, and then trying to find a very low-cost manufacturer that is small enough that Google can dictate terms to them.

They've taken on a different type of risk, geopolitical risk. Zeekr is a Chinese company. It used to be listed on the New York Stock Exchange until Senator Rick Scott started questioning whether it should be allowed on the exchange and questioning the national security implications of allowing a company with direct ties to the CCP being the foundation that powers potentially the premier robotaxi network in the US.

Instead of offering compliance and reassurance, Zeekr chose to delist from the New York Stock Exchange and relist in Hong Kong. So it's not that this partnership won't or can't work. It might. It would make sense for Waymo. They need a low-cost car producer.

But as we're going through this whole build-a-car thing, Waymo is playing whack-a-mole with problem after problem after problem. Tesla can build cars. Can they do the robotaxi part? Waymo can do the robotaxi part. Can they build the cars?

It is a race for the ages. Even Apple, who has tight control over its supply chain, doesn't own any manufacturing facility. So really, what's being tested between Tesla and Waymo? The system that has dominated the last three or four decades.

Design a product, write the software for it, make it somewhere else. Can that system compete against the company that changed "make it somewhere else" to "make it in-house"?