Very True by Verissimo

Alex (00:00)
Welcome back to another solo episode of Very True by Verissimo. Today we're going to talk about a super hot topic, which is the business model of AI. And if you've read my recent Substack or seen my recent LinkedIn posts, this is where I'm spending a lot of my time thinking, reading, opining, analyzing these days. So a certain idea came to mind, and that's what today's episode is about. So I hope you all enjoy.

Right now, almost every company raising money on AI is running a playbook. It's a good playbook. It worked once already spectacularly, and Uber wrote it. Raise more money than your competitors can dream of, lose it on purpose, change how people behave. Once you own their behavior, turn the prices back up and collect. But there's a piece of that playbook that everyone is quoting and nobody is copying. It's the piece that actually made it work. And I think

It's the difference between a subsidy that's an investment and a subsidy that's a prayer. So today I want to take Uber apart, the whole thing on one page, and then I want to hold AI up next to it and show you exactly which piece is missing.

Let me start with Uber the business, not Uber the story, because the business is simple. And that simplicity is the entire point.

Here's the whole business on one slide. I'll link the slide in the show notes so you can all see it. But basically, in a marketplace, your unit isn't one customer. It's the smallest balanced piece of the market that actually exchanges value. In this example, that's one seller and four buyers, because that's the ratio at which the thing clears.

That one seller moves four items per month. Say each item is $100 and the platform takes 10%. So this one little unit throws off $40 a month of contribution. Now, what did it cost to build that unit? Say $100 to acquire the seller and $25 for each of the four buyers. That's $200 all in. $40 a month against $200 to build it, you're paid back in five months.

After that, the unit just prints. now hold that number in your head five months because here's what actually matters. It's a number. You can write it down. You can argue about whether it's really four or seven. You can try to push the take rate up or the acquisition cost down. But the shape of the thing is sitting right there in front of you. That is what an understood business looks like.

Not optimized, understood. Now I realize that this might not be all that simple for people, but the contrast here is that in the vast majority of software businesses, the unit is just a customer. And we define the unit, like I mentioned, as where the cost of sales and the value of those sales meet at a least common denominator. So in a marketplace, it's not super intuitive. I give props to

To my good friend Amit Mukherjee, who I worked with at NEA for a few years, who actually explained this to me many, many years ago. But the main point here is that this model is understood. They figured this out, that this is the machine. You got a load balance between supply side and demand side. You have to figure out what that exchange is, how it changes over time. And then you got to figure out how long it's going to take you to basically pay that back to get it to Unit Economics Positive.

It's a multidimensional problem, but it's manageable, it's knowable, it's understandable.

You can also do something really important, which is exactly what Uber did, which is light the whole thing on fire. Not the actual understanding and the mechanics of it, but you can sink the ship on purpose, basically. So we'll get into that now. if every one of those units pays you back in five months and then prints forever, the strategy writes itself. You don't sit around optimizing. You go shopping. You buy as many units as you possibly can and as many cities as you can.

Before anyone else gets there.

I don't know if anyone remembers, but 10 plus years ago, there were billboards up in San Francisco where Uber was offering to pay $5,000 for a referral fee for a single driver. And this gets back to the idea that you need to load balance. Everyone knows that consumer behavior is king. And if you open the Uber app and there's no driver, then you close the Uber app. And that's a huge problem. And the same goes for the supply side. If you're a driver and you open the Uber app,

And nobody's asking for a ride, you close it. And obviously, there are other competitors at play here and other options. But when you're trying to affect behavior change, load balancing is important. And so is building the entire product and operation around that. But let's come back to this idea of the fuel and how we basically, like I said, sink the ship, light it on fire, whatever you want. It's a race, right? And they figured this out early, and you can win this race with a balance sheet.

whoever can lose the most for the longest is gonna win this game. And that's exactly what Uber did. I'll just run through a little bit of stats on Uber's fundraising history in case people forgot. Uber raised their first angel round in August of 2009. They raised seed in October of 2010, series A in February 11, series B in September 2012.

Series C August 13, Series D, June 14. The first mega round was actually the Series C at $361 million. By the Series D, they were raising over a billion. Series E in 2015 was $4.5 billion across multiple tranches. Series F another billion. Series G, another $3.5 billion. An extension of the series G for another quarter billion. And then another billion dollars of it. This is all equity capital.

Right. So Uber raised in equity capital.

like $20 billion, which now again, when we're comparing that to what you know, Anthropic raised in their last round, sounds like nothing, but that's my whole point and we'll get to it. they also raised a bunch of debt starting in 2015. They raised several billion dollars of of convertible notes and leverage loans and term loans and high-yield bonds, and then they continue to.

Play the debt markets and equity markets after going public in 2019. So by the way, they raised over $20 billion of equity capital in 10 years, which now seems like, whatever, but we'll see. That's actually my whole point. and a little bit on the financials, when Uber went public in 2019, their 2019 revenue was $13 billion. Their gross profit was around 53%. percent

And their operating income was negative eight and a half billion dollars. So was their their net income.

their adjusted EBITDA though was negative $2.7 billion when they went public. So in 2019, obviously they did their IPO, they raised a bunch more, capital. They began the year with 8.2 billion in the bank, they finished the year with 12.1 billion.

in the bank. but the interesting stats here are actually the non-gaap metrics, which are trips, and gross bookings are like two of their key non-gaap metrics that they disclose from you know IPO until until today in their filings. And in 2019, they did 6.9 billion trips for total gross bookings of 65 billion dollars.

So the bookings per trip, which is a basically our ARPU, was $9.41. And again, that resulted in an adjusted EBITDA of negative $2.7 billion. But again, this was a land grab, right? This is this is what everyone else is positing. Like, all right, we got to just own this market. But going back 10 plus years, people understood the unit economics here. Like this was there was an equation of what is this business model.

It's load balancing between supply side and demand side. It's figuring out our take rate and how we basically can increase our take by either increasing our bookings per trip, our number of trips, and or our

ARPU, which is that that ⁓ basically bookings per trip, which is ends up being the average price. so I think we all kind of know what happened next, and we all felt it as consumers. you actually lived through this, right? Which was that Uber got more expensive, right? They stopped paying you to use it.

Right. You were addicted. This is like the oldest business model in the game. I'm not gonna make all the references, but you get your customers addicted by subsidizing. And this is they they ran as much of a really smart business model strategy here as they did a awareness of the capital markets in which they existed. And that's again, I'm not gonna take anything away from that because it's one of the most impressive business stories of all time. but they actually changed behavior. I I remember like ⁓

Friend asked me once, I was investing in a company recently, and he was like, I don't get it. Like, do people actually buy that stuff that they're selling online? And I was like, No, that's the whole point. behavior change is the biggest opportunity in the world. Same with Airbnb. Like, do people really do that much couch surfing and like vacation rentals? It's like, no, but they will. Right. I said, nobody used their phone to hail a cab, and now that's what everyone wants to do all the time.

And this is a good thing. I think we and anyone who's used Uber, yes, there are complaints, but like compare it to trying to get a taxi. You know, I remember when surge pricing came out, people complained about it, but surge pricing is the best. in Israel, we have Gett and there is no surge pricing. And as soon as it rains, you basically just can't get a cab because they don't want to like actually drive somewhere to pick you up because they think they can just get a ride immediately because demand spikes. So

Why are why were we living in this socialist world where like demand spikes and we don't have a price increase despite supply staying the same? that's like ⁓ pretty simple. So I actually went through old school banking style and built a spreadsheet of Uber's metrics, some of which I just discussed going back to 2016, all the way through their 2025 fiscal year. And I have most of them here.

I'll just kind of give the headline metrics. but basically revenue in 2016 was 3.3 billion, revenue in 2025 was 52 billion. That's a 16x in 10 years. Pretty impressive.

Their gross margin, this is actually really interesting, went down from 54% in 2018 to 40% in 2025. Now, I'm not gonna focus too much on this. I think that's driven very significantly by the transition to to deliveries

Just to give an idea of gross bookings.

this is a quarterly number, but 2019 Q 1 they did $11 billion in gross bookings for mobility. For delivery, they did $3,071 and for freight they did $128M So that's you know our total of

14 billion 645 That's Q 1 of '19 percentage wise, that is 78% percent mobility, 21 % delivery, and less than 1% freight. So that just sets the stage for where they were in 2019, Q 1

Okay, we'll compare that to Q 4 2025 which we have the data right here. So mobility, they did

27, 442. Delivery they did 25, 431, and freight was 1,267

It

Became 51% mobility, 47% delivery, and 2% freight. that shift to delivery, those numbers are much lower. the ARPU is much, much lower on those compared to, like Uber Eats and Postmates versus actually just calling an Uber.

What's really interesting about this, if we look at the high-level number of bookings per trip, so like the dollar per trip and how that's changed over time. in 2019, it was $9.41, like I said. 2020 was $11.52. 2021, it was $14.20. 2022, $15.10.

And and then '23 through current is kind of leveled out at $14.50

I don't think they split them apart, which is really interesting. So you can't see the difference in price, but I think we can kind of all imagine. if you've ordered on Postmates or you've ordered on Uber, you generally know how much it's charging you. It's like a a few dollars, right? It's four or five dollars. Maybe it's ten dollars if it's like far. When you're getting in an Uber and you're going to the airport, it's thirty dollars, it's eighty dollars, i you know.

We've seen these price increases. And so as that shift has happened from 21% delivery to 47% delivery, that's actually driven the numbers down in terms of the ARPU. And that's actually also driven the gross margin down,

So just to kind of complete that circle, the result is that in 2020, Uber lost $2.5 billion EBITDA, negative 2.5 billion EBITDA. 2021, they had already started really turning things around and it was negative 774 million in EBITDA. By 2022,

They're at $1.7 billion of positive EBITDA. And in 2025, they did $8.7 billion in EBITDA. The reason why I'm using EBITDA instead of another metric is a whole different story. interestingly enough, they did not in the same like non-gaap table, they started disclosing free cash flow. And they didn't disclose it before when there was no free cash flow, right? If it's negative free cash flow, it's a little bit of a misnomer. doesn't really make sense. but in 2022, they had 390 million of free cash flow.

And in 2025, 9.8 billion of free cash flow. So this is all these numbers, you know, take from it what you will. But the point is they had a machine in place, right? This this unit economic machine where you've got this load balancing system of supply side and demand side. You have a number of transactions, you have a size of transaction, you have a take rate from those transactions, you have CACs for supply side and demand side. That all

Goes into a system, into a machine that tells marketing that they can pay $5,000 to acquire a new driver. And out comes an answer of Unit Economics of here's how here's what the cost of acquisition is. Here's what it pays us on a monthly basis. Here's how many months we need to keep this customer for it to matter. They took advantage of a very frothy financing environment and achieved what we'll call world domination.

Global brand recognition, mass behavior change, incredible. Like the the all the like best things that a startup can do. And they achieved it and they did it. It's those two things together that matter so much. It's that they brought these two ideas together. They took advantage of the capital environment, but it was layered on top of a business model that had its finger on the pulse. It was, it was measured. Every part of this was measured. Maybe I should

Go find a friend of mine who worked at Uber on the finance team in like the mid-stages and see to what extent they were really optimizing this and saying, hey, if we can recover CAC in two years, that's fine on this cohort or on this unit. because we have the money. Like, let's spend it. So let's get back to AI and back to that missing piece. Every AI company right now has the fuel, whether it's a lab, an application company,

They have more capital than Uber ever dreamed of, They've got the ambition, They have the domination, they have the behavior change, they have the vision, they have the distribution, you name it. And they've got the losses, like big time losses. If you, you know, I I always reference him because I'm a huge fan, but Cal Newport talks about this in his AI reality check, which I recommend for everyone.

They're talking about their revenue numbers, they're up, down, all over the place. They can't really nail it down. But we know how much they're burning. and we also know what that's doing to NVIDIA's stock price and their earnings and their backlog and everything. But what we don't have is that unity economics clarity, right? Like nobody can put the AI business on one page.

In a nice diagram in the way that I can do it for Uber and really for pretty much any marketplace. It's like it's it's not because they're lazy or stupid. You could argue these people are smarter than everyone else. They obviously have more tools, more instrumentation. It's because the numbers haven't settled. It's because we don't know this nice unit economic framework that we can apply. We don't understand the machine. They don't understand the machine. Forget what I understand. They don't understand the machine, right? Like they're they're coming up with it as they go.

You know, you know they're ideating on it. Like if you just look at like the names and titles of jobs and the backgrounds they're hiring at these big, you know, open AI, Anthropic, Google, right? They're hiring ⁓ teams of implementation consultants, like you name it. They're trying to figure it out and they're trying to hit multiple things, but they've raised so much more and they've burned so much more and they've worked so much harder on this like global domination, behavior change, brand recognition than Uber did.

For the level of figured out that the unit economics are. So, like the point is that people, what is the unit in an AI company? Is it just a user or is it an agent? Or is it a company? Or is it a query? Or is it a token? Like, is it a question answered, a lead delivered, is it results? Like everyone's talking about it, but no one knows. And

The cost to deliver any of those units is falling off a cliff while at the same time it's actually increasing, right? Like as these models get more complicated, like the paces are all moving differently in terms of what we can develop, how valuable it is, what is the hardware that we need, what's the power we need. All these are they're all in motion, right? We don't, we know they're all contributing factors, they're all in motion. And so it sounds great that like prices are coming down, but

You gotta also realize that the bar to stay in the game is is rising ridiculously fast. So the expensive thing that people invested a lot of time and money into build, like, does it really have staying power? this is where it just totally diverges from consumer behavior with lock-in, brand recognition. I myself, I started with ChatGPT, using it for a while, Gemini 3 came out.

Love Gemini 3. I'm all Google professional and personal. So like integrations that were great. And then all of a sudden, Opus came out and it was like a whole different ball game. I switched to Claude. also Gemini kind of seemed to have a lot of bugs for a while. They seem to have worked that out. but in either case, the switching costs are super low. Now, if you're building into enterprise tools, that's different. Obviously, me as a consumer using this stuff for professional and personal reasons.

Is different. You know, planning planning my next run and and meal and training session is not exactly the most valuable thing that AI can do. It is really, really good at it though. And I'll probably do another podcast on on how I use it and what I've gained from it. But the point is that we don't know what the unit is. We don't know how to price it. The business model is still being searched for. And to get this far and say we'll figure it out down the road.

Is a really tenuous place to be sitting with how much money's been burned and how much revenue has been earned. Because that underlying plan is is just not baked, right? So when people tell you that the AI unit economics just aren't optimized yet, that's the wrong word, Uber wasn't optimized, but it was understood. So you could draw the destination.

And then spend 10 years and $30 billion moving towards it with a clear strategy. AI is not just unoptimized, it's not understood. And the destination keeps moving. I'm not blaming anyone who's working at an AI company. It's a really hard problem to solve because it's changing and moving faster than anything we've ever seen, I think, in human history. So here's what I'll leave you with. Uber subsidized toward a place that it could see.

AI is subsidizing toward a place that it can't even draw yet. The first one is an investment. The second one, until someone draws the actual slide that outlines what the unit economics and the business model are here, is just a bet. Find me that slide, and that's the whole game. I'm working on it myself. I'm a huge nerd. I did this with with SaaS. I figured out.

Twelve years ago, that calculus actually underpins the entire relationship between ARR and recognized revenue and all the costs and how they align is just one big calculus problem. And as a mechanical engineer, that was actually very intuitive to me. I've talked before and I've actually just written again and I've written before about why the SaaS game fell apart. and it's simple competition, supply and demand, that's the whole thing the whole thing.

For AI, I'm I'm trying to, I'm cooking. I'm cooking. I'm nerding out hard. I'm going back to my textbooks from mechanical engineering and figuring out what physical phenomenon we've observed and can draw towards. you know, understanding the past, not just reading it, but understanding the mechanism that we can use to describe it is how we actually build a model that we can pressure test things on in the future. instead of just wasting a bunch of money figuring it out.

feeling our way through the darkness, we can actually start to build a business model in the same way you would build a CAD model for a car or a bike or anything else for that matter and figure out what can actually work. So I'm excited to continue to work on this, continue to discuss it. you know, Uber is just one great example because it's a great company and they did this extremely well. You know, Airbnb is is right there with them. You could argue even Tesla, also very, very similar in terms of what they've done.

So I hope this was interesting and remember to stay true.