Loading summary
Transcript41 lines
- [00:00]
A
What it seems to me has happened is that the world's largest ever private funding round just closed for a company that loses 14, that lost $14 billion this year, is projecting to lose another 143 through 2029. It just shut down Sora, its flagship non chatbot consumer product product because it only ever made 2.1 million. And one third of the money coming in this round literally must be repaid by next year.
- [00:24]
B
Yes.
- [00:25]
A
Okay, cool. Is a lot riding on this?
- [00:27]
B
Yes.
- [00:28]
A
Oh shit.
- [00:29]
C
Would it be the entire or less?
- [00:33]
A
Is it bad if that doesn't work?
- [00:35]
B
Well, that's the thing though. Like this is like being. The markets are riding on this, but the economy isn't. And if OpenAI shits its pants, if so, if SoftBank dies, that could really be bad for the Japanese economy.
- [00:48]
C
You know, the bank part, the soft part's fine.
- [00:51]
B
Yeah, the soft parts, that's just their economics. It's just when I think about this stuff, I feel crazy because when I try and talk to journalists about it, they're like, yeah, it's fine, mate. Every time. Like, it's fine, mate. Yeah, they'll work it out. Amazon Web Services lost money. No, it didn't. Not the same thing.
- [01:05]
A
They say Uber lost money as well. Yeah, but also the. What I also think is like, but Uber had a use case also Uber
- [01:12]
B
didn't lose anywhere near as much money and Uber didn't have to. Uber's business model is the same as it was at the beginning. They're just not charging $5 a ride. Users were. They didn't have a subscription where you paid $200 a month and could get two and a half thousand dollars worth of rides. Like Anthropic's doing rate limits during this is so funny as well. They're doing rate limits during peak hours, but those peak hours are 5am to 11pm so just like the entire day. As long as you don't use this during the day or the early morning, you are fine.
- [01:45]
A
I also noted that their peak hours don't move by time zone.
- [01:49]
B
No.
- [01:49]
A
So it's all just. It's Pacific time.
- [01:52]
C
So if you're listening, Freaks in San Francisco. Incredible.
- [01:55]
A
It's freaks in San Francisco and then normal people elsewh.
- [01:58]
C
It would be really, really funny if Uber had done this right back when Uber had launched. If they had. Because I know they were operating sort of like at a loss, you know, in order to like undercut taxis and stuff. But if they were operating at this much of a loss, if they were like, okay, Cool. It's this new app, it's called Uber. We give you a Maserati, you can keep it.
- [02:18]
B
Really? It really is like that, though. It's really insane to me.
- [02:23]
A
What you're describing, Nova, is you're describing MoviePass, which is a famous debacle.
- [02:27]
C
Oh, I didn't know. This is going to be a famous debac. That's not good.
- [02:30]
A
Yeah. So MoviePass tried to do the same thing as Uber, where. Because, like, Uber, like, again, what they did was a regulatory play where they undermined taxicab regulations and just sort of replaced licensed, metro municipally regulated cabs in every city. That's the other thing they had. Yes, they had a product, but also they had a value proposition. The value proposition was evil, but they had one. MoviePass was just like, okay, were you subscribed to us? And we'll buy you tickets for every movie that you want to go to. You can just go to any movie.
- [02:59]
C
That's a pretty good deal. Do they still, like, honor that or.
- [03:03]
A
No, I was a fiasco and they went out of business. Yeah, but they weren't backed by every single venture capital and asset manager in the world. That's the difference. Okay, but Open's counterargument to this, and we're going to sort of go through this in a bit of a structure, is we have a compounding effect. Better infrastructure and better models lower the cost of delivery, will improve products and deeper enterprise deployment, increase the revenue unit per compute as utilization increases and the platform matures. This drives meaningful operating leverage over time. So what are you going to write on now that you've been proven wrong?
- [03:37]
B
Golf, if that's something that I would say. If I was in a car accident and I got out and someone asked me to describe economics, I'd be like, yeah, leverage the AI, it's going to be a compounding fighting.
- [03:50]
D
Yeah, you know, he had a lot of the language models. You know, they're large now, but it gets smaller and you get the good pews. And Sam Altman, he's. He's a little boy and, you know, you give the money to the little boy and then he makes the, you know, the. Back in 1942, when OpenAI, when I was the CEO of OpenAI, we made motor cars.
- [04:14]
B
Yeah, exactly like that.
- [04:15]
A
You can slip into Biden for longer than most people. I know.
- [04:19]
B
For longer than Joe Biden can.
- [04:22]
A
No, but, but OpenAI's argument here is, well, our economics don't matter because as our product gets better, yeah, it's really good Wait, wait, wait.
- [04:31]
B
Sorry, take a step back. It really is though. It's like, hey, don't worry about the economics, right Be.
- [04:38]
A
That's why their argument is the economics as they appear now don't matter because there's a kind of magical force that's going to make them better, which is every time we make a better model lowers the cost of delivery, which by the way isn't true because as it gets cheaper, people use it for more.
- [04:52]
B
So they, they haven't done that.
- [04:54]
A
Yeah, they also didn't do that.
- [04:55]
B
They haven't done that. They haven't made them cheap. They're more expensive. Like they, they may change the rate of the tokens they offer. They, the price of the tokens. That doesn't mean it's cheaper to run.
- [05:05]
A
And also they say, well, improved products and deeper enterprise deployment increase the revenue per unit of compute. And we're going to talk about that towards the end. So just put a little. That's the sound of putting a pin in that idea. So they go on in their announcement of this gigantic funding round. Moments like this do not come often, thank goodness. In past generations, capital markets helped build the systems that defy modern economies. From electricity to highways to the Internet. This is that moment again. The capital being deployed today is helping build the infrastructure layer for intelligence itself. And over time that value will flow back into the economy, to communities and increasingly to individuals. Flow back is the new trickle down. I think this is if you just replace trickle down, that's just trickle down economic backwash economics.
- [05:44]
C
We're all going to get clawed.
- [05:46]
A
Backwash, yeah. Delicious. I can't wait to contract meningitis. But also what I want to know is like, like the capital being deployed today is helping build the infrastructure layer for intelligence itself. I don't think that's anything. I don't think that means anything. Infrastructure layer for intelligence itself sounds to me like nothing.
- [06:03]
B
In opinion it doesn't mean anything. None of this means anything. The data centers that OpenAI is meant to be building. It's really funny. Sam Altman proudly actually going to pull this up because it's fairly recent. If I was the CEO of OpenAI and things were this delayed, I would not quote, post this and say the first steel beams went up this week at our Michigan Stargate site with Oracle and related digital. That means that datacenter will not be built. I don't think it'll be built ever. But before 2030 they just like, they don't have an answer because if they did, they give us it because if they had an answer, they'd just be able to be like, yeah, this Generally, I find when someone has an answer, it's simple, it's straightforward. You can just be like, yeah, what's going to happen is we have a plan and we are going to do this. And they don't even bother to do it. They don't even have the honesty to be like, yeah, we're kind of working through it. They don't even have a kind of vague plan where they're like, oh yeah, well, this specific efficiency gain that we're working on today will do it. They're just like, you know, GPUs and all that, you know, efficient better.
- [07:09]
A
And then, yeah, don't worry about, hey, another 152 billion. But I also want to talk about right there's this one idea underpinning these entire recent article you wrote about in 2024, the The Pale Horses of the Apocalypse. They've updated now with some tests of some of the, of the predictions. And in my, in my view, all of these horsemen are elaborations on this theme, which is how to make a dollar look like 10 or 100 and then abscond with as much of it before someone asks for any of so you talk about the, the comparisons with the subprime mortgage crisis, which I think is worth going into here, which you've been thinking about. But subprime mortgage era, the whole trick was the same, which is how do you make $1 seem like 10 or 100? It's the same thing as well with the cryptocurrency Web3 and DeFi and all this. How do you make $1 seem like more by valuing not just the stuff and the things that are useful to people, the goods and services, but like the transactions themselves and expected value of nested bets that are all sort of against one another. And the simple brute fact is in the subprime era, people's mortgages had their houses valued at many multiples of their worth, and hedging that risk required increased complexity and obfuscation. That's how you get CEOs as we get various celebrities explaining tranches to us in a movie or me in a podcast. For some reason, my understanding is in the air equivalent. This is done by the combination of subsidized pricing. It's like you mentioned, anthropic starting doing rate limiting, which basically means we are Uber. We've been giving you a Maserati, but we have to say you can only drive the Maserati at two in the morning. Yes, to specifically disguise the cost of this thing that by everyone's reckoning is essential to grow in the economy by embedding in everything and making companies seem like the center of a lot more actual economic activity, providing something that actually has an end consumer by throwing the same billion dollars back and forth between them forever and ever. So far, so familiar. Is that the right comparison at a high level?
- [08:50]
B
My way of putting it is a bit different. Like you're mostly on the right track. But the way I put it is this, which is the subprime mortgage crisis was caused by a facade of sorts in that the housing market was artificially inflated by the easy availability of money. That was the big thing. It was housing was not worth what it was worth, other than in the existence of unlimited resources. If anyone could get a mortgage, of course houses could keep being sold. And indeed, people built their lives around and took on these mortgages they either didn't understand or thought they could get out of, using very flimsy mythologies such as I will be able to refinance. Housing value will only ever go up. And by the way, if you look at the news from around the time, that's basically what people say.