Oracle had earnings yesterday and it was an absolute blowout. The stock's up 10%. The numbers...
The god candle. I like that.
It's still way down from the crazy highs of last September. So, it was a $470 billion company now, not bad. It's not quite the almost $1 trillion company it was last year. But they're still up over the last 12 months. So, they've sort of ridden this crazy curve up, but the stock basically doubled and then sold off by half, but it's ticking back up. And the main thing is that, like, you know, was it overhyped, underhyped? I don't know. But this earnings was important because a lot of people were worried about the CapEx, the infrastructure build-out, was there going to be demand? What was the timing of...
Yeah, the timing was the big thing. Because when you had that initial announcement around the RPO last year, the criticism was, hey, you're basically going to try to build AWS in like two years.
Yeah. And no one, like, I mean, obviously the market at the time was quite excited about it, but there was some concerns around just how aggressive the timeline was. So, seeing them hit their timelines at this stage is super important.
Yeah, it was like 500 billion of RPO, which is like, that's AWS-size numbers, and that's a lot of what I think is going on here. That's actually the dynamic, like the market is demanding another AWS. And so, let's go through the top-line numbers first. So, analysts estimated 86.7 billion on revenue. Oracle said that they will hit 90 billion for fiscal year beginning in June.
Now, people don't care as much about the top-line number. Everyone's obsessed with the infrastructure business. So, the previous quarter showed growth of 68%. That's not bad. Analysts said this quarter we're looking for 79%, and Oracle delivered 84%. So they beat estimates, that's why the stock popped, of course. The infrastructure business is particularly important for two key reasons. So, first, Amazon, Microsoft and Google, like, there's more than just them when it comes to hyperscalers, but they are unique among the hyperscalers in that they have AWS, Azure and GCP. They have true cloud platforms, cloud infrastructure platforms. OpenAI does not yet. Meta does not yet. Meta is a hyperscaler, like they build huge data centers, but they don't have this flexibility that comes with operating something like AWS. And so who's signed big deals with Oracle? It's OpenAI and Meta. And fortunately those deals are going better than people expected. People were worried and people were saying, is Larry going to get caught holding the bag? Well, it looks like things are penciling out so far. So Meta and OpenAI both have massive ambitions around AI. They both signed on with Oracle to ramp up data center capacity. Oracle is just this extra burst of capex capacity in the system and they're ramping up. So 50 billion in capex in the current fiscal year and they're consistently outpacing analyst estimates. In the fiscal third quarter, analysts predicted 14 billion in capex, Oracle spent 18.5. So Larry is certainly opening the pocketbook, digging for coins in the couch cushions, but we'll get into the long-term implications of like what does this mean for becoming over-leveraged, debt, you know, drawing down on cash flow. There's actually a lot of interesting structure going on within the financials that shows you how even though he's definitely risk-on, definitely AGI-pilled, all full tilt ahead into the build-out, it doesn't feel like financial recklessness because of the structure of the deal. So, the second reason is that that infrastructure is so important right now that I think people on the outside are sort of missing is that compute is still growing exponentially, but AI adoption curves often look like S-curves. And so the classic one is consumer LLM usage. So OpenAI came out with ChatGPT and the numbers were just like insane. It was like 20 million users overnight and then 100 million and then 200 million and it was just like, okay, this curve is going completely vertical, we're going exponential. But of course there's only 8 billion people on earth and a lot of them just don't use really computers apparently because Meta over two decades has only gotten three and a half billion users to use like the whole suite of apps including everything. So just WhatsApp, you can just be a DAU of Instagram and you count in that. Like we all, everyone in this room counts in Meta's DAU for sure, even if you're like, I'm not really that big on Facebook or I don't use WhatsApp that often. Like you definitely, they got you, except for half the human population which is not on Meta platforms for a variety of reasons, some geopolitics, some economic. But basically there's going to be a slowdown. So OpenAI shot up to basically a billion MAU, monthly active users. The official number that's getting trotted around right now is 920 weekly active users, but everyone behind the scenes says like, yeah, they're well past a billion MAU. And so that curve is slowing down. Like there's just no way you can be like, yeah, actually, like we're expecting 10 billion ChatGPT users next year because like there aren't 10 billion people. So there's going to be deceleration in consumer AI usage.
Sufficiently aligned superintelligence might want to stimulate the growth of human population.
Good take, good take. Yes, yes. You go to ChatGPT and it says like, 'Hey, I know you've been thinking about having a fourth kid. You should do it and here's why.'
Maybe, but at least for now, we are seeing deceleration. And I think for just a lot of people, they're like, 'Yeah, I use AI. I have an app on my phone, maybe Gemini, maybe Claude, maybe OpenAI, maybe ChatGPT. I use it, maybe I use Grok, maybe you see it vendored into different systems. I interact with it on Instagram, but like I'm in, but I'm not like blown away by the growth of this thing because everyone started using this a year ago and we're still all just using it.' But compute is scaling very differently because when we went from LLMs and ChatGPT to reasoning models, that probably 10x'd the amount of tokens that people were generating with o1. And then once GPT-5 came out, the reasoning models became much higher usage rates. And then the agents framework, the OpenClaws, the Codexes, and the Claude Codes, that all did another 10x in token volume. 10x is like a very rough number, but it's basically growing exponentially. So you have these double exponentials is what I'm talking about.
