00:00Imagine a number, then treble it. You know, that's where I'm at right now. I just can't get my head around it all, Mandeep.
00:05How are these companies going to spend $650 billion this year? What aren't?
00:10Yeah, look, I mean, I think everyone highlighted supply constraints and the fact that they didn't have enough capacity.
00:18Otherwise, they would have grown faster. So that sort of justifies the increase in CapEx.
00:23But you're right. I think the kind of increases we are seeing, probably Google is the only one who has earned it because their cloud revenue grew 48 percent and the infrastructure side actually was north of 50 percent.
00:37And they're doing it at, you know, incremental margins of at least 30, 40 percent compared to an Amazon where the margins are going down.
00:46And yes, the growth did accelerate, but it's coming at the expense of margin.
00:51And so to my mind, you know, there is this rush going on in terms of securing supply at TSMC.
00:58And the bigger you are, I think the chances are you have a better shot at securing supply.
01:03But if you add up all the CapEx, which you have on the screen, TSMC cannot make as much in terms of, you know, the needs that these companies are highlighting with their CapEx numbers.
01:16So they have to significantly ramp their own CapEx, you know, the semi-CapEx equipment purchases to be able to service this kind of demand that they're seeing from the hyperscaler.
01:27So it'll be interesting to see what NVIDIA guides to because NVIDIA hasn't moved.
01:32I mean, NVIDIA should be up, right, with all the CapEx increases.
01:35So there's something that at least limits right now in terms of what kind of upside NVIDIA may end up seeing because of these CapEx increases.
01:45So, Mandeep, here's my question for you.
01:46What do you think we hit first, serious supply constraints or the limits of investor tolerance?
01:51I think we are already seeing some sort of investor panic in terms of the free cash flow expectations for this year and beyond if this rate of CapEx continues.
02:04Because for public cloud businesses, the baseline CapEx was more around $40, $50 billion.
02:11Now we are three to four times that.
02:13If that doesn't translate into revenue at the same pace, and yes, you have to invest up front to bring the capacity online, but at some point you have to ask yourself, if I'm doing it, you know, at the expense of margins and free cash flow, how long do I sustain?
02:30And that's where the questions were asked for Oracle.
02:33I mean, there was a pushback with Oracle because everything was tied to OpenAI.
02:38I don't know who are going to be Amazon's customers for the $200 billion in CapEx.
02:44Is it all inferencing or is there a big training component?
02:48Because inferencing workloads are the only one that translate into cloud revenue.
02:53Training is more of a sunk cost.
02:54Yes, you're going to get a better model, but you're not going to be able to monetize it right away.
02:59So all these questions, I think, will become clearer as we go through probably the next two, three months.
Comments