Nvidia's financing schemes and declining token growth second derivative signal a potential top; stock drops after earnings despite beats.

So, let's be clear, like he sounds really good, and I I respect Zach for what he's done and what he does. All this sounds really good, but see how quickly we gloss over these things.

I mean, he says three words. I got to freaking pause it and explain.

Token generation, when you look at the Google IO event, has already turned negative on the second derivative. Now, part of that is law of large numbers. But if you're already going negative on the second derivative already, and we're supposedly early, then you're no longer exponential.

You can't have a negative second derivative and call it exponential. What is a negative second derivative? It's a really positive second derivative is the rate of change is getting faster. when the second derivative, so the rate of the growth, how much the growth rate is changing, turns negative.

It means you're still growing. So the first derivative is positive, but you're growing slower. So it's still going up, it's just going up in a more flat way, [laughter] right? Uh so you can't you can't just say, "Oh, exponentially."

Hey, you're protected against that. But Jake, could I could I shift gears for a second to tell you what's so brilliant about what Jensen did here? There's there's one part. There's one other part in return for that. They get a revenue share above that floor.

Who who gets the revenue share above the floor? This is by the way now a little bit of a a tangent a tangential discussion on Nvidia. We'll listen to part of it and we'll keep getting uh in about 1 minute we'll hit a little bit on the payback periods and then we'll get on to the actual signal for the artificial intelligence.

Yeah. So they I think the way I think the deals will work is say we're willing to guarantee offtake at you know nowhere near $30 or $50 a watt. We're willing to guarantee I'm going to fast forward this.

Gavin basically suggests that Nvidia is giving a 25% residual value guarantee to some of their customers which has been publicized and that there's an opportunity for them to potentially take uh you know some revenue share if revenue is above a certain level. basically a way of continuing this circular financing to make sure that cloud companies like you know Neb Nebius or otherwise can can continue to expand their buildouts and in other words buy more Nvidia chips.

It's a little funny when we get into this circular financing, but let's listen to Sax's reply

that the TAM is getting constrained by the ability to finance this build out. And what he's doing is alleviating that finance constraint so that he can grow uh as big as the the TAM actually is, right?

It's removing that constraint. So for just to take one example, Elon wants to add somewhere around 6 to 8 gawatt next year. We know that that would cost3 to400 billion dollar of capex.

Remember what I wrote on my SpaceX documents? On the SpaceX documents when the earnings came out, I said on the filing, they're going to have to come up with about $414 billion of extra debt.

And their preference is debt over stock issuance, which is good and bad. Uh there you go. It's the same number. It's no secret. We're all seeing the same stuff. It's just how we essentially you make draw conclusions from that information.

Like that's a lot of debt that still needs to be sucked out of somewhere. And we saw last time we had a suckening. It was June. It led to the Leo fold in July. All right. The company just raised a hundred billion in its equity and debt offering.

So obviously they would have to go out and finance that somehow. And as we talked about on our previous episode, the simplest way to finance it would be to get seller financing from Nvidia, especially given that the payback period could be as quick as one year. Could be. Could be.

See a lot of the assumptions that have been layered in here. Okay. Politics will stop overbuilding. Token growth is exponential and the payback period could be as low as one year.

That those are huge assumptions. and you know the payback period is related to that $30 to50 billion of of revenue that Elon wants to pick up from compute maybe it'll happen but look at the assumptions that are built in and when you see the assumptions then you can kind of make the determination like okay we getting closer to uh you know the top doodla they actually well sachs comes up with a very similar thing that I've been harping on for a while in terms of the potential top uh and how to measure it and it comes after 3 minutes of yapping from Gavin and the others.

And it is right at 109. I'll play from about here. Yes.

No, they're not. Okay. Again, this is where David Saxs reads the crap on X and I don't think he's putting his critical thinking hat on. I think he's scrolling on X too much. They're not making hundred billion of ARR right now.

Annualizing out the last day of July, they're at 65 supposedly. Okay, we don't know if that's just because of the co-work explosion. We don't know uh if that's going to last, if that's going to get whittleled away by openweight models.

We don't know. But David Saxs here just literally told you they're making a hundred billion. That's enough to pay Elon's 30 to 50. But they're not. They're making on one or two gigawatts of compute, they're making about 6465.

Now, what's the problem with that? Well, the problem with that is they have massive compute coming online. What I also wrote over here that I didn't talk about yet is they have 5 gawatt coming from Amazon and another 5 gawatt coming from Google and Broadcom.

Now, not all of those 10 gawatt will get built. But even if half of this gets built, that's another 5 gawatt, which is 2 and a half to five times as much power as they have now.

So, is their revenue going to be able to scale to keep making what they're making now this many more times over? Well, according to David Sachs, demand is exponential. So, in that case, sure, why not?

But that's not what we're seeing. We're not seeing exponential demand. We're seeing declining second derivatives on token usage. There's still really big numbers. [laughter] But that's a concern.

Okay. If compute, that's why they're able to pay SpaceX say 50 billion a gigawatt. No, they're losing money. They're not able to. They're losing money.

You know, a spot price for compute. That's why SpaceX is then able to pay 30 billion to Nvidia for chips for the buildout. That's why Nvidia is able to

That's why Nvidia is able to the buildout is going to be way more than and this is the other problem. THE NUMBERS COMPRESS SO MUCH MORE THAN WHAT David Sachs is saying and and is fine.

Maybe maybe the exponential ramp will come but it's not 30 billion that you're putting into the buildout. That's the problem. You know, he's telling you, oh, they're making a hundred and they're going to spend 50 to Elon and Elon's going to give 30 to Jensen and everybody's going to win.

But the reality is Elon's probably spending, you know, 50 for the entire data center, right? We forgot that there are other components to the data center. It's not just the Nvidia GPUs, it's the CPUs, it's the memory, it's uh the flash storage, the hard drives, uh it's the server racks, it's the building or whatever else.

You know, 30 billion might very well just be the GPUs, but realistically, that's probably going up to closer to $40 billion on the latest generations of GPUs.

Well, then you got all the other crap that goes into the data center as well. And this hundred billion dollar figure is not validated.

It's not Kevin Worsh. It's not Nvidia earnings. Nvidia's gone down four out of four of their last earnings after earnings despite beating expectations.