This Week in Startups
00:47:37 · D.A. Wallach, S0, Dave McClure, Ben Black
[00:47:37] S2: Riskiest hand that's ever played putting [00:47:40] S0: 400,000,000,000. That's a very good point there, D. A. He's 400,000,000,000 into CapEx spending being essentially [00:47:48] S2: balls out, dude. And the guy's... He's 80. Shaky, man. Yeah. I mean, could we imagine taking that kind of risk when we're [00:47:56] S0: 80? And, by the way, spending another whatever... Putting whatever tens of billions on the rest to buy every dying media asset. No. You had [00:48:03] S3: to put it [00:48:04] S0: on the same charge. Giving it to your son. [00:48:05] S4: I mean Mhmm. [00:48:06] S3: Yeah. Son's doing the same shit. [00:48:08] S0: They are spreading the treasury. There... It's it's the treasury. This, I think a lot of empires spread the treasury a little too far and had too many conquests going on at once. Let's talk about who won here. Lon, we have to get two things before we wrap up. Number one, who won? Sure. Who won in the old venture capital game here at This Week in Venture Capital? Then number two, convertible notes hitting public markets for the first time in a in a in a significant way. Go ahead.
00:48:44 · Lon Harris, S0, Dave McClure
[00:48:44] S1: Yeah. Alright. Well, we'll start we'll start with that one. So, Anthropic, lot of headlines were saying they burned $80,700,000,000. That was the net loss for the first half of twenty twenty six, but 72,400,000,000 of that comes from just one line, quote, change in fair value of convertible notes, derivative liabilities, and safes. In this case, it was mostly convertible notes plus derivative liabilities that were tied to Amazon and Google note purchase agreements we were previously discussing. Anthropic elected to carry the notes at fair value instead of splitting them into debt and equity pieces, or every reporting period the company now remeasures these liabilities. If its value has gone up, the shares owed to investors are worth more, hence more liability. The increase is recorded as an expense on the income statement then added back onto the cash flow statement as non cash. So this is raising a lot of questions about which of these numbers matters, which doesn't, and how do you all think private markets are going to... Or public markets are going to learn how to interpret these, numbers. [00:49:47] S0: Dave. Yeah. This is your wheelhouse. You just net out that 72,000,000,000. [00:49:51] S3: I think my partner did the analysis of this. And again, I think they're in fine shape on a profitability basis. There's some, you know, noncash expenses here that people are are looking at that are worried about too much. But on actual cash out the door, they're they're doing fine.
00:49:51 · Dave McClure, Ben Black
[00:49:51] S3: I think my partner did the analysis of this. And again, I think they're in fine shape on a profitability basis. There's some, you know, noncash expenses here that people are are looking at that are worried about too much. But on actual cash out the door, they're they're doing fine. [00:50:06] S4: Yeah. But I I heard though... I mean, that that there is a negative kind of implication that that those notes... A lot of the notes were due to their Amazon deals with a... AWS and were basically subsidizing their compute costs. They gave lower cost. They gave they gave lower compute compute costs. So this goes back to the point of, like, do we really understand cost of compute here? Because that's my understanding of it.
00:50:06 · Ben Black, S0
[00:50:06] S4: Yeah. But I I heard though... I mean, that that there is a negative kind of implication that that those notes... A lot of the notes were due to their Amazon deals with a... AWS and were basically subsidizing their compute costs. They gave lower cost. They gave they gave lower compute compute costs. So this goes back to the point of, like, do we really understand cost of compute here? Because that's my understanding of it. [00:50:28] S0: Well, here's the great news. Sunlight will figure this all out, and we're about to have essentially two of the three, the the father, son, and holy holy ghost here. We're about to figure this out. Once they're public, then we just have to somehow pencil in one hand of poker at the table, which is what cards are Sam Altman holding. We know what cards Elon's holding. We understand his compute costs. We understand what he's building. We have perfect insight into that. We know what Amazon's doing. We know what Oracle's doing. So really is just going to... And we know what NVIDIA's doing. When we get these two companies public, we're going to know the answer to all these questions. We'll keep at it there. Then let's talk about the winners. Sure. Dave, you were kind of rolling your eyes there, I'm assuming because this is off balance sheet loans and stuff like that.
00:51:17 · Dave McClure
[00:51:17] S3: Yeah. Yeah. Yeah. I I think what we're not able to clearly understand right now is the amortization of that cost of capital. Like, you're putting a shit ton of money out there on CapEx, and the question is, though, how quickly or how not quickly does that CapEx, amortize, you know, depreciate over time? If you're talking about chips, you know, people are saying, hey. These chips are gonna make money for us for four or five, maybe even six years. For the neo clouds, I don't know if that's, like, completely true. I think they're spending a lot of money, and those, you know, things that they're leasing and buying may be making money in the future, may not be making money in the future. I think we saw from, SpaceX q three earnings, SpaceX is in great shape. Like, Elon's, you know, payback is less than twelve months. He thinks he can get to six months. And, you know, I think made some analysis about, like, all the other neo clouds in trouble because their payback was not, anywhere close to that. Was two or three years, maybe not even. Yeah, but
00:52:16 · S0
[00:52:16] S0: I mean, two or three years seems completely reasonable as well, D. A. And when we look at the history of compute and the amortization of computers, we see a couple of observations here. I want your comment on D. A. You're Google or you're Amazon, you have new technologies emerge and optimized software. So what was used as a frontline computer eventually becomes used for, I don't know, cold storage, background jobs. And then eventually, the equation is, in a data center, is this providing enough value in your eight, nine, or 10 of serving web pages or Gmail? Is it worth sending a human to pull this out of the rack or should we just let it keep grinding and the average lifespan of compute at like other data centers? We're talking about seven, eight, nine, ten years for some of these servers, And it's only the human cost of swapping it out that becomes the cost. I think these machines are going to be doing reasonable things in years seven, eight, and nine with optimized software, you know, to move them to inference, to move them to storage jobs, background jobs, etcetera. What's your take on how long this compute has a lifespan I