This Week in Startups
00:02:51 · D.A. Wallach, S0
[00:02:51] S2: Is everybody feeling don't know what the hell is going on. You know, it's like you're seeing these little fragmentary views of the elephant every week and every Gavin Baker interview and every all in podcast or whatever, but you still never get to see the whole elephant. So, of course, that's what investing is. You got to make sense out of fragmented information and try to paint a picture of what's happening, but I still have no idea what's happening. [00:03:16] S0: Which is very similar, by the way, to poker. You are trying to make really good decisions with partial information, Like I know my whole cards, but I don't know who yours or the other seven people at the table. Maybe no three cards on the flop, but I still don't know the turn in the river to come. I think that's actually a great way of framing it. Dave, when you saw the numbers drop and you start unpacking the hand, do you feel more bullish on the AI trade? Do you feel like this is making people more confident in what's being built out here and the sustainability of it? Or do you feel like you have more questions rather than answers?
00:03:16 · S0, Dave McClure
[00:03:16] S0: Which is very similar, by the way, to poker. You are trying to make really good decisions with partial information, Like I know my whole cards, but I don't know who yours or the other seven people at the table. Maybe no three cards on the flop, but I still don't know the turn in the river to come. I think that's actually a great way of framing it. Dave, when you saw the numbers drop and you start unpacking the hand, do you feel more bullish on the AI trade? Do you feel like this is making people more confident in what's being built out here and the sustainability of it? Or do you feel like you have more questions rather than answers? [00:03:54] S3: I I feel like they did themselves a favor. You know, I think there's been a lot of concerns in the last quarter that revenue might have been plateauing. I I think we still have those concerns. You know, questions about whether OpenAI is taking share, whether OpenWeight models are taking share. Company looks like they are profitable. Looks like they've been, you know, on an operating basis profitable for the last two quarters. It looks like CapEx spend is crazy. That... That's, you know, over $500,000,000,000 of CapEx. I I don't know over what period of time, but, you know, $100,000,000,000 plus commitments to both Google and Amazon, which are locked in. [00:04:30] S0: Well, maybe unpack that for a minute then, Dave. If that's where the first thing that came to your mind is crazy, how does one go from crazy to sustainable connect the gap between CapEx spend and revenue and profitability? Take a moment there.
00:04:30 · S0, Dave McClure
[00:04:30] S0: Well, maybe unpack that for a minute then, Dave. If that's where the first thing that came to your mind is crazy, how does one go from crazy to sustainable connect the gap between CapEx spend and revenue and profitability? Take a moment there. [00:04:46] S3: If they had gone public last quarter, I think everybody would have been happy with putting a 2,000,000,000,000 to $3,000,000,000,000 valuation on the company. But if you look at... I don't know if we can share the graph of ticker trends. This is Anthropic revenue, this is OpenAI revenue. [00:05:01] S0: Okay. So when we're looking at the Anthropic revenue, this is companywide annualized revenue run rate. We just did a story about this in newsletter, actually. This is R as in a pirate, A R R. When you see that, which is not to be confused with ARR, with three Rs, annualized run rate is typically done by taking the last month or the last quarter and timesing it either by that period of time, twelve or four. And here what we're seeing is it's flat at 76,800,000,000 in ARR. And you see here, even on this website, they put ARR, which would lead you to believe it's annualized reoccurring revenue. Annualized reoccurring revenue is for subscription products like Netflix or your Verizon account, it's distinctly different than this. So what's your takeaway here? Obviously we see flat, but what is that based on? Where did they get that number?
00:05:01 · S0, Dave McClure
[00:05:01] S0: Okay. So when we're looking at the Anthropic revenue, this is companywide annualized revenue run rate. We just did a story about this in newsletter, actually. This is R as in a pirate, A R R. When you see that, which is not to be confused with ARR, with three Rs, annualized run rate is typically done by taking the last month or the last quarter and timesing it either by that period of time, twelve or four. And here what we're seeing is it's flat at 76,800,000,000 in ARR. And you see here, even on this website, they put ARR, which would lead you to believe it's annualized reoccurring revenue. Annualized reoccurring revenue is for subscription products like Netflix or your Verizon account, it's distinctly different than this. So what's your takeaway here? Obviously we see flat, but what is that based on? Where did they get that number? [00:06:03] S3: You looked at the growth from January to July, you'd be like, holy shit, that is crazy growth. That's almost hyperbolic. But then it tailed off a little bit, and people were wondering, like, why did that tail off? Was that because of OpenAI releasing models that were more competitive and taking share? Replit recently announced they've been switching their default to OpenAI instead of Anthropic. Is it because open weight models are taking share? I I would still say that Anthropic is [00:06:30] S0: Or maybe there's a data problem here. Like, maybe this website doesn't have who knows where they're getting their data from here?
00:06:30 · S0, Dave McClure, Ben Black
[00:06:30] S0: Or maybe there's a data problem here. Like, maybe this website doesn't have who knows where they're getting their data from here? [00:06:38] S3: Yeah. I think my partner just did a teardown of the Anthropic S1 and it still looks like they are growing. I think they added at least 5,000,000,000 of new revenue in the quarter. So I'm not concerned that Anthropic is not growing, but they're not growing as fast. And I think you'd kind of wanna look at that and, like, compare that to the CapEx spend. Can they just, you know, keep up with the spend that they're doing, and is that revenue gonna continue? They've previously been talking about $203,100,000,000,000 dollar kind of run rates by the end of next year. I'm not sure. I'm still a believer that they're gonna hit 300. I might I might guess it's gonna be between $1.50 to 200 next year. We'll see. Yeah. But they did kind of answer a lot of questions about profitability that seemed like they had profitability under control. I think I'm still a believer in Anthropic, at 1 and a half trillion, not at 2 or two and a half trillion. That's just my, you know, hot take. [00:07:35] S0: All right, Ben. Your takeaway when you saw the s one and, the ensuing discussions? [00:07:41] S4: When I saw the s one, I really was so impressed with how how much analysis could be done in such short period of time with a... With with AI. There are so many analysis, [00:07:51] S3: I can
00:07:51 · Ben Black
[00:07:51] S4: barely keep up with them all. It was like... In in my analysis. But, really, you know, when I... I read the s one who said, you know, we need new measures. I think, you know, this is really the first time we've had financials from a company that does what Anthropic does that are getting now to be, torn down by public market analysis... Analysts. And how the public market analysts, you know, I would say with private companies, this is the moment where the rubber hits the road. And right now, we're just seeing literally, like, the first inkling of all the... Of, you know, of the way that we're gonna analyze these companies. We're still just like what you did. We're trying to take SaaS analytical tools like ARR and slap them on a completely new type of business model. It's gonna take a while before I think that the world of business analysts catch up with Anthropic and OpenAI, where essentially, you're you're selling compute at some sort of margin, and you're selling it across multiple models that have different values and different costs, all of which are falling. And I frankly am not sure I understand this business yet. I think that's the biggest takeaway I have is that, you know, we have to have a whole new set of measures for these businesses that really help us understand what's going on because I think the s s one can be characterized in one way by one group of people, another way by another group of people, And who win the fray... Who wins the framing in the next two weeks is gonna determine whether or not 1,500,000,000,000 is a good number or not.