Big Technology Podcast
00:35:43 · Alex Kantrowitz
We're we're about five days past. I don't think we'll spend too much time on the s one this week. But it was really a Rorschach test, I think, for people in terms of if you believed in Anthropic, found plenty of evidence to show you that the company was very healthy. Booming growth they went from 1,500 customers in December 2025, spending $100,000 or more to $11,900 in June. So that's massive increase. They also increased their customers spending 1,000,000 or more. They had something like $110,000,000,000 cash on hand. So if you thought Anthropic was just going to incinerate all its money and desperately crawl to the IPO, Obviously that's not what happened. They're making $6,600,000 per employee, so they're very efficient. Their efficiency was going up. They used to make $0.40 for every dollar of compute they spent. In the S1, they showed that they are now making $2.08 for every dollar of compute they spent, so they completely flipped that on its head. But the detractors will point to two things: the fact that Q3 revenue was not included in the S1 and Q3 revenue looks like it will massively decelerate from something like more than 100% quarterly revenue growth to in the 40% quarterly revenue growth field, and of course that spells the end of tokenomics. Then there's also the 500,000,000,000 or so that they've committed to spend, much of which is non cancelable. When you looked at this blot of ink on the paper, what did you see?
00:37:19 · Ranjan Roy
[00:37:19] S1: What I saw was a business that is doing very well, and I think, like, actually met a lot of the height... Heightened expectations. What I didn't see in the most frustrating part of this to me, all I want is to understand the unit economics of Anthropic's business to give a reflection of what a great AI business can look like. And I did not get that. And I'll I'll tell you why. So I actually don't think the q three numbers, again, a lot of people are... There's like a lot of conspiracy theory ing going on, I feel, around, like... It actually makes sense. Like, audited financials, it would have... Quarter would have ended, September 30. So there's... They're going to release it based on... I think it's November 11. If they don't IPO before then, which they likely won't, they're going to have to release q three. So we'll get those numbers. We'll we'll see what... To me, the numbers actually were so good that that actually raised questions for me. And what I I mean by that is, again, I mean, you kinda started to say it, but like, the whole story has been that like cost of compute is gonna be so dramatic, and that's why we need to just raise so much money. And that's why we need, like, the the the... It's a very expensive thing, and that's why this is all gonna be very expensive. They lowered their cost... Compute as a percentage of revenue, and q one was 71.6% to 48.1%. It was 225% of revenue a year and a half ago. So, like, all these numbers show us that actually the economics are almost magical. Like compute is decreasing, revenue is increasing, compute as a percentage of revenue is decreasing. To me, I don't think we yet understand what the actual true cost of compute was relative to their revenue. And the reason being the most glaring example was like, there's all this talk around the $80,000,000,000 loss, is it real as an accounting treatment? Within that, the big part of that was, it was the increase in value of convertible notes from Amazon and Google. But we talked about this a year and a half or two years ago, actual calendar year and a half, two years ago, not Alex's six months is a year framework.
00:39:47 · Alex Kantrowitz, Ranjan Roy
[00:39:47] S0: I just wanna say, I don't mean six months is a year. I just wanna say that when people say six months is a... When people say that happened last year, they really mean it happened in April. They've lost the ability to keep track of time. [00:39:59] S1: Well, the only reason this was actually related to, I think it was like summer twenty twenty five convertible note issuance. But remember, we were to... Actually, this was back in late twenty four. Amazon's putting in 4,000,000,000. They openly are saying part of this is heavily discounted compute. A lot of the deals with Broadcom, with all these other people that are both investing into Anthropic, they're receiving significantly discounted compute. And it was just a reminder, I think what we're seeing here is not reflective of what this is actually gonna cost and we're still not getting a good representation of, like, what are the unit economics. That's my only frustration. I think, like... Which also is gonna be really important because when we talk about these 500 bill... What is 325 or 5... Whatever the future committed spend is. [00:40:52] S0: It's in the neighborhood of 500,000,000,000. Yeah.
00:40:52 · Alex Kantrowitz, Ranjan Roy
[00:40:52] S0: It's in the neighborhood of 500,000,000,000. Yeah. [00:40:54] S1: Yeah. We still don't know what what does Anthropic's business look like. Like, if they were paying market rates for compute relative to their revenue, we know they aren't, but we don't know how much that discount is. So my my ask for investors is to demand that they disclose and provide a better understanding of actually what they were paying over the last number of quarters and year and what kind of discounts they received. Otherwise, we still don't know what is the actual kind of, like, picture of this business. [00:41:29] S0: But wouldn't you agree, by the way, that if you think about the amount of money that Anthropic has raised, they still have $100 plus billion in cash. So it couldn't have been that much. Right? [00:41:43] S1: No. No. That that that... That's exactly what I'm saying. It's actually, like, insanely efficient. It's so... Feel... For everything we've talked about for the last two years, suddenly you look at this and you're like, wait. This is the most cash efficient, compute efficient thing imaginable. Like, the the number that they've only burned $9,500,000,000 in this entire last few years is actually... Like, we know it's not exactly correct because they receive below market rates as part of these investments, but we don't know how much less. And we know they're not going to continue.
00:41:43 · Ranjan Roy, Alex Kantrowitz
[00:41:43] S1: No. No. That that that... That's exactly what I'm saying. It's actually, like, insanely efficient. It's so... Feel... For everything we've talked about for the last two years, suddenly you look at this and you're like, wait. This is the most cash efficient, compute efficient thing imaginable. Like, the the number that they've only burned $9,500,000,000 in this entire last few years is actually... Like, we know it's not exactly correct because they receive below market rates as part of these investments, but we don't know how much less. And we know they're not going to continue. [00:42:22] S0: Let me press you on this. Google said it was gonna invest up to 40,000,000,000 in Anthropic. We don't know if they've invested all 40,000,000,000. Amazon said they agreed to invest 25,000,000,000. So let's say that was all Amazon and Google giving Anthropic compute for free. You know, that's just $65,000,000,000. They have a 100 plus billion in the bank. [00:42:46] S1: No. No. But but they spent 5,500,000,000 in compute last quarter, 3,500,000,000 in q one, 8,000,000,000 total in the first half of the year. So like if they're investing, if Google and Amazon are discounting by in that order of magnitude of billions, that fundamentally changes the margins of what Anthropic's business is. So the cash on hand I think is separate. This is what should they have paid and what would their actual expenses ratios look like relative to what they are able to charge.
00:42:46 · Ranjan Roy, Alex Kantrowitz
[00:42:46] S1: No. No. But but they spent 5,500,000,000 in compute last quarter, 3,500,000,000 in q one, 8,000,000,000 total in the first half of the year. So like if they're investing, if Google and Amazon are discounting by in that order of magnitude of billions, that fundamentally changes the margins of what Anthropic's business is. So the cash on hand I think is separate. This is what should they have paid and what would their actual expenses ratios look like relative to what they are able to charge. [00:43:19] S0: So your suggestion is that Amazon and Google both invested in the company and not dollar for dollar, but just said, we want you to grow. And so outside of our investment, we're also going to provide you with discounted compute. [00:43:31] S1: Amazon, that was explicit in their convertible note. They said it. And we talked about this way back. And it was kind of a joke. Like we're like... And back then I remember we were like, is it cash or is it credits? Is it like... But now you're actually seeing the result of that is we still don't know of what is Amazon business going to look like going forward. And I think to me that's the only... Otherwise, everything looks so good. It's almost like unbelievable. And that's why to me, they should disclose around here is actually like our average... What this discounting looked like in the past so people can understand what it will look like in the future.