Anthropic’s reported $80B loss was not cash burn

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 6:44

   So number one is that this $80 billion loss number that we've heard so much about, the media has constantly been pumping and harping on as proof that Anthropic was either a fake business or that we were in this sort of whole bubble. That narrative basically just got completely refuted because it turns out that that $80 billion loss is a phantom number. It resulted from the fact that Anthropic had issued convertible notes to basically their infrastructure partners like Amazon and so forth.

2. 7:16

   And as Anthropic's valuation increased, those convertible notes got recorded as a loss on the balance sheet and that flowed through into the income statement, but it's not a real loss. If they had just issued equity, that loss wouldn't be there. So it was not a cash loss. It's basically a fake number. And in fact, what was revealed in this S1 is that Anthropic has been an extraordinarily capital-efficient business.

3. 7:42

   So they've raised something like, I think it was around $110 billion, roughly. And they've only spent about $9 or $10 billion throughout the entire course of this business. And they still have, like you said, J. Cal, about $100 billion left of cash or cash equivalents on the balance sheet. Everyone thought, based on the previous reporting, that this was a huge loss-making business. In fact, it ended up being a very capital-efficient business.

4. 8:08

   And then the other thing that you saw was that their margins have been improving. I think their gross margin is now something like 85%. And what you see is that for a given task or capability, they're showing roughly 10x improvement year over year in their ability to deliver that task. the cost effectiveness of that task. It gets 10x cheaper year over year. And that allows them to cut the cost of that model or that task by, let's call it 50%. And so the consumer recognizes a huge cost savings, but they are able to realize an even greater savings because they're able to capture the rest of that efficiency in their margin. And that's what we basically saw is that, again, they're cutting the cost of their tokens significantly, but their margins are going up even faster because they're able to produce these models that are more and more efficient.

5. 9:08

   And they call them compute multipliers. It's very interesting the way that Anthropik sees their business. The framework that Dario has is that there's raw compute, which are scaling very rapidly as well as a result of better and better chips and more and more of those chips. And then there's the compute multipliers, which are things like algorithmic improvements or better curation of the data, or it can be things like better architectures, more MOE architectures, things that make all these models more efficient. So they're able to give you the same answer for one-tenth the cost, or they can take those efficiencies and create greater capabilities. So this is what we're seeing is that that business is actually improving.

6. 9:53

   And I think it is a little bit of a refutation to all the people who've been saying over the past year that they're getting commoditized away, that open models are basically putting them on a hamster wheel where they're basically going to zero on gross margin, or that it's a business that only works because it's heavily subsidized by venture capital. In fact, I think there's a lot of reasons to believe that Anthropic has a pretty big lead. And if you believe that recursive self-improvement has kicked in, as they are claiming it has, then their elite could actually grow. Freeberg, your thoughts on the impact of this massive deluge of capital coming in to