Latest / In the Loop AI / The AI Dark Horse: Anthropic's $300B Secret Formula

Transcript

The AI Dark Horse: Anthropic's $300B Secret Formula · In the Loop AI

0:00
Ejaaz: Imagine spending years designing, building, perfecting your app or product only
0:04
Ejaaz: for an AI model to come along and build an identical replica in seconds.
0:08
Ejaaz: Well, this is exactly what Anthropic has achieved.
0:11
Ejaaz: They have the best coding AI model clawed in the world.
0:15
Ejaaz: And they've fended off multi-billion dollar attacks from the likes of Google and OpenAI.
0:19
Ejaaz: In fact, they're so good that they've 10x their annual revenue three years in a row.
0:24
Ejaaz: They're on track to achieve $10 billion in rev this year and they're projecting
0:28
Ejaaz: $70 billion in revenue by 2028, which is just insane.
0:33
Ejaaz: But if that still hasn't convinced you, get this.
0:36
Ejaaz: 50% of AI prompts today are just people vibe coding and 80% of those people, they're using Claude.
0:42
Ejaaz: But here's the crazy plot twist.
0:45
Ejaaz: Anthropic superpower isn't that they have the best coding AI model in the world.
0:49
Ejaaz: It's what you can do with that coding AI model. Hear me out.
0:53
Ejaaz: Imagine never having to open an app or a website ever again.
0:57
Ejaaz: Claude just generates the exact digital experience that you desire.
1:01
Ejaaz: Imagine not needing to scroll endlessly looking for the perfect pair of shoes that you want to buy.
1:06
Ejaaz: Claude just generates the UI and presents you with a buy now button that you
1:11
Ejaaz: just click and it completely evolves the way that humans interact with the internet.
1:16
Ejaaz: Now I wish that this was the only good thing that Anthropic was good at.
1:20
Ejaaz: But unfortunately, there's so much more. In fact, Microsoft and NVIDIA are investing
1:24
Ejaaz: around $15 billion in Anthropic, and they're winning the enterprise market share as well.
1:30
Ejaaz: All of this on the fact that the or rather the rumor that they're going to IPO
1:35
Ejaaz: next year at a crazy valuation, which might just mean that Anthropic is the
1:39
Ejaaz: best investment opportunity in the recent decade.
1:43
Josh: Yeah, this episode kind of stems from the idea that I think more people are
1:47
Josh: bearish on Anthropic than they should be.
1:49
Josh: And it mostly comes from a misunderstanding because I'd place myself in that category
1:53
Josh: I hadn't cared that much about Anthropic. I heard a lot of headlines about them
1:56
Josh: being kind of bearish and decelerationist, and it rubbed me the wrong way.
2:00
Josh: But upon doing the research for this episode, it became clear that there's actually
2:04
Josh: some really interesting,
2:06
Josh: advantages and cool opportunities that exist inside of Anthropic for an investment
2:10
Josh: opportunity, but also for just personal productivity and use cases on a daily basis. It's very cool.
2:15
Josh: So before we get into that, I want to set the origin story, kind of do some
2:18
Josh: context setting for the cast and characters here.
2:20
Josh: You could think of Anthropic as kind of the rebel house that walked out of OpenAI.
2:24
Josh: We like to do the Game of Thrones talk.
2:26
Josh: OpenAI was the king house. Dario and his sister, Danielle Amode,
2:30
Josh: were actually senior leaders at OpenAI, the company.
2:33
Josh: And in fact, on screen, we're showing one of the founding papers that he wrote while he was at OpenAI.
2:37
Josh: So if OpenAI is how stark, then Anthropic is like this group of ex-Starks that
2:42
Josh: said, yeah, we're just going to go build our own kingdom over there on the other
2:45
Josh: side of the hill. Now, why did they do this?
2:48
Josh: The divide comes from the underlying ethos of what we see Anthropic talking
2:52
Josh: about a lot, which is safety.
2:55
Josh: And a hot take is that safety, their obsession with safety isn't actually a break,
3:00
Josh: but instead it's their enterprise sales strategy in disguise,
3:03
Josh: which we're gonna get into because a lot of the market is enterprise sales and
3:06
Josh: they require a serious amount of safety and constraint that a lot of models don't actually have.
3:11
Josh: So Dario, he has this kind of heretic belief in Silicon Valley where he thinks
3:15
Josh: benchmarks, they're mostly Cope and that coding is the only benchmark that matters.
3:19
Josh: And through this foundational value set, we start to see where Anthropic shines, EGS.
3:25
Josh: And I think that's probably where I'm going to pass it off to you to talk about
3:27
Josh: how impressive some of these models have been recently, particularly around
3:32
Josh: the unique advantage that they have, which is coding.
3:36
Ejaaz: Yeah, I mean, I'm with you, Josh, that I was also a hater of Anthropic.
3:41
Ejaaz: I just always saw them as kind of like a corporate NARC model.
3:44
Ejaaz: And I kind of like dismissed coding primarily because that's not the majority
3:48
Ejaaz: of the activity that I do on a daily basis, right?
3:51
Ejaaz: Versus some like software engineering oriented friends that really see the benefit of this model.
3:56
Ejaaz: And then I came across this crazy stat in prepping for this episode,
3:59
Ejaaz: which is, Josh, just last year, the percentage of tokens.
4:03
Ejaaz: So that's ai tokens the the stuff that you kind of fuel with prompts that composed
4:08
Ejaaz: of coding um guess guess what that percentage might have been oh
4:11
Josh: Man uh coding 30.
4:13
Ejaaz: It was 12 last year oh wow okay and now 12 months later it's 50 and called code
4:20
Ejaaz: accounts for 80 of that right so um that was the latest stats release whoa hold
4:26
Josh: On pause there wait so 80 of 50 of the tokens that are being generated,
4:31
Josh: are now generated by one company.
4:33
Ejaaz: Correct.
4:33
Josh: Oh, man. Okay. Whether it's through cursor.
4:36
Ejaaz: Whether it's through some kind of integrated development environment or through
4:39
Ejaaz: Claude Code directly, which is like primarily where people are going to, to Vibe Code.
4:43
Ejaaz: 80% of that is Claude Code, which is just an insane stat and such a rapid rate
4:49
Ejaaz: of increase over the last 12 months.
4:51
Ejaaz: This is stats from the recent report from Open Router.
4:54
Ejaaz: Shout out to Alex Atala. We did an amazing episode with him a few months back.
4:57
Ejaaz: Definitely go check it out. But the main takeaway from this,
5:01
Ejaaz: Josh, is that they have the best coding model.
5:03
Ejaaz: And it's not like this is an old coding model. This got released,
5:06
Ejaaz: I think, a week and a half ago in the aftermath of Gemini 3 Pro's kind of like
5:13
Ejaaz: crazy entrance where it just broke every single benchmark, including coding
5:17
Ejaaz: So Anthropic released this model and said, hey, hey, we're still the kings here.
5:22
Ejaaz: Step back. And I think this summary from McKay Wrigley, who is a guy,
5:27
Ejaaz: Josh, you were telling me this before, who gets access to these models kind
5:31
Ejaaz: of well in advance of them being released.
5:32
Ejaaz: So he gets to test them out and really give an in-depth review.
5:35
Ejaaz: So what we're reading here is, you know, something that is really practical and true to the bone.
5:40
Ejaaz: He explains the kind of bigger picture with this coding model.
5:44
Ejaaz: The main one being that it just allows autonomous development of pretty much anything.
5:50
Ejaaz: I think a lot of people look at Anthropic and they think, oh,
5:53
Ejaaz: they're just the coding model.
5:55
Ejaaz: I don't want to use it. It's got nothing to do with me because I don't code.
5:58
Ejaaz: And what the argument that he's trying to make in this post here effectively
6:01
Ejaaz: is if LLMs enabled you to write words autonomously and then the first iterations
6:08
Ejaaz: of Claude allowed you to code,
6:10
Ejaaz: then this latest Claude Opus 4.5 release will be the enabler for AI agents.
6:17
Ejaaz: It'll allow them to autonomously do work for you.
6:19
Ejaaz: All those crazy kind of examples that we've come up with in the past now become
6:23
Ejaaz: a reality with this new model.
6:25
Josh: Yeah, one of the things that I found shocking, well, outside of that stat,
6:28
Josh: that 80% of 50% of tokens are anthropic, is that people don't care because people
6:33
Josh: don't use it. And I find myself in that bucket.
6:35
Josh: But when you read reviews like this, when you see those numbers,
6:38
Josh: you start to realize that even if you're not using it directly,
6:41
Josh: this is a massively large deal.
6:43
Josh: And one of the things that he mentioned in this post is just the revenue growth.
6:46
Josh: And it starts to make sense now that we have these pieces where the first year
6:49
Josh: was one to a hundred million.
6:51
Josh: The second year was a hundred million to a billion. The third year,
6:55
Josh: one billion to 10 billion.
6:56
Josh: Can they go to a hundred billion? Well, they're projecting 70 billion.
6:59
Josh: So maybe not another 10 X, but this massive gargantuan improvement in revenue
7:04
Josh: year over year because they have this monopoly.
7:07
Josh: And as I was reading through this review, and we're not going to go through
7:10
Josh: the whole thing because it's fairly long,
7:11
Josh: This is a developer who is completely in awe over the fact that the entire paradigm
7:16
Josh: has kind of shifted in a way that wasn't previously possible.
7:19
Josh: The new model introduces this thing called planning mode. And if I was playing
7:22
Josh: around with it a little bit because I was just fascinated to see how it works,
7:25
Josh: where EJS, if you've ever done deep research in open AI, you'll recognize that
7:29
Josh: you share the prompt of what you want, but then it asks follow-up questions
7:32
Josh: to kind of make sure that you're doing the right prompting and make sure that
7:36
Josh: you're getting the right result that you want.
7:38
Josh: What planning mode does with Claude is it does the same thing.
7:41
Josh: So if you have a query, if you have a prompt, and you want it to be one shot,
7:44
Josh: you want to just say what you want and get it on the back end,
7:47
Josh: the new version actually has a path of getting this.
7:50
Josh: And it seems to me that this is the public sentiment everywhere that...
7:54
Josh: In a way, Claude almost has a monopoly on code, where if you really genuinely
7:57
Josh: want the best model in the world at writing code, this new 4.5 model from Claude is it.
8:04
Ejaaz: I particularly like this sentence that he writes here, where he describes Opus 4.5 like Waymo.
8:12
Ejaaz: He says, you tell it, take me from A to B, and it takes you there.
8:17
Ejaaz: After a few of these experiences, your brain realizes, oh, okay,
8:20
Ejaaz: we live in this world now, and now you're hooked.
8:23
Ejaaz: So in the same way that Waymo, you can step into a Waymo car and it autonomously
8:28
Ejaaz: takes you from site A to site B,
8:31
Ejaaz: he explains it in the same way that Claude code can kind of take you from problem
8:34
Ejaaz: set that you described to it to an output in a matter of seconds.
8:38
Ejaaz: And that's super powerful. Your point around the prompting stuff,
8:44
Ejaaz: Josh kind of reminded me of this concept of meta-prompting, where you kind of
8:49
Ejaaz: write a prompt to ChatGPT or to Claude, and you're kind of like,
8:53
Ejaaz: hey, make this prompt sound better.
8:57
Ejaaz: Write this in an expert way. Try and say what I'm trying to say.
9:01
Ejaaz: And the LLM kind of evaluates your prompt and then rewrites it and then submits
9:05
Ejaaz: it to itself. And it's this kind of like recursive momentum that Claude has
9:09
Ejaaz: really sort of nailed in the coding vertical, right?
9:13
Ejaaz: And then I kind of think of another thought that Andre Carpathy came up with
9:18
Ejaaz: a while ago where he says the new language of coding is actually English, right?
9:23
Ejaaz: And we see this development of people getting really, really good at prompting
9:27
Ejaaz: and they structure it in such a way that it's kind of like a hybrid between
9:32
Ejaaz: the English language and coding.
9:34
Ejaaz: And I see kind of like the future of AI coding, not just being this thing that
9:37
Ejaaz: is just kind of relegated specifically to software engineers and those who have
9:41
Ejaaz: a certain software engineering background, but anyone and everyone can get access to it.
9:46
Ejaaz: It just kind of like converges into this specialized form of the English language,
9:50
Ejaaz: which is both kind of like
9:52
Ejaaz: unnerving in a way because it's kind of like undiscovered in uncharted territory,
9:55
Ejaaz: but really, really enabling once you figure out that, oh, that can unlock so
9:59
Ejaaz: many other things if like now any kind of like 10-year-old can figure out how to code.
10:03
Josh: Yeah. And maybe it's also helpful to explain why coding has become the default
10:08
Josh: interface, because I think it's not really obvious when you talk about Andre's
10:12
Josh: quote where English is the final language.
10:15
Josh: Well, natural language is great for asking questions, but code is great for actually doing things.
10:20
Josh: So we're converging on this pattern where like a user will describe their intent in English.
10:24
Josh: The AI turns that into code. And then the code is the interface to everything,
10:28
Josh: to data, to APIs, to payments, to UIs, to all of that.
10:32
Josh: And actually on the topic of UIs in particular, EGIs, I know you have some takes
10:36
Josh: on this that I would love to hear about because your idea of what the future
10:41
Josh: of UI is, and for those who aren't familiar,
10:42
Josh: user interface is basically when you engage with an application,
10:45
Josh: the way the app looks and feels and functions, that is the user interface.
10:50
Josh: But Ijaz, you have some interesting takes on the future of UI and what that's
10:54
Josh: going to look like as it relates to Claude in particular. Yeah.
10:58
Ejaaz: Yeah, so I think this is the number one ball case for Anthropic.
11:02
Ejaaz: It isn't the fact that they have the best coding model in the world.
11:06
Ejaaz: It's what that coding model enables.
11:08
Ejaaz: And it's this term I've been kind of playing around with called ephemeral UI or ephemeral AI.
11:15
Ejaaz: And what I'm describing here is, well, typically people are used to navigating
11:20
Ejaaz: to websites, going on apps, scrolling on their phones, tapping icons.
11:23
Ejaaz: I argue that in the future, apps don't exist, Josh.
11:28
Ejaaz: Websites don't exist. Scrolling doesn't exist.
11:31
Ejaaz: All that happens is your AI listens to what you want and it generates a user
11:36
Ejaaz: interface personalized to exactly what you're asking for in milliseconds.
11:42
Ejaaz: So it generates the website from scratch where it didn't exist before you prompted.
11:46
Ejaaz: It generates the buy button.
11:48
Ejaaz: Heck, it might even create a product that didn't even exist before that you
11:52
Ejaaz: can click order and then it sends it to a manufacturer or factory in the real
11:56
Ejaaz: world where a bunch of robots or people create it and then deliver it to your door a few days later.
12:01
Ejaaz: Now, this was kind of like crazy to wrap my head around.
12:04
Ejaaz: And I found this really great tweet here that I'm showing you from Justin Murphy,
12:09
Ejaaz: which kind of captures the excitement.
12:10
Ejaaz: I'm going to read an excerpt for it. Claude code is personal AGI.
12:15
Ejaaz: You can't use this thing for more than a weekend without realizing it's completely over.
12:19
Ejaaz: At first, you make a graphic user interface app and you're like, okay, cool.
12:24
Ejaaz: Then you're like, wait, graphic user interfaces are a waste of time.
12:27
Ejaaz: Let's just make a terminal app.
12:29
Ejaaz: And then you're like, wait, apps are a drag.
12:32
Ejaaz: What if I just ask Claude Co to do the thing directly?
12:36
Ejaaz: Works immediately. And then you're like, damn, now asking Claude Code to do stuff feels like a drag.
12:42
Ejaaz: Can I maybe have Claude make a system that says this stuff for me?
12:45
Ejaaz: And he goes on and on and on and he reaches the end point where he's like,
12:49
Ejaaz: oh my God, this is going to completely change the way that we interact with
12:52
Ejaaz: the internet and the digital world that exists today.
12:55
Ejaaz: We're not going to open up our laptops and navigate to a website or a browser.
12:59
Ejaaz: We're going to have just AI funnel this entire thing for you.
13:03
Ejaaz: Josh, is anything that I'm saying makes sense? Like, it's hard for me to wrap
13:07
Ejaaz: my head around. So I can imagine it might be similar.
13:10
Josh: Yeah, this makes total sense. In fact, my question, like, is not whether this is correct or not.
13:14
Josh: It is, is Claude the company that will be able to do this?
13:17
Josh: So it seems like at the limit, when you're exploring, like, where are the outer
13:21
Josh: bounds of where AI goes to?
13:23
Josh: The end state is probably something similar to photons in, photons out.
13:26
Josh: Meaning, like, you get sun from the sky, it turns into intelligence through
13:30
Josh: a transformer and it just outputs some sort of information,
13:34
Josh: What does that information look like? Well, is it text on a screen?
13:38
Josh: Is it a visual interface?
13:40
Josh: Is it audio? Is it all of the above? And what we're seeing on screen here is
13:44
Josh: Andre Carpathy, who is the legend that we all know and love.
13:47
Josh: He is describing one possible thing, which is just what you were talking about
13:50
Josh: is, is this dynamic creation of user interfaces. Now, I just want to note the
13:55
Josh: fact that this is a Google example and not an Anthropic example,
13:59
Josh: which is where my one little bit of concern comes in.
14:02
Josh: So Anthropic is unbelievably good at coding models, large language models,
14:06
Josh: which traditional LLMs, they predict text token by token, kind of like left
14:11
Josh: to right. And they rely heavily on prior tokens too.
14:13
Josh: So like, it's just, it's a very linear way of thinking.
14:16
Josh: But what Google is doing and what some of these other companies are doing that
14:20
Josh: have more of a physics-based approach versus a code-based approach is this thing
14:23
Josh: called diffusion models, which instead of relying on tokens,
14:27
Josh: they kind of diffuse everything and they start with noise and then they iteratively
14:31
Josh: refine the entire output all at once. So it's much faster. It's much more dynamic.
14:35
Josh: It feels like it's appropriate for building these types of user interfaces.
14:38
Josh: I think you're right in the sentiment. I'm not sure Claude is going to be the one to get there.
14:41
Josh: And that's kind of where I start to find little pieces of doubts in this thesis
14:45
Josh: is, well, Claude's great for code and code has a huge marketplace,
14:48
Josh: but there's a lot more in the world that will be impacted by AI that is not written by code.
14:53
Josh: Like in the world of atoms, like any sort of visuals that we're doing,
14:58
Josh: anything that's not built by code.
14:59
Josh: Claude isn't really there. But is that a problem? Probably not.
15:03
Josh: I mean, hey, what was the number? 80% of 50%?
15:06
Josh: That is a huge number and the revenue isn't slowing down.
15:10
Josh: So maybe what we're seeing here, and I think this is probably what we're seeing
15:13
Josh: here, is again, we talk about this all the time, the narrowing of focus and
15:16
Josh: scope in these AI companies where Google has a real world reality model.
15:20
Josh: Claude is focusing highly specifically on coding. And right now, they're the best by far.
15:27
Ejaaz: Maybe Anthropic ends up providing the fuel for this next iteration or evolution of UI, right?
15:35
Ejaaz: So maybe they won't end up building the user interface that people interact
15:40
Ejaaz: with or whatever app surfaces this to people, but they supply the code.
15:45
Ejaaz: They supply the engine to kind of harness and power this. And hey,
15:48
Ejaaz: that's like a very, very valuable thing, right?
15:50
Ejaaz: If they can create their kind of like own market monopoly on a layer between
15:55
Ejaaz: GPUs and AI apps and they sit in between them and they funnel everything between
16:00
Ejaaz: them just through their own coding model.
16:02
Ejaaz: That's a pretty valuable company as is. The other thing that really impressed
16:06
Ejaaz: me about what they've been able to build, Josh,
16:09
Ejaaz: is this coding model was built without the billions and billions of dollars
16:15
Ejaaz: worth of resources that Google and OpenAI has.
16:19
Ejaaz: Remember, OpenAI is a $500 billion company or private company.
16:23
Ejaaz: It's the most expensive private company that we know of right now that's been
16:27
Ejaaz: like officially confirmed.
16:28
Ejaaz: And they have so much in funding, capex funding from their investors, right?
16:32
Ejaaz: Google, on the other hand, is a cash cow. They get so much money from all their
16:37
Ejaaz: other profitable businesses that they could just plow it into AI and make all
16:41
Ejaaz: of the AI apps free, right?
16:42
Ejaaz: And we saw that with Gemini 3 Pro coming out, that they just smashed all benchmarks.
16:46
Ejaaz: Despite that, Anthropic, with lesser funding, with lesser support,
16:50
Ejaaz: was able to pull out a better coding model.
16:53
Ejaaz: And what's impressive about this now, Josh, is that they're being funded a hell of a lot more.
16:59
Ejaaz: In fact, they're starting to make some pretty big expenditures.
17:03
Ejaaz: This news was released this week
17:05
Ejaaz: where they signed a $52 billion deal to purchase 1 million Google TPUs.
17:12
Ejaaz: That's Google's version of their own GPUs.
17:14
Ejaaz: And the reason why this is super interesting is TPUs specifically are built
17:20
Ejaaz: for very specific purposes, very specific functions.
17:25
Ejaaz: An anthropic specific function, as we've been saying repeatedly on this episode, is coding.
17:31
Ejaaz: I see this as a direct means for them to extend their lead and strengthen their
17:36
Ejaaz: moat as the leading coding AI model provider.
17:39
Ejaaz: This is pretty impressive when you take it to the fact that it's not just Google
17:44
Ejaaz: getting involved in this, Josh.
17:46
Ejaaz: It's also all the other Game of Thrones participants. You've got NVIDIA and
17:50
Ejaaz: Microsoft investing $15 billion.
17:54
Ejaaz: I have a crazy fact for you, Josh. What do you got? Guess the percentage of
17:59
Ejaaz: ownership that Google has of Anthropic.
18:03
Josh: Just guess. Wait. So, again, I cheated in the research for this.
18:06
Josh: It's outrageous. 14% of the company is owned by Google.
18:10
Ejaaz: Yes.
18:10
Josh: That is so much of the company. Which is funny because, I mean,
18:14
Josh: Anthropic just wired Google $52 billion for a TPU order to cut their influence bill by two thirds.
18:20
Josh: So, again, like you said, this is very incestual. There is a lot of incest going
18:23
Josh: on. In fact, there's another person on the cap table, which I also learned, which is fascinating.
18:28
Josh: And that company is named Amazon, who owns somewhere between 15 to 19 percent of Anthropic.
18:34
Josh: So between Amazon and Google, one third, they own one third of Anthropic,
18:39
Josh: just about, which is an outrageous amount.
18:41
Josh: And it made me laugh because, again, so much of this industry,
18:45
Josh: so much of this AI bubble,
18:48
Josh: for lack of a better word, is built on top of these reoccurring Lego blocks
18:53
Josh: where every house in the Game of Thrones, it's all very incestual.
18:56
Josh: They're all related. They all have interpersonal relationships.
18:58
Josh: They all have a vested interest in the other person succeeding.
19:01
Josh: So when Anthropic is wiring,
19:03
Josh: $52 billion to Google. Well, Google owns 15% of the company already.
19:08
Josh: And they're going to help Anthropic cut their inference bills by two thirds
19:12
Josh: because TPUs are very efficient.
19:13
Josh: And like, it's all this self-fulfilling loop, hopefully upwards in a rising
19:18
Josh: tide that lifts all boats.
19:20
Ejaaz: It's really funny. Anthropic is kind of like the golden child that's getting
19:24
Ejaaz: spoiled by his parents and all the aunties and uncles. You've got NVIDIA that's
19:28
Ejaaz: saying, hey, please take my GPUs.
19:31
Ejaaz: In fact, do you want to know a stipulation of the $5 billion investment in Anthropic
19:35
Ejaaz: that Jensen specifically put forward to Dario?
19:38
Ejaaz: It was the fact that they would dedicate an entire NVIDIA team towards designing
19:44
Ejaaz: their next model of Claude.
19:46
Ejaaz: So Claude5 is going to be co-signed by NVIDIA because he was so anxious about
19:52
Ejaaz: Google TPUs taking over his market share for Anthropic specifically, right?
19:57
Ejaaz: Then you got Google coming in with 1 million TPUs, and then you got the Amazon
20:01
Ejaaz: investment, which means that their Tranium chips, which they just announced
20:04
Ejaaz: last week, is also going to train Anthropic.
20:06
Ejaaz: There is no means for them to lose from an infrastructure perspective.
20:10
Ejaaz: It's theirs to basically lose, and it's just an insane thing to see.
20:15
Ejaaz: But they're being even more aggressive. It's not just GPUs that they're focusing, Josh.
20:18
Ejaaz: Did you see this news leak from the information this week, I think they're going
20:23
Ejaaz: on a acquisition spree, dude.
20:26
Ejaaz: They're going to start acquiring a bunch of developer companies to further strengthen
20:30
Ejaaz: not only their coding AI mode, but the application layer that developers will end up using as well.
20:36
Josh: Yeah, okay. So this is interesting to me because what's the constraint that every company faces?
20:41
Josh: They don't have enough CPUs. They don't have enough compute,
20:44
Josh: whether it be energy or they can't get enough GPUs from Jensen.
20:46
Josh: They're compute constrained. It feels like Anthropic is kind of compute flooded.
20:50
Josh: They have an abundance of options they they don't feel constrained by compute so,
20:56
Josh: The question I'm asking myself is if like, hey, if you give the king of coding
21:00
Josh: a mountain of chips and ask what happens, well, my bet is they just continue
21:03
Josh: to solidify the coding lead.
21:04
Josh: And as they do this, they will continue to climb the stack into probably more
21:09
Josh: enterprise apps, more directly up against other companies like OpenAI. So this is their way in.
21:14
Josh: Are they going to compete on a broad scale across all the features? No.
21:18
Josh: But as a focus on code, if they continue down this route, I mean,
21:22
Josh: that monopoly on coding tokens is like pretty serious. And as the number of
21:26
Josh: tokens goes up every quarter, I mean, like Dario said, they're on pace to $70 billion of revenue.
21:33
Josh: That's like a tremendous amount of money. So it seems like they're in a really good spot.
21:38
Ejaaz: Okay, so if I were to get a gut check between the both of us right now,
21:42
Ejaaz: Josh, I would say we're pretty bowled up, right? Would you agree?
21:46
Josh: I am bullish. And if I am bullish, then surely there's a way for me to see some
21:51
Josh: upside in this, right? Based on the rumor mill.
21:53
Ejaaz: Well, that's the puzzle piece that's been missing for so long,
21:57
Ejaaz: right? Like it's one thing for our listeners to listen to this and be like, okay, great.
22:00
Ejaaz: Anthropik's an amazing company, but it's another thing to be like,
22:02
Ejaaz: well, I can't invest. They're a private company.
22:04
Ejaaz: Well, the rumor mill has been churning as it always does every single week.
22:10
Ejaaz: And apparently Anthropik is filing for an IPO as soon as next year.
22:15
Ejaaz: Now, some have even said, Q1 of next year. I think they are probably too bullish.
22:19
Ejaaz: But somewhere in the kind of realm of 2026 is where we might see one of the
22:25
Ejaaz: biggest IPOs. And that's not a phrase that I came up with.
22:28
Ejaaz: That's coming directly from the Financial Times.
22:30
Ejaaz: Dario is separately trying to raise and confirm a $300 billion private valuation
22:36
Ejaaz: of Anthropic with the aims of going public.
22:38
Ejaaz: If it does, Josh, I think this might be one of the best AI investments that
22:44
Ejaaz: we've seen in the recent decade, right?
22:46
Ejaaz: We haven't even got an OpenAI IPO. And I think they might even IPO later.
22:49
Ejaaz: And there's a few reasons why Anthropic is going to IPO as soon as this.
22:53
Ejaaz: And it's, well, actually, it's one main reason.
22:56
Ejaaz: They're becoming profitable way sooner than any other AI lab.
23:00
Ejaaz: Like you said, they're projecting $70 billion by the end of 2027.
23:04
Ejaaz: And the main reason why they've been able to do this is because they've dominated
23:07
Ejaaz: the enterprise AI market.
23:09
Ejaaz: So if all things go on and they keep on following this pattern of 10Xing their
23:13
Ejaaz: revenue for the last three years, and they do it for another fourth and fifth
23:16
Ejaaz: year, then they're the likely candidate for a crazy IPO.
23:19
Josh: Yeah, it's like if you missed AWS and you missed early NVIDIA,
23:23
Josh: then Anthropic really is one of the only shots left to own a pure play in AI
23:28
Josh: infrastructure and the coding story at scale.
23:32
Josh: Like this is really one of the earlier opportunities to get in at any of these
23:35
Josh: companies under a trillion dollar valuation, which is outrageous to say,
23:39
Josh: but that's probably the reality.
23:41
Josh: And I guess if I'm thinking about how to evaluate this, whether I would want
23:44
Josh: to invest or not, it's if you're evaluating Anthropic on their ability to create
23:49
Josh: a coding model. They're fantastic.
23:51
Josh: Revenues are high. They have a clear path to expanding. I mean,
23:54
Josh: $70 billion is an outrageous number for a company this young.
23:59
Josh: Outside of coding, it's not that interesting. I'm not sure where the applications
24:03
Josh: lie, but coding is so large.
24:05
Josh: And again, I'm trying to wrap my head around 80% of 50% of all tokens generated.
24:11
Josh: It's like such a huge number. And so much of the world runs on code that having
24:15
Josh: a pseudo monopoly on that for now is a really big opportunity and i guess like it,
24:21
Josh: Depending on the valuation, I mean, like OpenAI had a trillion. That seems high.
24:25
Josh: Anthropic, maybe lower. I mean, they're planning to raise the $300 billion.
24:28
Josh: What's the multiple on that prior to IPO? I don't know.
24:30
Josh: But it's going to be interesting. Ejaz, what is your take on the public offering?
24:35
Ejaaz: Well, before we do it, do you want to hear another crazy stat,
24:38
Ejaaz: Josh? You've been loving the 80% and 50%, right?
24:41
Josh: That's such a huge number. Because the number of tokens is going straight vertical.
24:46
Josh: And if they're owning a majority of that, that's huge.
24:48
Ejaaz: Well, not only that, like the percentage of those tokens that are just going
24:51
Ejaaz: towards Vibe coding is increasing.
24:54
Ejaaz: It 5X'd over the last 12 months from 12% to 50%.
24:57
Ejaaz: But get this, if you look at just clawed code, siloed as its own product,
25:03
Ejaaz: it is the new quickest product to reach $1 billion in annual recurring revenue.
25:09
Ejaaz: It did it in under six months.
25:11
Ejaaz: That title now is owned by Anthropic.
25:14
Josh: Yeah, that's insane. And the numbers are not going, they're not slowing down.
25:19
Josh: That was the thing, like that 1 million to 100 million 100 to a
25:22
Josh: billion billion to 10 10 to 70 it's outrageous
25:25
Josh: they have i mean according to that number 40 percent of all tokens generated
25:29
Josh: they don't need to spend as much money because they're getting the google tpus
25:32
Josh: and tpus are two-thirds more efficient than gpus so they can do this all at
25:36
Josh: a lower cost and they have this efficiency and the company is it's presumably
25:39
Josh: smaller at scale in terms of employees and it is this whole perfect storm,
25:43
Josh: if you are into code and if you are bullish on code for Anthropic.
25:47
Ejaaz: But Josh, what's the bear case for this? Like, how will Anthropic...
25:52
Ejaaz: Like, this sounds all great if it pans out like it is.
25:55
Ejaaz: But I feel like there's a few obstacles that Anthropic kind of needs to navigate
25:59
Ejaaz: before it kind of, like, fulfills that sentiment. Do you have a bear case for us?
26:04
Josh: I mean, the bear case for me, at least, is that they're just good at code.
26:08
Josh: You don't really want to use Cloud for anything else. Maybe writing.
26:11
Josh: But again it doesn't have they
26:14
Josh: have no diffusion model they have no real world physics engine
26:17
Josh: they don't really process things multimodally they don't
26:20
Josh: do images or videos or voice really and
26:23
Josh: they kind of have small signs of this here and there but they are really hyper
26:26
Josh: fixated on this one thing and that is to their benefit because they earn this
26:30
Josh: monopolistic advantage in the marketplace but in the case that there is an innovative
26:34
Josh: breakthrough let's say ilia who works at safe super intelligence basically a
26:38
Josh: funded research lab that has raised tons of money with no product.
26:41
Josh: Their only job is to come up with unique innovations in how you create models
26:45
Josh: to give you better outputs.
26:46
Josh: And if a company like Ilias comes out with the breakthrough that is able to
26:51
Josh: leapfrog Anthropic in terms of how well it's able to code at a cost per token that's lower.
26:57
Josh: I mean, that completely shatters the business model. So there is a lot of fragility built into this.
27:02
Josh: The upside to this is that, well, they're a enterprise company.
27:07
Josh: They're not really selling to us i mean sure their their
27:10
Josh: users are developers and coders but they're oftentimes you can
27:13
Josh: think of kind of like an aws their their enterprise sales software it's
27:16
Josh: going to companies and companies are slower to to move
27:19
Josh: and to pivot so they are building this moat so i think the bear case is probably
27:22
Josh: that coding isn't a sticky moat and they're one breakthrough away from having
27:27
Josh: this moat crumble but in the case they're able to retain this moat which is
27:31
Josh: 40 of all tokens like they are in really good shape going forward i'm curious
27:36
Josh: if you agree disagree where you're you You kind of stand on this.
27:39
Ejaaz: I agree with you on the delivery part.
27:43
Ejaaz: I'm not convinced that Anthropic is going to be the one to turn that vision
27:48
Ejaaz: of coding to ephemeral UI that we described earlier towards the end consumers, the end users.
27:55
Ejaaz: I don't think they're like a consumer product company.
27:58
Ejaaz: I think they're more developer-oriented and enterprise-oriented, as you described.
28:02
Ejaaz: But I don't think that's a bad thing at all, because the more that I think about
28:07
Ejaaz: it, Josh, I see coding as a very necessary substrate to build AI apps.
28:16
Ejaaz: So in the same way that NVIDIA GPUs, and they own the market monopoly on those,
28:21
Ejaaz: are the substrate needed to build the AI model in the first place,
28:24
Ejaaz: I think Anthropics coding capabilities,
28:27
Ejaaz: their coding engine, will be the substrate needed to build any and all future AI apps.
28:33
Ejaaz: And people will kind of like, as long as they maintain this monopoly,
28:35
Ejaaz: kind of paid the Pied Piper.
28:37
Ejaaz: They'll pay the toll in order to get access to frontier level coding AI.
28:42
Josh: And I guess at the very end of the limit, the final question would be,
28:45
Josh: is there going to be apps?
28:46
Josh: Like I think about OpenAI's hardware device next year, where there's no screen
28:49
Josh: and like, I don't know, it's gonna be weird.
28:51
Josh: But I think we could probably leave it at that where Anthropica is doing good things.
28:55
Josh: And this is kind of like the, the explainer as to why they are more important
29:00
Josh: than I think most people perceive, ourselves included.
29:03
Josh: I mean, prior to doing the research for this episode, I was much more bearish
29:08
Josh: on Anthropoc just because of the sentiment around them. They were kind of known
29:10
Josh: as the decelerationists.
29:12
Josh: And I didn't love that. They were super into safety.
29:14
Josh: And while that's still true, they are undeniably generating tremendous amounts
29:19
Josh: of real world value for people like us, for companies like Google and Amazon,
29:25
Josh: and for the whole world who writes code.
29:27
Josh: So in that, You have to admire what they're doing and the growth of this unbelievably impressive company.
29:33
Josh: I mean, that's really, that's all it is. That's Anthropik.
29:36
Ejaaz: Yeah, well said. Now, for the astute listeners of this show,
29:41
Ejaaz: some of this information might have sounded familiar to you,
29:44
Ejaaz: but that's because you read our essay in our record-breaking newsletter,
29:49
Ejaaz: which was released two weeks ago, where we wrote about the bull case for Anthropik.
29:55
Ejaaz: If you guys aren't signed up or subscribed to the Limitless Newsletter, you definitely need to.
30:00
Ejaaz: We'll link to it in the description below.
30:02
Ejaaz: And for the 80% of you that still
30:06
Ejaaz: aren't subscribed but watch this show week in week out it would mean the world
30:11
Ejaaz: to us if you subscribe and sign up um josh do you have any latest stats for
30:16
Ejaaz: subscriber count i saw that we were up like a couple hundred over the last couple of days
30:20
Josh: We're getting subs quickly so thank you for subscribing please if you have not
30:23
Josh: pressed the button and turn on the bell next to it which will it'll actually
30:27
Josh: notify you every time an episode comes out the three times a week we try to
30:29
Josh: keep them short if you if you watch all three episodes per week there is not
30:33
Josh: much that you're missing in the world of AI. You are caught up at the frontier.
30:36
Josh: In fact, you are even in front of the frontier because Ejaz,
30:38
Josh: like you said, a lot of times in the newsletter or sometimes in these episodes,
30:42
Josh: you will hear us talk about, and I'm very proud of this point because I'm like,
30:45
Josh: okay, we're actually doing it.
30:46
Josh: You'll hear us talk about these things prior to them being the mainstream news.
30:49
Josh: Like even just last week, we had our space episode about AI data centers.
30:52
Josh: And over the weekend, it became this huge fallout where Kathy Wood was talking
30:56
Josh: about it. Elon was talking about it. They were building new theses.
30:58
Josh: And we had just discussed the entire bull case for it earlier in the week.
31:02
Josh: So if you want to be early, if you want to be right, if you want to be on time,
31:04
Josh: or at least, you know, we're going to try our best to be right.
31:06
Josh: But if you want to be early to the information, please subscribe to the newsletter,
31:10
Josh: subscribe to the podcast, do all the good things, share it with your friends.
31:12
Josh: And yeah, we'll be back again for the next episode in a few days. We'll see you guys then.