Latest / In the Loop AI / AI Bubble Update: Kimi K3 Sends China into Chaos

Transcript

AI Bubble Update: Kimi K3 Sends China into Chaos · In the Loop AI

0:00
Josh: China just released KimiK3, the most powerful model to ever come out of the
0:03
Josh: country. The question on everyone's mind is, is this going to create a bubble
0:07
Josh: or is it already happening in front of us?
0:09
Josh: KimiK3 is the most impressive model leapfrogging the frontier American models
0:14
Josh: and some benchmarks and doing so in a way that is significantly cheaper and open source.
0:20
Josh: So the conversation over the entire weekend is not only is there a bubble,
0:23
Josh: but how credible, how strong is the American moat as it comes to China.
0:27
Josh: We always discuss the timelines and how America is normally ahead by,
0:31
Josh: I mean, let's say six to 12 months.
0:33
Josh: It seems like that timeline has kind of shrunken a little bit.
0:36
Josh: And you see how these two countries are kind of juxtaposing the way they're
0:39
Josh: going about it. There was two speeches that happened on Friday.
0:41
Josh: One was from America blaming China on rigging votes. And the other was China
0:46
Josh: very strongly presenting themselves as the up and coming AI leader.
0:49
Josh: So there's a lot to unpack here, probably starting with the market,
0:52
Josh: which was nothing but red over the last couple of days because you have to assume
0:57
Josh: there is a little bit more uncertainty than normal as it relates to
1:00
Josh: just ai models in general and how how it turns into china i mean look at this
1:04
Josh: look at the screen share that is a wall of red
1:07
Ejaaz: Yeah, so last week was basically China's retaliation to the American frontier
1:12
Ejaaz: models, and it happened in a pretty big way.
1:14
Ejaaz: As you mentioned, Moonshot Labs, which was the creator of this model called
1:18
Ejaaz: Kimi K3, which is completely open source, is now the largest open source model ever.
1:22
Ejaaz: It comes in at around 2.8 trillion parameters, and it's super cheap to use when
1:27
Ejaaz: you compare it to Fable 5 and GPT 5.6.
1:29
Ejaaz: Now, the question on everyone's mind is, is it necessarily good enough?
1:33
Ejaaz: If you look at the NASDAQ market and how it responded last week,
1:36
Ejaaz: it shaved off hundreds of billions
1:38
Ejaaz: of dollars, like it was down completely in response to this model.
1:42
Ejaaz: The reason being is, it doesn't matter if it competes with Fable 5 or if it's
1:46
Ejaaz: as good as GPT 5.6, what matters is the cost.
1:49
Ejaaz: When you look at Fable 5 specifically, you're paying around 50 to 56 bucks per million tokens input.
1:57
Ejaaz: And when you compare that to Kimmy K3, you're paying around 50 cents to a dollar.
2:00
Ejaaz: So if you're an enterprise that is burning tens to hundreds of millions of dollars,
2:04
Ejaaz: or if you're a software engineer that is spending all your tokens on an expensive
2:07
Ejaaz: $200 a month membership, you might want to consider an alternative model.
2:11
Ejaaz: And in this case, it'll be a Chinese model. Now, the craziest part about this
2:16
Ejaaz: is the valuation of Moonshot Labs in January of this year was $4.3 billion.
2:24
Ejaaz: Now, after they released Kimmy K3 just last week, they announced that they're
2:28
Ejaaz: raising at a new valuation of $30 billion.
2:31
Ejaaz: That is still 30x cheaper than OpenAI's $1 trillion valuation.
2:37
Ejaaz: And Anthropic is valued at the same amount. So the question in everyone's mind is,
2:42
Ejaaz: are the American labs overpriced or are the Chinese labs underpriced?
2:46
Ejaaz: We're going to have a series of Chinese IPOs coming from the AI labs,
2:50
Ejaaz: such as DeepSeek and Moonshot over the next roughly six months or so.
2:54
Ejaaz: And the Chinese president Xi Jinping reinforced this in a very big way on Thursday,
2:59
Ejaaz: where he took the stage in Shanghai in person at the World AI Conference,
3:03
Ejaaz: the first time since 2018 to make his first public commentary on the AI landscape.
3:08
Ejaaz: And he had some pretty major takeaways for everyone.
3:11
Josh: Yeah. Like you mentioned, this is the first time since 2018 taking the stage.
3:14
Josh: This is a fairly big deal. And his kind of thesis could be summarized in one
3:18
Josh: line, which is AI should not be a solo performance by a single country.
3:22
Josh: It should be a symphony of international cooperation, followed by a shot at
3:27
Josh: Washington, which said we should jointly oppose overstretching the national
3:31
Josh: security concept in the field of AI or placing one country's security over that of others.
3:37
Josh: And it was really interesting. I mean, I mentioned this in the intro briefly
3:40
Josh: to see that kind of juxtaposed against the other speech that was happening simultaneously
3:44
Josh: on Friday night, but back in Washington, where President Trump,
3:47
Josh: he went on to accuse Beijing of the largest compromise of election data in history.
3:51
Josh: So while one president is standing up in front of the AI conference for the
3:55
Josh: first time, offering the world basically free frontier intelligence through
3:58
Josh: these open source weights, another one is complaining that we've been robbed and
4:03
Josh: That we're not simultaneously while this kind of lead that we have is continuing to shrink.
4:08
Josh: And there's this post he says, I shared it with you just before recording.
4:10
Josh: I loved it. It was from Travis Kalanick.
4:12
Josh: And for those who don't know, Travis Kalanick, founder of Uber,
4:15
Josh: he's the guy who has famously went into China and convinced them somehow to
4:19
Josh: allow him to let robo taxis into this country is very familiar with how they were culturally.
4:24
Josh: He said, in my experience, China rules different on these topics.
4:27
Josh: If you're in China and you tell a business colleague, dude, they frauded me, they took my stuff.
4:31
Josh: The natural response is, why'd you let them get your shit together?
4:34
Josh: And I think this is kind of beginning to show the two differences in cultures.
4:38
Josh: We know that China doesn't have any regard for IP and therefore is able to distill
4:42
Josh: the models, is therefore able to rip off all of our IP. And that's how they're
4:45
Josh: able to accelerate so well.
4:47
Josh: I think my question now, Ejaz, is are they actually going to ever be able to
4:51
Josh: flip the United States in terms of capabilities if they continue to just distill and follow along.
4:57
Josh: It feels like they can maybe get that lead closer and closer,
5:00
Josh: but in order for them to actually break through and create something better,
5:03
Josh: it will require some technical breakthrough that they haven't quite shown.
5:07
Josh: So when we look at the market, does it feel overreactive that it's down so much,
5:12
Josh: even though they don't actually have a frontier model and getting a true frontier
5:15
Josh: model would imply a lot of breakthroughs that they've never been able to do.
5:18
Josh: I mean, they can't even serve this model yet. They're overflowed with just Kimmy requests.
5:22
Ejaaz: Yeah. And I think what you're referring to in that last sentence is Moonshot
5:26
Ejaaz: Labs announced over the weekend that they don't have enough GPUs to support
5:31
Ejaaz: everyone's signup and request.
5:33
Ejaaz: So they've had to pause new subscriptions. So it's an over-demand.
5:36
Ejaaz: And I actually have this really cool chart from the folks at Open Router who
5:42
Ejaaz: basically allow anyone and everyone to get access to these open models as well
5:46
Ejaaz: as American frontier models.
5:47
Ejaaz: And it shows that 58%, near 60% of users in America now are using open source Chinese models.
5:55
Ejaaz: So we're now flipped from using American frontier models in the majority of
5:58
Ejaaz: token expenditure to Chinese models, which is very significant.
6:01
Ejaaz: But to answer your question more directly, Josh, it depends how you define what a breakthrough is.
6:08
Ejaaz: Now, a breakthrough in itself is Kimi K3. It may not be Fable 5 worthy,
6:13
Ejaaz: but the The fact that it's been able to catch up to within like,
6:16
Ejaaz: you know, striking distance of Fablefy, but not quite as good,
6:20
Ejaaz: but at a fraction of the cost,
6:22
Ejaaz: could be considered a breakthrough. If you look at their white paper,
6:26
Ejaaz: you've seen that there's a bunch of new research things that they've been able to figure out.
6:29
Ejaaz: I think they call it Kimi Delta, attention attenuation, or whatever the hell
6:32
Ejaaz: that's called, which allowed them to efficiently scale two and a half times
6:37
Ejaaz: more efficiently than any other frontier lab has done.
6:40
Ejaaz: Now, if you look at America, they haven't been able to do this.
6:42
Ejaaz: What they have been able to do is throw tens to hundreds of billions of dollars
6:46
Ejaaz: at the problem and hope that it would fix itself.
6:50
Ejaaz: So more efficiently for a much cheaper cost.
6:54
Ejaaz: Now, if you ask me if the American labs stopped putting out new models,
6:58
Ejaaz: so let's say we never got anything later than Fable 5 from Anthropic going forwards,
7:03
Ejaaz: I think it would take Chinese labs significantly longer to catch up with them
7:08
Ejaaz: simply because they have been distilling. You know, you talk about their disregard for IP.
7:11
Ejaaz: That is primarily the way that they've been able to catch up to America frontier
7:15
Ejaaz: models. Now, my question that I throw back to you maybe, I'm curious what your
7:18
Ejaaz: thought is, is distillation necessarily malpractice?
7:22
Ejaaz: I think everyone is kind of doing it, like I looked at,
7:27
Ejaaz: Thinking Machine Labs last week, which is Mira Morati, ex-OpenAI CTO's new lab.
7:34
Ejaaz: And they announced this new open source model. It's their flagship model.
7:36
Ejaaz: It got a lot of attention.
7:38
Ejaaz: Every American spokesperson was like, hey, American open source is so back.
7:42
Ejaaz: Look at this new cool model.
7:44
Ejaaz: Now, if you look at the white paper, they basically distilled DeepSeq v4's architecture from China.
7:49
Ejaaz: So we're kind of like all distilling and copying everyone's kind of trends and
7:52
Ejaaz: research to collectively make better models.
7:54
Ejaaz: And I don't necessarily know if it is a bad thing. The final thing I'll say
7:58
Ejaaz: is Fable 5 was restricted and banned for a while.
8:02
Ejaaz: It's only actually been fully live for like three weeks right now.
8:05
Ejaaz: There is no way Moonshot Labs was able to fully distill the model and create
8:09
Ejaaz: Kimi K3 in that shorter time span. So they're doing something different, is what I'm saying.
8:13
Josh: Yeah, well, Fable is like one model. I'm sure they have been doing this with others.
8:18
Josh: And I think when it comes to research, like algorithmic improvements,
8:21
Josh: similar to what Murmurati's company was doing, it's a little bit different than blatant IP
8:26
Josh: copying, where like, sure, there are these technical techniques,
8:29
Josh: like the transformer that everybody uses, but then there is the actual IP built on top of it.
8:34
Josh: And that is where I mean, it is just illegal to do these things is illegal to
8:37
Josh: steal other companies information like this.
8:40
Josh: And it's obvious that China doesn't have any regard for this.
8:44
Josh: So I would say it is different in the sense that this is just like blatant IP
8:48
Josh: ceiling. And while it is sure a gray space and a lot of companies distill from a lot of places.
8:53
Josh: This is just blatant, illegal...
8:56
Josh: Disregard for copyright. And I think that is like where you can comfortably draw a line.
9:00
Josh: If you are the United States, I worry that perhaps there is an over-indexing
9:04
Josh: coming just before we were recording, Ejaz, you mentioned there's a possibility
9:07
Josh: that open source models are going to start to have some pushback,
9:10
Josh: perhaps from the government.
9:12
Josh: There is some resistance happening towards that.
9:14
Josh: And should that happen, I do worry what the world looks like then,
9:19
Josh: because when you think about the costs of these things, like frontier intelligence
9:22
Josh: models, You're paying, what, $10 to $50 per million tokens.
9:27
Josh: And for these frontier open source models, these Kimi models,
9:30
Josh: you're paying, what, $1, $0.50, maybe $2.
9:33
Josh: Yeah, $0.50. And it's like that, if those start to get banned here in the United
9:39
Josh: States, you're almost in a way forcing consumers to pay a significantly higher
9:43
Josh: price for similar levels of intelligence.
9:46
Josh: Like if you're paying a 50 times multiple for a model that is perhaps 10% better,
9:51
Josh: that seems like it is problematic for just general users in the United States
9:55
Josh: who don't want to pay perhaps $50 per million tokens, but they also want access
9:59
Josh: to close to the frontier.
10:00
Josh: There isn't really an option for them right now. And banning these open source
10:04
Josh: models also doesn't seem like a good place to be either.
10:07
Ejaaz: I think people are just tired of overpaying and they're starting to kind of
10:11
Ejaaz: like look at their wallets and budget spends and thinking, well,
10:14
Ejaaz: maybe if there is a sexier alternative, I'll go use it. I don't care if it's
10:18
Ejaaz: Chinese, it's open source after all.
10:20
Ejaaz: Now, it hasn't escaped me that the irony of China being the open source proponents,
10:27
Ejaaz: the fact that in Xi Jinping's speech over the weekend, he basically said like,
10:32
Ejaaz: hey, we need to be openly collaborating with each other.
10:35
Ejaaz: This can't be a solo kind of like country thing. And we also can't be imposing
10:38
Ejaaz: export restrictions. We've just got to just got to let people build, man.
10:41
Ejaaz: Compare that to like Trump's America or rather, you know, capitalist America,
10:46
Ejaaz: who is supposedly meant to be proposing the same kinds of things,
10:49
Ejaaz: but they're actually like, you know, we might actually end up banning the Chinese
10:52
Ejaaz: open source models. Like, it's just ironic to me.
10:56
Ejaaz: If you're looking at this objectively, the question that you should be asking
10:59
Ejaaz: is, are these models actually good? Like, what are they actually good for?
11:03
Ejaaz: If they're not Fable 5 worthy, like, what can I be using it for?
11:06
Ejaaz: And I think there's a really good example from David Sachs over the weekend here where he goes,
11:10
Ejaaz: Kimi K3 just fixed 15 critical security bugs that Codex and Fable refuse to
11:16
Ejaaz: because of cyber guardrails.
11:18
Ejaaz: And over the weekend, I was using Kimi K3 quite a bit. And I will tell you,
11:22
Ejaaz: the number one feature has got nothing to do with how smart it is,
11:25
Ejaaz: or how long it is, or how cheap it is.
11:27
Ejaaz: It's the fact that I can ask it a question, and it just answers,
11:31
Ejaaz: or I can give it a task, and it just does it. There is no relegation to a lower class model, Kimi K2.
11:37
Ejaaz: It just kind of like does the thing. If I did the same thing for Fable 5,
11:40
Ejaaz: if I did the same thing for GBD 5.6, there is a very high probability that they
11:44
Ejaaz: would just say, sorry, I can't answer that.
11:47
Ejaaz: Or sorry, I'll give you a cheaper, crappier model to do your thing. And it's quite freeing.
11:52
Ejaaz: It's kind of like the feature in itself. Like I don't really care about the
11:54
Ejaaz: benchmarks. It's the fact that I have a near enough really good model that can
11:57
Ejaaz: just do the work. Is it slower? Yes. Does it use more tokens?
12:01
Ejaaz: Yeah. Is it as cheap as I could have hoped?
12:03
Ejaaz: Maybe not, but I don't really care because it's unrestricted.
12:06
Ejaaz: I don't know if that makes sense, but like, yeah, that's my experience.
12:09
Josh: Yeah. And this is the difficult part too. It's because like that wasn't really
12:13
Josh: always the intention either.
12:14
Josh: It's like if they wanted to release this model in the public,
12:17
Josh: the government would not allow it without these safeguards.
12:19
Josh: And they tried and it got banned for almost a month. And that's,
12:21
Josh: that's part of the problem that I think a lot of these labs,
12:23
Josh: and that's kind of what David Sachs is referencing now,
12:26
Josh: is that if you have a frontier model, you are now burdened by the approval process
12:30
Josh: of the U S government before comes out, which will oftentimes include these
12:33
Josh: restrictions that do set things back.
12:36
Josh: And the reality is, is that, I mean, Kimi K3 is basically close to frontier
12:42
Josh: level cyber capability, and it's going to be freely downloaded soon.
12:45
Josh: And if it's not Kimi K3, it will be Kimi K4 in a couple of months.
12:48
Josh: And the reality is, is that while these safeguards are very important to have
12:52
Josh: when they are truly at the frontier, the moment it becomes freely and openly
12:56
Josh: available elsewhere, it loses all the advantage and then therefore puts everyone
13:00
Josh: who is burdened by those safeguards at a disadvantage. And that's this really difficult thing is
13:05
Josh: If you're the AI labs, you want safe intelligence that is aligned with our best human interests.
13:11
Josh: And you also want to offer people the best tools possible.
13:14
Josh: And the US government is saying, well, this is great, but they need to be safe.
13:17
Josh: We need to approve these things.
13:19
Josh: And all the while, this is slowing things down. I mean, how long did it take for us to get GPT 5.6?
13:23
Josh: It was sitting there ready for many, many weeks until they approved it.
13:27
Josh: And during that time, we got Kimmy K3. And there's this really difficult disconnect
13:34
Josh: that's getting hard to wrap my head around.
13:35
Josh: As it relates to the Chinese frontier labs, just kind of burning, scorching the earth.
13:41
Josh: I mean, this was very much Meta's plan when they first were trying to come up
13:44
Josh: in the world of AI, where they were just going to scorch the earth,
13:46
Josh: create the best open source models, distribute it to everybody,
13:49
Josh: and lower the margins of the frontier labs.
13:51
Josh: And the reality is, is that China's just kind of taking that and grabbing it
13:54
Josh: by the horns and running with it.
13:56
Josh: And I'd say doing a pretty good job so far.
13:59
Ejaaz: Yeah. Okay. So something that we have discussed a lot on this show, Josh, is valuations.
14:05
Ejaaz: It's ipos it's generating wealth let's talk is there a bubble or not you just get into it,
14:11
Ejaaz: well okay yeah there's definitely a bubble the question is,
14:14
Ejaaz: is should the bubble be popped and this is a very visceral example that we're
14:18
Ejaaz: showing on the screen right now right so you have opening eyes valuation at
14:21
Ejaaz: the top which is near a trillion dollars
14:23
Ejaaz: and then you have deep seek which is a fraction of that at 71 billion dollars
14:28
Ejaaz: and then moonshot this week who got a valuation upscale to $30 billion.
14:33
Ejaaz: So it's sitting at 30x less the price of OpenAI, despite being near frontier
14:38
Ejaaz: capability of their flagship model, GPT 5.6.
14:41
Ejaaz: Now, we know that there is a ton of frontier American IPOs that are happening
14:46
Ejaaz: by the end of this year. You've got OpenAI, you've got Anthropic,
14:50
Ejaaz: and you've got a few more chip companies as well.
14:52
Ejaaz: And then you've compared that to the Chinese IPO.
14:56
Ejaaz: Aiming for a valuation that is considerably less than that, you've got Moonshot,
15:00
Ejaaz: Kimike2, Kimike3, sorry, doing the same exact thing.
15:04
Ejaaz: And the question is, which one is overpriced or underpriced?
15:07
Ejaaz: Now, my guess, if I had to have a take here, would be that,
15:11
Ejaaz: I think the Chinese frontier labs are going to bring the valuation of these
15:16
Ejaaz: open AIs and anthropics down in the medium term. And what I mean by that is
15:21
Ejaaz: they have proven that you can get frontier capability for a fraction of the cost.
15:26
Ejaaz: But the premium that sits with American frontier labs sits in the research,
15:31
Ejaaz: sits in the talent, and sits in the fact that whoever is closest to building
15:35
Ejaaz: self-improving models will end up capturing quite a bit
15:40
Ejaaz: of future economic valuation that any of these AI models create.
15:43
Ejaaz: And right now, the best bets, if you had to place it on Polymarket or wherever,
15:47
Ejaaz: would be with Anthropic and OpenAI. So it's very important. You can't just look at models right now.
15:53
Ejaaz: That's just a snapshot. You need to be thinking of Fable 6. You need to be thinking of GPT-6.
15:58
Ejaaz: Can those models build themselves and build the next versions of themselves?
16:02
Ejaaz: If the answer is yes, then you could justify some of these valuations.
16:05
Ejaaz: But it is a bit of a leap. It's a bit of a thought boy kind of take,
16:09
Ejaaz: I guess. I don't know. Do you think differently?
16:11
Josh: Yeah, no, it could be true that both OpenAI is overvalued and Moonshot is undervalued.
16:16
Josh: And the reality is, is that they should exist somewhere in between.
16:19
Josh: The big difference, the reason why we see this 28 times multiple on these different
16:23
Josh: companies is because one has a tremendous amount of revenue and a predictable
16:27
Josh: future amount of revenue, and the other does not. We just saw how Kimmy had a huge success.
16:32
Josh: They were unable to satisfy the demand. They had to shut down the signups of
16:36
Josh: new people using their product because they just don't have the GPUs.
16:39
Josh: So there is just not enough compute available to them in order to earn a higher
16:44
Josh: multiple at this moment.
16:45
Josh: They're offering it for free. There isn't predictable revenue drivers on it. It's just not possible.
16:50
Josh: OpenAI, on the other hand, is trading at a very high multiple on its revenue.
16:54
Josh: But the reality is, is that like you mentioned, it is somewhat close.
16:58
Josh: You have to imagine getting closer and closer to this like self-recursive learning
17:02
Josh: loop. And we have no idea what that's going to look like, But we can suspect
17:06
Josh: and we can hypothesize that the totals demand for tokens is going to continue to be up only.
17:11
Josh: Basically forever or for at least a very long time. And should that be the case,
17:15
Josh: there are only a few companies that are able to satisfy that demand,
17:19
Josh: let alone satisfy that demand at a high quality.
17:21
Josh: So it's really this race for GPUs. And it's funny because at the end of the
17:25
Josh: day, the real winner here is Jensen.
17:28
Josh: And I have a feeling this is why he's been so supportive of open source in general,
17:33
Josh: because it just, it requires a tremendous amount of GPUs. it makes AI more accessible to more people.
17:39
Josh: And as other companies begin to make their own custom ASICs or their custom
17:42
Josh: accelerators like Tranium and like the Google TPUs, there is going to be this,
17:48
Josh: again, increasing demand throughout open source communities of the GPUs
17:52
Josh: And I don't know if I'm looking at the market right now, I'm seeing things getting crushed.
17:57
Josh: I'm kind of, I'm kind of bullish. I'm kind of like, wait a second,
17:59
Josh: this is just a discount. None of the market sentiment, or I guess none of the
18:03
Josh: market forces have really shifted in a meaningful way.
18:06
Josh: The only thing that's changed meaningfully is the sentiment because people are
18:08
Josh: a little bit concerned about where China stands.
18:11
Josh: But the reality is to me is that while China has pretty great models that are
18:15
Josh: very efficient, very open source, they're still so far away from the real frontier
18:20
Josh: because the frontier so much more than just intelligence per token.
18:23
Josh: It's the infrastructure.
18:24
Josh: It is the sales teams. It is the business development. It is the reputation.
18:28
Josh: It's the American institutions trusting your product.
18:31
Josh: And so much of that is already baked into OpenAI, into Anthropic,
18:34
Josh: into these large AI labs that it doesn't seem to me like there is a problem.
18:37
Josh: Like, okay, these models are great, but they can't serve them.
18:40
Josh: They aren't going to pass the frontier.
18:43
Josh: And Anthropic and OpenAI, in the meantime, are just going to continue hurling
18:47
Josh: so fast forward with very predictable revenues that are going to,
18:50
Josh: I mean, expect it to go up only.
18:52
Josh: And I think that makes me feel like this is a pretty actually decent buying
18:55
Josh: opportunity and that maybe there is like a little bubble pop,
18:58
Josh: but that happened for 72 hours and now we're back on our way.
19:01
Josh: It seems to be like the market has become very resilient. And even though it's
19:05
Josh: short-sighted, there is this kind of weighing machine over a long period of
19:08
Josh: time that will work itself out to continue to guide upward, I think is the idea.
19:13
Ejaaz: Yeah, I feel the same. When I look at valuations, when I see how volatile the
19:18
Ejaaz: stock market is, I do think it is based on a lot of people's emotions.
19:22
Ejaaz: But the fact of the matter is, whether you end up with a frontier model that
19:26
Ejaaz: costs a lot of money, or if you end up with an alternative model that costs
19:29
Ejaaz: a fraction of that, people are going to use the thing.
19:33
Ejaaz: People who want frontier intelligence and can't afford the Kimi K3 intelligence
19:38
Ejaaz: deficit will end up using Fable five and paying for the extra cost because they
19:41
Ejaaz: get so much more, they make so much more money or they create so much more output on the back end.
19:46
Ejaaz: The people who are hobbyists, who are young founders and startups who can't
19:50
Ejaaz: afford the expensive models will use Kimi K3.
19:52
Ejaaz: It'll take longer, but they'll end up building amazing things.
19:55
Ejaaz: And Jevin's paradox will play out at the end of the day.
19:58
Ejaaz: People will want more tokens. They'll want to consume as much AI compute as
20:02
Ejaaz: they can for various different use cases. The winners at the end of the day
20:06
Ejaaz: are you who are listening to this. The winners at the end of the day are the
20:08
Ejaaz: businesses who want to use AI to accelerate their own purposes.
20:12
Ejaaz: And I think that if you're looking at where, you know,
20:16
Ejaaz: parking your money might make sense, again, like not investment advice,
20:19
Ejaaz: it would be kind of like at the bottom of the stack, the people that are creating
20:23
Ejaaz: the infrastructure, the people that are helping serve these tokens,
20:25
Ejaaz: the people that are helping power these GPUs, the people that are creating the
20:28
Ejaaz: memory to remember all the context that all these different models are going to create about you.
20:32
Ejaaz: The fact of the matter is, if you look at Trump, you look at his speeches and
20:35
Ejaaz: the policies that he's making. If you look at Xi Jinping, these guys are adversaries.
20:39
Ejaaz: He's got the opposite kind of take. The ultimate unifying thread is more consumption,
20:45
Ejaaz: more demand for more AI, more GPUs, more infrastructure. And that is technically.
20:51
Ejaaz: The sector or the companies that are going to end up winning.
20:53
Ejaaz: The most important Kimi announcement over the weekend is this one.
20:58
Ejaaz: Kimi K3 has received far more love than we expected and our GPUs are feeling
21:01
Ejaaz: it. They've stopped signing members up because they've run out of GPUs.
21:04
Ejaaz: Now, whether they're using Huawei's GPUs, which was announced at the same conference
21:08
Ejaaz: that Xi Jinping had a speech at, or whether they're using NVIDIA GPUs, the point is,
21:12
Ejaaz: infrastructure is the moat and that's where capital is headed so whether memory
21:17
Ejaaz: stocks are down whether these gpu stocks are down i think in the long term i'm
21:20
Ejaaz: i'm pretty bullish and i know i'm pretty bullish relatively on this show but.
21:24
Josh: I think we both are
21:25
Ejaaz: Please prove me wrong like i'm i want to see the counter thesis.
21:28
Josh: Yeah well listen it's been correct so far because i mean since the conception
21:31
Josh: of the show we have essentially been up only and the market forces continue
21:34
Josh: to point that way you know it's like there there is somewhat of a case to be made that perhaps the
21:40
Josh: there being just two AI labs isn't going to work out great for everyone else.
21:44
Josh: But the people who it works out for, like you mentioned, is the people who are
21:47
Josh: providing the infrastructure to build this out.
21:49
Josh: Because even in the case, let's say Kimi K4 is a true frontier model,
21:54
Josh: they're still going to need a huge amount of compute and all the other labs
21:56
Josh: are going to need a tremendous amount of compute to keep up.
21:59
Josh: And the thing with this competition is, like, so long as the demand is there,
22:04
Josh: the amount of resources required to supply that demand is just going to continue to be up only.
22:09
Josh: And regardless of these outcomes, regardless of the open source starting to
22:13
Josh: ship away at closed source or closed source maintaining its moat and just shooting
22:17
Josh: ahead through the self-recursive learning, both of those realities require
22:20
Josh: so much more compute, both to serve the larger models and both to keep up with this increased demand.
22:26
Josh: Clearly, China can't do that. So I'd say safe from disaster.
22:29
Josh: Once again, we've survived our second deep seek moment, perhaps.
22:33
Josh: It only took us 48 hours to kind of rebound and get back to it.
22:37
Josh: But we'll see what comes next. I mean, there are rumors, EJs,
22:40
Josh: and I don't want to speculate, but I have heard that Ilya is beginning to wake
22:44
Josh: up and has been receiving some pressure from investors to share what they've been working on.
22:49
Josh: Ilya Sitskever, yes, Safe Super Intelligence, that guy, co-founder of OpenAI,
22:53
Josh: he's been working quietly for how many years now?
22:56
Josh: And I'm seeing on X all the time. It's like, hey, the investors at this company
22:59
Josh: are kind of like getting a little restless.
23:02
Josh: They know what's going on, but the world doesn't know and they want to get that bag secured.
23:06
Josh: So I'm exceptionally excited for announcements like that, where perhaps there will be this huge sort of
23:13
Josh: Improvement in efficiency or algorithmic improvement or something new and novel
23:16
Josh: that can really push things forward in a big way so we are we are always a single
23:20
Josh: breakthrough away from this catastrophic
23:23
Josh: event either a catastrophic up or down we don't know but for now
23:27
Josh: safe from explosion of the bubble for now and again like i mean china's just
23:31
Josh: on their way and they're going to continue to keep firing with no regard for ip
23:36
Josh: or any any sort of american laws and i think that's kind of where we could leave
23:39
Josh: it is like hey kimmy k3 is a great model um you can't use it right now because
23:43
Josh: it's a little too good um but it's a great model nonetheless and and the american
23:48
Josh: markets felt it and hopefully it's um
23:49
Josh: it keeps everyone on their toes
23:50
Josh: moving forward very fast when it comes to iterating through these models
23:54
Ejaaz: Yeah that's pretty much it i think if if there was a single sentence that I
23:57
Ejaaz: would want to leave you folks with is we can't keep ignoring China.
24:01
Ejaaz: Like those days are gone. They are here to play in a very meaningful way.
24:05
Ejaaz: And they're leading the open source frontier, which means that,
24:07
Ejaaz: you know, you can own your AI at home. These, the cost of hardware is going
24:10
Ejaaz: to get a lot cheaper and you'll be able to have the privacy access and the cheap
24:15
Ejaaz: efficiency costs that comes with that.
24:17
Ejaaz: So keep an eye on the pulse, on the radar. If you can't do that,
24:21
Ejaaz: if you're not on X the entire time, hey, we've got this really fun podcast called
24:24
Ejaaz: Lemon Rhapsody. Subscribe to us.
24:27
Ejaaz: We do four episodes a week. We post twice on a newsletter to hundreds of thousands
24:31
Ejaaz: of folks to cover anything and everything that is happening at the frontier of AI and technology.
24:35
Ejaaz: So if you want to keep up to date, make sure you subscribe to us.
24:38
Ejaaz: Leave us a comment if you disagree with anything that was said today.
24:41
Ejaaz: Or if you agree, we'd love to hear from our fans as well.
24:44
Ejaaz: We're always looking for where we might be wrong or different investment angles
24:46
Ejaaz: and stuff like that. So please let us know in the comments.
24:49
Ejaaz: If you're listening to us on Spotify or Apple, again, please rate us,
24:52
Ejaaz: leave us a comment. It helps us out massively. And I think the final note I
24:55
Ejaaz: will say is Limitless has officially spun out. We are on our own.
24:59
Ejaaz: We're keeping the lights on at home ourselves.
25:01
Ejaaz: So if you are a company that has a product or service, or if you know of someone
25:05
Ejaaz: that has a product or service that you would like to share with our very palpable
25:09
Ejaaz: audience, please let us know. Reach
25:10
Ejaaz: out to us. Our email is linked in the description below or DM us on X.
25:15
Ejaaz: And I think that's it, Josh. Anything else?
25:17
Josh: That was pretty good. Dude, you got the outro dialed now. I think that covers
25:20
Josh: just about everything and then some.
25:22
Josh: Clearly we've been doing this for a lot. This is what, episode 210 now?
25:25
Josh: So yeah, if you've made it this far, you know the deal. Thank you guys so much
25:28
Josh: for watching and we'll see you in the next one.