Loading summary
Transcript55 lines
- [00:02]
A
Hello and welcome to another episode of the Capital Cycle podcast. This is Edward Chancellor and I have with me again Laura Fife, who's an emerging markets analyst with Marathon Asset Management. Welcome, Laura. Thank you. Now, Laura, perhaps the biggest economic story from China this decade has been the collapse of its epic real estate bubble. But there's another, probably more important story that you outline in your latest contribution to the Global Investment Review. Namely, that's the extraordinary technological advances being achieved across many different sectors in China. And in some ways this resembles Japan and Korea in the post war period, but it's on potentially a far larger scale. And you start your piece with example of how China has jumped on the AI bandwagon.
- [01:02]
B
Yes. So China's actually demonstrated what I find to be quite striking efficiency with respect to certain links in the AI value chain recently. In January last year, we saw quite a vivid reminder of how even the most impenetrable seeming of moats can be breached quite suddenly. That was when DeepSeq released its R1LM and Nvidia shed nearly 600 billion of market cap, which was at the time the single largest destruction of value or stock in history.
- [01:33]
A
And you cite some token numbers for the Deep Sea model which I don't quite understand. Can you explain them?
- [01:40]
B
Yes, yes. So the R1 was priced at $0.55 per million input tokens and about $2.19 per million output tokens. And it cost reportedly about $6 million versus incumbents billions of dollars to train and released open source. So what was disruptive about this, I guess, is that capability that was assumed to require vast capital talent that was really scarce and the best chips was actually replicated at one tenth of the cost and then given away.
- [02:13]
A
So what's interesting about that episode with the arrival of the Deep SEQ model is that as you say, Nvidia stock took a hit, but it's bounced back, as you know. And the pre market valuations of the US frontier models, OpenAI and Anthropic have gone through the roof, I think. Anthropic recently valued at close to a trillion dollars. But interest in Chinese open source AI models is picking up again because they're said to deliver around 90% of the performance of the Frontier models at about 10% or less of the cost. And we are coming to the end of something called token maxing in which employees were told to prove their AI skills by spending as much money on tokens as possible. Well, that era is over. So these cheaper Chinese open source AI models are looking increasingly attractive. Anyhow, back to Your piece. You're saying that advances are taking place in other areas, technological advances, say for instance in the auto business.
- [03:22]
B
Yes. So we've seen some examples of Chinese competition surprising incumbents elsewhere, especially in autos. So good enough or actually increasingly superior physical products are increasingly becoming available at a fraction of the price and coming to the market more quickly. In my essay I use the example of a Chinese carmaker, Cherry's JKU 7 model which has been nicknamed the Chinese Range Rover.
- [03:47]
A
Yeah, which we see instantly on almost every road. Have you noticed that driving along the
- [03:53]
B
JQ7 became the UK's best selling car in March and has a plug in hybrid range which is actually superior to the far more expensive Range Rover Evoque. So it's about a third cheaper than the Evoque and along with Chery's other brand, Omoda J Crecoo became the fastest emerging auto brand to hit a million units and did that in only three years. So both Deepsea and Cherry have quite recently illustrated just how rapidly and significantly the cost curves can be lowered as Chinese competition adds capacity. And this is something we've seen across time and industries at Marathon. But what's perhaps interesting now is the speed that it seems to be taking on and the higher up levels of the value chain that it might be.
- [04:41]
A
And you cite some other examples.
- [04:43]
B
Yes, I make the point that Chinese competition is done so some damage, especially in commoditized industries. So those ones where incumbents moats are mostly capital based. So some consumer facing examples of products that listeners might be familiar with as a result of this are Anker chargers, Xiaomi phones, DJI drones, anything pretty much from Temu and clothes from Shein. But on the more industrial side and in some newer applications, low cost Chinese capacity has also slashed pricing and shipping over the longer term out as well as solar and humanoid robots by up to about 90%. So comparable to those levels that Deepseq disrupted incumbent models by and with the disruption timeline like I mentioned, seeming to shrink in each of those instances.
- [05:33]
A
And China's competitive dominance in the world of solar panels is well known but still striking. And the robots have a similar type of efficiency gains.
- [05:43]
B
Yes, yes. So the 90% or so pricing disruption in solar took over a decade, but in humanoid robots it took barely a year actually for Chinese humanoid robot manufacturer Unitree to come up with its G1 model selling at $16,000 versus rivals $150,000 for example.
- [06:03]
A
And you think there's more to come? Namely that Chinese companies are set to pose competition in the worlds of memory chips, the hottest sector in the year, aiming to supply bottlenecks at a time of rampant demand.
- [06:20]
B
Potentially, yes. So at the moment there's a memory oligopoly which includes Samsung, SK, Hynix and Micron, who control about 95% of global supply. And they're basking in an AI driven scramble for what's called high end high bandwidth memory. So hbm, and to make that, they're diverting capacity away from low end commodity dynamic random access memory, which is dram, thus causing a shortage.
- [06:49]
A
As anyone will know, these stocks now together, their aggregate valuation is $4 trillion of aggregate valuation and on $500 billion of profits. So these are businesses that, owing to the, as you say, to this oligopolistic structure, are now super profitable. But you think that the Chinese competition will start to upset that profitability?
- [07:19]
B
Potentially, yes. Because what's important here is that DRAM supply, dynamic random access memory supply, the commoditized part of the value chain, isn't a problem of especially scarce technological know how, but rather of capital and time, more so. So the product is fungible and its physics are well understood. So the oligopoly incumbents, when they're earning operating margins above 60% on an average basis, offer somewhat of an invitation to Chinese competition.
- [07:51]
A
So China is coming to upset this oligopoly, you believe?
- [07:54]
B
Yes, exactly. So the invitation, so to speak, does seem to have been accepted in Hefei, where a state subsidized company called Chongqing Memory Technologies, which, which is abbreviated Cxmt, is scaling capacity from about 100,000 towards 300,000 wafers per month, with its revenue up already over 700%. And it's expected to garner about 15% of global DRAM within the year. So as we've seen elsewhere in those other examples, it can flood the commodity tier, push down pricing and erase billions of market value. And many global customers are actually already adopting Chinese dram.
- [08:38]
A
So it's interesting, Lauren, because if you remember in the capital returns book which I edited, we had a global investment review, I think, from 2013 that pointed out the problems of the digital semiconductor cycle, namely that when returns were high or abnormally high, capital was sucked in. And that capital cycle in semiconductors or memory, let's say, seems to have been perhaps interrupted somewhat in the first years of the current century when competition eroded in Japan, which is why those memory makers have been a nice position recently. But here the capital cycle is re establishing itself and pretty fast. And you have to bear in mind also that those memory makers are Themselves, the leading companies are also expanding capacity. So there are some areas, some real pinch points in the world of semiconductors, but that strongest competitive moat actually appears to belong to asml, which provides the technology that's vital for producing, as far as I understand, most technologically advanced chips. Can you explain that position?
- [09:52]
B
Yes. The pinch point, to use your word, between commodity and premium memory, is a technology as advanced as its name suggests, actually called Extreme Ultraviolet Lithography, abbreviated euv. And the Dutch company asml, which is a marathon holding as you point out, monopolizes the business of producing the ultra high tech tools and equipment which make EUV possible.
- [10:16]
A
Why is this technology so vital?
- [10:18]
B
It's vital because it's the only practical way to draw the tiniest features on today's fastest chips. And it's hard to copy because doing it requires a highly complex machine where every part, and there are many parts to it, much beyond my understanding, are at or beyond current engineering limits. So ASML's EUV capabilities have taken decades of physics and supplier learning and service knowledge and customer integration and feedback to build, which is extremely difficult. And I would say it's fair to say that ASML as it stands wields true technical scarcity.
- [10:57]
A
But you think even here, China, with its abundance of capital and patience and rapid technological advance, could make inroads?
- [11:08]
B
Yes, and this is actually the crux of my, of my essay, because price erosion of commodity moats is one thing, but the more unexpected, challenging of more sophisticated moats, whether or not the challengers themselves are economic, is another thing. And I found China's EUV ambitions quite striking. Recent estimates place China's EUV handicap at as few as five years, so less than half what was previously assumed under even the most optimistic of assumptions. And from this sort of starting point, its progress might even quicken because China's allocating US$300 billion to an AI build out plan and it'll give a piece of that to euv. At the moment, China only has a prototype, but it doesn't even need to close the gap with ASML entirely to do real harm. And it may in fact narrow it faster than expected.
- [12:05]
A
And you say that CXMT is going down a different route to produce more sophisticated chips.
- [12:12]
B
So CXMT actually expects to begin mass production of HBM so the the higher end chips by the end of this year. And it's hoping to do that by substituting less sophisticated deep ultraviolet lithography or duv, for euv, which is a costly workaround which has attracted some skepticism from
- [12:34]
A
industry experts, Laura, beyond the realm of semiconductors, you point to Chinese entrants having graduated from imitators to global standard setters. I mean, you don't mention actually in your piece who are way. But that almost is a poster child for that. But you talk about some other examples.
- [12:52]
B
That's right, yes, I would say imitators are increasingly innovators in pharma, for example, and 1/4 of the world's newly licensed assets now originate from China, which is quite impressive given it used to be just a low cost manufacturing hub. AstraZeneca, for example, agreed to license almost $19 billion from China's CSPC in January. Elsewhere in electric vehicles, BYD outsells to Tesla, which again you might see on the streets around you and our listeners might see on the streets around them. It pretty much dominates the low end of the cost curve for EVs and in batteries. A company called CATL supplies 40% of the world market being based in China and dictates the storage economics as well, which is an emerging technology to watch. And then beyond that, DJI and Unitree, the robotics company I mentioned earlier, lead on drones and humanoid robots by volume.
- [13:50]
A
And this is happening at a time when American tech companies are themselves extremely highly valued, especially compared to the Chinese equivalents. And SpaceX, after its recent IPO being case in point. But you don't think this evaluation discrepancy will last forever?
- [14:09]
B
Yes, and it's quite a valuation discrepancy. So SpaceX listed at 100 times loss making sales, surpassing Sun Microsystems and Nvidia's 10 times and over 26 times peak ratios in 2000 and 2005. So on this basis, SpaceX looks pretty much priced to have an impenetrable moat, despite capital cycles theory telling us that where there's a larger potential profit pool, competition generally seems to be lining up to take a piece of it. So I suppose time will tell whether AI and the conquest of space are immune to this phenomenon.
- [14:44]
A
So Laura, how does this thinking on Chinese technological advances competitiveness influence Marathon's portfolios?
- [14:51]
B
Well, on one hand we own a couple of globally competitive Chinese challengers which have proven that they can disrupt global profit pools. And on the other, we seek to avoid the companies that are vulnerable themselves to low cost capacity.
- [15:04]
A
So can you give me an example of the former?
- [15:07]
B
A company called Sani Heavy Industry and it has been the world's leading manufacturer of excavators and concrete machinery by volume since 2020. It competes with Caterpillar and Komatsu, which listeners might know of. And having survived China's brutal post property construction equipment glut, which aligns well with your opening point on the property boom and bust in China, it now earns more than 60% of its sales abroad and it actually does so at higher returns on invested capital. So something impressive it's done is turn the battle scars of what was a vicious home market capital cycle into a durable global cost advantage. And most recently it showed 28% gross profit margin. So it doesn't offer much rent for a new entrance to come and compete away from it. Or to use the example that we touched on earlier, it's quite an unappealing invitation.
- [15:59]
A
But you think the scope for Sany to continue to what you call move up the value chain.
- [16:05]
B
Yes, there is actually scope for Sany to continue to move up the value chains as it gains share abroad. So it's very profitable service business is only 10% of its sales versus its competitors having more than 30. And there's also scope for it to consolidate the market it continues to lead at home, given it has a superior receivable position.
- [16:25]
A
And then you point to electrification being another area of potential interest.
- [16:29]
B
Yes, I actually think that's potentially the most intriguing bit. So SANY is the first mover in electric construction equipment with a leading market share in electric mixers, electric dump trucks, electric excavators, cranes, and earning a greater share of its sales from these products than either of its main peers as well. Further validating this, in addition to being cheap, it's proven it's trusted because it got a major agreement for 100 electric excavators from a marquee global customer which many listeners may know called Holsim.
- [17:02]
A
Which is what, a cement company?
- [17:04]
B
Yes, it's also partnered with the battery behemoth Catl, which we mentioned. So it has especially efficient battery systems that are globally competitive. And at the moment when we last spoke to management, they view the current installed batteries capacity is able to support twice their current sales base. And Sany, in contrast to SpaceX, trades at less than two times sales.
- [17:26]
A
You believe that many moats priced as unassailable, they are merely yet to be assailed by Chinese capital at scale.
- [17:35]
B
That's right. So true inimitability, so to speak, is rare and it rarely endures. Deepseek as the example we started out with, dramatized our long held belief that supernormal returns tend to invite competition. And the other examples also also support that.
- [17:53]
A
What about the moats you think can endure?
- [17:56]
B
Well, that's what we spend a lot of our time trying to discern. But perhaps one of the best ways to be immune to deep pocketed competition would of course be to do something that capital can't replicate. A couple of examples of this might be scale based networks or beneficiaries of overall capacity growth. So metaphorical toll roads or companies with moats predicated on geographically unique physically constrained capacity. So those companies operating in the physical and material world, and you're thinking here
- [18:27]
A
of say for instance like the copper mining where there's a shortage of copper ore and mines to extract it.
- [18:35]
B
Yes. Also those companies that have networks which are hard to replicate, those type of businesses can hopefully endure against imitation, risk by AI or China or both.
- [18:45]
A
And you cite two examples of Chinese holdings that benefit from network effects, Shenzhou International and Tencent, which is owned via the South African company Naspers. Start with Shenzhou. Tell us what you think there.
- [19:01]
B
So Shenzhou operate what I described as a toll road like business model which enables them to harvest the fruits of disruption without actually needing to win said disruption. As the vertically integrated knitwear manufacturer for leading sportswear brands like Nike and Adidas, it enjoys a scale advantage and it actually earns about mid-20s gross margins and it does so regardless of fashion fluctuations. But what's interesting about these mid-20s levels is that they don't attract competition, and that's quite the opposite of what we're seeing with memory at the moment. It's likewise immune to supply additions further downstream. So if Anta, which is a Chinese sportswear brand, were to usurp, for example Nike, Shenzhou is pretty much indifferent. And Tencent, Tencent's WeChat ecosystem is similarly to Shenzhou, brand and sector agnostic, and it acts as somewhat of a distribution layer that gets to tax whoever wins across gaming, payments, advertising, AI, what have you. And on top of this, it has stakes in the broader Chinese tech complex, which gives it optionality on companies like Deepseek that are disrupting other incumbents and even within China. In this way, Tencent's able to be asset light as an incumbent and not need to outspend any rivals.
- [20:24]
A
And Laura, give me some examples of the second type of company. You alluded to geographically unique businesses.
- [20:31]
B
An example in the emerging markets portfolio might be Copa Airlines, which is a Panamanian airline uniquely positioned on the isthmus of Panama, which is something that just cannot be replicated with capital as well. You've spoken with Alex on previous podcasts about our holdings in copper and platinum group metals.
- [20:50]
A
The final takeaway how to think about these competitive moats and Chinese competition.
- [20:57]
B
I think the discipline is to distinguish the more underwhelming barriers of capital and time, which Chinese companies have preyed on, from the genuinely scarce ones. We try our best to select the likely survivors, especially those that have already endured cycles over incumbents whose premiums rest on the sometimes optimistic assumption that no one will build next door.
- [21:20]
A
Great. Thank you very much. Laura.
- [21:22]
B
Thank you.
- [21:23]
A
Thank you for your time today. I hope you will listen to the next edition of the Capital Cycle. This communication is provided for information purposes only. Please refer to Marathon's website and the Global Investment Reviews for further information, including important disclosures.