The N/T ratio made yet another new high as the tech-heavy Nikkei 225 outperformed TOPIX. The yen continued to weaken as the rate differential widened. And the carry trade continued as the BOJ appeared to be behind the curve in fighting inflation while the yen intervention threats by the MOF continued to fall on deaf ears. And market volatility remained high. Yet the market felt noticeably different from the previous few months.
 
First, despite the SOX continuing to drive higher, Nasdaq was actually down (as was the S&P 500). The former consists of semiconductors and semi equipment companies while the latter is driven more by the hyperscalers. In June, the market became even more concerned about monetization of AI. The market shifted from “everything AI is awesome” to “everything where AI scarcity exists is awesome”. But even in that context, there was some bifurcation. Within SOX, for example, most of the winners were equipment companies while names like Broadcom and Qualcomm were down (despite the stellar AI revenue growth announced at Broadcom). Even NVIDIA was down, albeit slightly. As the hyperscalers continue their insatiable capex, the market rewarded the picks-and-shovel makers to feed that capex, which is why the Nikkei 225 outperformed as did Taiwan and Korea, albeit less convincingly than previous months. Within Japan, there was similar bifurcation with names like Softbank down hard while semi equipment names like TEL and SCREEN were strong (the MSCI Japan/Technology Hardware & Equipment index was up +4.6% while the MSCI Japan/Software & Services index was down, coincidentally, -4.6%). 
 
Secondly, while the US-Iran ceasefire should have calmed markets, the focus moved to sticky inflation and its effect on growth despite a sharp decline in oil prices . As such, despite the weaker yen, traditional yen beneficiaries like autos and (non-tech) cyclicals were down. The decline in commodities and gold pushed the trading houses significantly lower as well. And yields remained high. According to Google Trends, the usage of the word “stagflation” doubled vs Feb~May which itself was a double vs the months before the Iran conflict (which, again, was a double vs most of 2025).
 
As such, despite the firm Japanese market in June, it felt distinctively weaker which might explain the rising volatility vs previous months.
 
Within that backdrop, we mildly underperformed for the month in our flagship, concentrated portfolio though we continued to outperform in the diversified portfolio, the only reason being that we had taken profits from most of our tech hardware names in our concentrated portfolio while we left some placeholders in our diversified book. But they both benefited from the barbell approach that we have been employing with the defensives and less-macro sensitive positions strongly outperforming. Also, despite the factor headwind, one of our software names announced strong guidance for the coming year with earnings to grow 2.8x despite heavy spending, causing the stock to rise +33% in June (but still trading at a reasonable 22x -yr forward earnings even after the move).
 
I am becoming more worried about slowing growth. We maintain our secular themes in our portfolios based on labor scarcity, automation, productivity gains, and industry consolidation as Japan permanently escapes deflation and we enter an era of rising wages, interest rates, and ultimately real growth. These themes will, in my view, continue regardless of slower growth. But one additional theme has developed, that of economic activity consolidation. I’ll drill down more in upcoming monthlies as we further dive into our themes and why I believe them to be incredibly powerful.
 
Masaki Gotoh
 
*** We will be co-hosting the first inaugural SOHN Tokyo Investment Conference on Oct. 16. We hope you can all attend and help support a great cause with the Karen Leung Foundation *** ​​​

​

===

 

​​“For centuries alchemists tried to make gold from base metals. Today, we make microchips from silicon, which is common sand, but far better than gold.’” – Max Zorin, played by Christopher Walken, in A View to a Kill.
 
I remember one of the first solo meetings I had when I started my career on the buy-side was with Elpida Memory which had just listed. It must have been destiny as my first job out of college in 1992 was at Hitachi in one of their ASIC labs called the Device Development Center which also housed engineers from Mitsubishi Electric ("Melco" for short). Most of the non-memory engineers in our lab and other Hitachi labs (and those from NEC and Melco) would eventually be folded into what is now Renesas, whereas the memory guys would eventually become Elpida.
 
I met the charismatic Sakamoto-san for the first (and only) time at the IPO. I remember quite distinctly his pitch about “premium DRAM” and I couldn’t help but raise my eyebrow. I didn’t know much about accounting or fundamental investing at the time. But I had common sense. And common sense told me that there was no such thing as premium DRAM. If any chip was a commodity, DRAM was it. In the early 80s, the US dominated the market. By the late 80s, Japan would own over a majority of the global market. In the 90s, Korea starts to take over the market along with many Taiwanese players. By the 2000s, the US and Europe would eventually leave the market (excluding Micron). Japan tried to hang on via Elpida but eventually went bust (and acquired by Micron).
 
But memory isn’t very hard. One just needed scale. You needed a huge fab, cutting edge lithography (in order to reduce costs through die shrink), and massive R&D budgets. But the underlying technology is very simple. So, the industry would go through ruthless cycles. Demand would rise, prices would rise, someone would invest in more capacity, others would follow, the industry would overinvest, prices would fall, weaker players would be bankrupted (and/or acquired), supply would shrink, and repeat. Like oil where countries had varying costs to extract the same commodity, DRAM manufacturers had differing “cash costs” (essentially cost excluding depreciation on the fabs and equipment) that would define who would survive each downturn.
 
Samsung, with their abundant cash flow from other businesses and being not just a producer of memory but a major consumer of it, could outspend everyone. The Taiwanese were all too small to achieve the scale that Samsung had and would periodically drop out. If Samsung was Saudia Arabia, the Taiwanese were like Venezuela or Canada’s oil sands. The Japanese (in this analogy might be Brazil), oddly enough, prioritized profitability unlike Samsung who accepted lower profits and continued to invest to achieve scale and technology in advanced nodes regardless of the cycle. Japan ultimately gathered what memory players remained to create scale via Elpida, but, with its inherently higher cost structure, they couldn't compete. Instead, they chose to make a higher priced commodity which worked for a while, thanks to Apple. LCD panels also consolidated for similar reasons and Japan’s answer via Sharp was a similar “premium panel” using the Kameyama brand which equally failed against the scale/cost game (looking back, I never shorted Elpida, which I should have, but did short Sharp).
 
While Hynix and Micron are marveled now, they were equally in difficult positions. Hynix had to be saved. But what kept Hynix and Micron alive (other than the fact that they had fairly low costs after Samsung) was probably because they were pure memory makers with nothing else to fall back on, unlike the Japanese. So Micron continued to invest via acquisitions of other failing DRAM enterprises. Hynix went through several near bankruptcies and Micron actually tried to acquire them once. They refused, underwent a painful restructuring, and came out better for it. It was further invigorated when SK Group invested and, by then, there was only Samsung and Micron+Elpida (and was able to continue investing in a niche technology called High Bandwidth Memory). If Micron had succeeded in acquiring Hynix, who knows? Maybe it would’ve been Samsung, Micron+Hynix, and Elpida … though I doubt it and Elpida would’ve collapsed regardless.
 
And then there were three.
 
And this is why I worry about stagflation. I don’t care about HBM that much. Admittedly, I don’t know the technology in depth, though I’m generally a believer that all technologies can eventually be replicated by latter entrants if one spends enough and/or willing to forgo profits for a period of time. What worries me is that, because HBM is so profitable and in such high demand, the three have reduced supply of conventional DRAM. While not a cartel, it is similar to what happened with the 1973 Arab oil embargo leading to the first oil shock. And, like the 70s, higher inflation is leading to slowing global growth from supply side constraints.
 
A few months ago, I had to buy a new PC for a new member of our team who wanted 64 GBs of memory. I just bought a couple last year so didn’t think much of it. But when I checked, the prices were 2x. I had to buy a non-branded PC to get a “reasonable” price and, even then, there was only a single online store (which I do not planning on sharing with anyone in case I need to buy another one; I might buy one or two to keep in stock). And like oil in the 70s, memory is used in everything, not just PCs.
 
The tightness is likely to continue. The latest Micron earnings call seemed to support that view where they stated that supply will remain constrained throughout 2027 despite aggressive spending. And the tightness isn’t just HBM but DRAM and NAND too (specifically, enterprise SSDs). The fact that customers are signing multi-year agreements to secure supply would suggest that Micron isn’t just talking up their own book. I’d be watching carefully for what Samsung and SK Hynix say this quarter.
 
Assuming this is true, what happens next? As we learned from the second oil shock, it is not inflation but the perception of higher inflation that caused stagflation. We need something to cause prices to fall (or at least perceive to fall).
 
Unlike oil which can be conserved through higher fuel efficiency, we will always need more memory. If we could find a replacement such as what nuclear power did for oil, perhaps that will alleviate some pressure. But current such technologies will take decades to produce in scale and even longer to replace DRAM if ever. The other method would be to stave off demand. If markets truly start to question the monetization of AI and/or hyperscalers become FCF negative (a point we are fast approaching), perhaps AI investment would slow, leading to HBM production shifting back to conventional DRAM. Or, perhaps, we simply need a healthy rate hiking cycle as inflation stays high, maybe even a Volcker-style shock treatment. It is interesting timing that the new Fed chair is much more inflation conscious.
 
The other possibility is new supply that could disrupt the imbalance. The recent news that Apple is requesting a waiver to source memory from a politically sensitive supplier, ChangXin Memory Technologies or CXMT, feels very significant. A viable fourth player may suddenly arise, one that already produces about 10% of total DRAM and could expand fast as they prepare their IPO. They had acquired Qimonda’s patents and technology to become a viable threat very quickly. And, as I had mentioned earlier, conventional DRAM isn’t that complex. CXMT might become the Permian Basin of DRAM.
 
However, this is a binary question and is dependent on the White House. If the waiver is given, my stagflation fears might be put to rest. If not, I may start hoarding DDR5 DRAM.
 
[Separately, I believe Bain made a good call in selling the remainder of their holding of Kioxia. NAND is no HBM. It’s very easy to produce. And there are more players including another Chinese manufacturer, YMTC, that is already producing in larger scale in terms of market share than CXMT. It’s also easier to increase bit output without increasing wafer count. I think the market is excessively bullish on Kioxia. But it may simply be because I’m annoyed that it’s the top company in our Topix Mid400 benchmark … talk about bad timing, we just switched benchmarks this year and it’s outperforming our old benchmark by +3% in the first 6 months alone.]
​