What to do with a problem like Korea

Research evidence excerpt

What to do with a problem like Korea

LLMs are not 120

as efficient - gap in tokens per intelligence is not as large - the writing 20Jan-20 Jan-21 Jan-22 Jan-23 Jan-24 Jan-25 Jan-26

appears on the wall that fully-integrated monopolistic LLMs are less likely. 20222025 EPSEPS 20232026 EPSEPS 20242027 EPSEPS

This implies that investors might be seeing peak LLM values. But, lack of

Source: Datastream; Macquarie Global Strategy

monopolies is good for all derivatives of the AI stack, including applications,

infrastructure and memory. Lower costs and greater flexibility should MSCI Korea Forward PER (x) - down

raise and widen demand. Also, the current backlog (~US$2-3trn) extends to 6-7x

for at least three years, with shortages forcing buyers into long-term

contracts, including price floors. As for supply, while Samsung and Hynix 16

have ambitious plans (>US$1.5trn), new capacity is unlikely before early '30s. 12 + 1 SD

There is a DRAM threat from China (CXMT), but it will take several years

8 Average

to have a material impact. There are other issues (e.g. circular earnings, - 1 SD

investments by Korean corporates flush with cash), neither appear to be a 4

concern right now. 3. Unlike Taiwan, and akin to Japan, Korea offers a wider May-01 May-02 May-03 May-04 May-05 May-06 May-07 May-08 May-09 May-10 May-11 May-12 May-13 May-14 May-15 May-16 May-17 May-18 May-19 May-20 May-21 May-22 May-23 May-24 May-25 May-26

range of exposures to our favorite themes that are only tangentially related MSCI KOREA-12m fw PER

to AI, including "bullets & prisons", "opium of the people", and "augmentation Source: Datastream; Macquarie Global Strategy

of humans". A concentrated market cannot withstand derating of its two