11.16.24
The results and commentary from the DRAM makers regarding the third calendar quarter were great news for memory shareholders. All of the big three DRAM makers said they will hold back inventory to support pricing rather than selling into market weakness. The DRAM market is bifurcated into AI-related products (HBM and DDR5) and everything else (PC, mobile, traditional servers, automotive, industrial, consumer). Demand is high in the AI segment and tepid everywhere else. The companies are moving bits over to AI as fast as they can, which is lowering supply in all other segments. This has been the reason for much of the increase in pricing seen in the last four quarters. Hynix has been in a DRAM upturn for six quarters. Micron, Samsung, and to a lesser degree Nanya, have enjoyed about a year of rising DRAM prices. The difference is because the HBM DRAM market turned six months before the other types. Hynix had a monopoly in HBM and are still far out ahead of everyone else. Micron is in second place and said they will match their HBM bit share to their overall DRAM bit share sometime in calendar 2025. Samsung is the laggard, but they are as big as Micron and Hynix put together, so when the full weight of their capacity comes into the HBM market, it is going to be a tsunami. This will happen sometime in 2025 as well.
DRAM ASPs in the quarter revealed a spectrum of increases, depending on how much HBM the company has in their mix, as well as an x-factor. At the weak end, Nanya saw their DRAM ASPs up mid-single digits %, but bit shipments were down 20%. Nanya is all legacy nodes and has no HBM capacity. They are getting squeezed by the weak non-AI DRAM market on one side and by new legacy supply coming online from indigenous China. Nanya is at risk of not having an upturn and may be in real trouble. DRAM ASPs at both Micron and Hynix were up mid-teens % in the quarter and bits were about flat. The x-factor is how Micron was able to match Hynix’s ASP growth with a lower mix of HBM. My guess is they are holding the line on pricing with customers more strongly than the rest and are willing to grow inventory to do so. Samsung’s DRAM ASPs were weaker, up only mid-single digits %. They increased bit shipments 10% sequentially, so their market share increased in the quarter. Samsung has been willing to take a smaller price increase from customers in order to ship more bits. I hope this is not a trend. On the other hand, Samsung said they believe customer inventory levels for PC and mobile DRAM have normalized and the fourth quarter will be as good or better than the third as far as supply-demand balance.
NAND is also a spectrum of pricing strength, though there is not a clear reason for this as there is with DRAM. Micron’s NAND ASPs were up high single-digits % and bit shipments were also up high single-digits %. NAND at Hynix saw ASPs up mid-teens % and bits down mid-teens %. Samsung’s NAND ASPs were up low single-digits % and bits were down low single-digits %. WD NAND ASPs rose 4% sequentially and bit shipments rose 14%. Note that only WD gives like-for-like pricing changes. The other three are blended. WD’s blended NAND ASPs were down 6% sequentially. Overall, NAND price increases are slowing, which typically means they are approaching their peak. However, there has been no rise in NAND capacity investment, so this isn’t coming from a supply response. Likely, we are seeing price elasticity affecting NAND demand, following several quarters of increasing ASPs.
In summary, the DRAM makers are saying the right things about holding inventory to support rising prices in non-AI products. If AI demand holds, we are in for at least two more quarters of strength in the DRAM market. NAND is mysterious. Will the continued lack of investment in capacity allow prices to rise further, albeit at a slower rate, or is this market fundamentally unhealthy such that even capacity discipline can’t save it?
– Smooth Hughes (cyclical long MU)