8.28.24
Sanjay Mehrotra (CEO) and Manish Bhatia (VP of Operations) in attendance.
- · The smartphone and PC end markets “remain stable,” which means they continue to be weak. Customers in these segments have built inventory because they see pricing trending upward, they see coming supply tightness as bits are shifted to HBM, and content per unit continues to grow. The third reason is always there, though memory company executives regularly talk about it as if it is news.
- · Consumer retail, industrial, and automotive segments are also weak now, weaker than mobile and PC. The CEO called out overall China weakness also.
- · In calendar 2023, HBM consumed about 1.5% of the DRAM industry bits and the TAM was $4B. In 2025, the company sees HBM consuming around 6% of the industry bits with a TAM of more than $25B. So Micron sees 4x the DRAM bit share producing more than 6x the total revenue. Assuming even flat pricing per bit of HBM, which is optimistic, that means the company is forecasting a large increase in DRAM bit supply between 2023 and 2025. 2023 was an historically weak year, but that is a lot of bit growth in two years.
- · FQ3-24 (the period from March to May of 2024) was Micron’s first with more than $100M of HBM revenue. For FY-24 in totality, which ended on August 29th, they will have several hundred million of HBM revenue. Putting these two pieces of information together implies that FQ4-24 will have somewhere above $200M in total HBM revenue. In FY-25 they will have multiple billions of dollars in HBM revenue, which probably means more than $2B but less than $3B. If it were going to be $3B or more, the executives would say “several billion dollars of HBM revenue.” This is all reiterating past statements.
- · The CEO also restated that they are sold out of HBM capacity in 2024 and in 2025. This indicates the strength of forecasted demand from customers, but it is far from a guarantee of sales and pricing. If end customer demand were to drop for HBM, Micron’s customers would back away from those purchase commitments. Micron simply doesn’t have sufficient leverage in their relationship with customers to make them take the bits if they don’t want to.
- · Their 1-beta DRAM node has been mature for eighteen months now. That is the node which produces the die used in their HBM3E product.
Summary
Overall, just AI server demand is good. All other segments have varying degrees of weakness. The most useful information from this interview was Micron’s assessment that 1.5% of total DRAM industry bits were HBM in calendar 2023, and they see 6% of industry bits going to HBM in 2025. This puts HBM at around 3% to 3.5% of industry bits in calendar 2024. The DRAM market only started turning in the late summer of 2023, in part from bits being converted to meet the new HBM demand. At a 3:1 trade ratio, the 3% to 3.5% of total DRAM bits going to HBM would have require 9% to 10% (rounding to recognize the imprecision of these numbers.) That is quite a lot of bit supply coming out of other segments. Using their 2025 projection of 6% of industry bits, that is 4.5% more total industry share than 2023, when the downturn bottomed. Thus, the shift of bits from non-HBM to HBM will reduce non-HBM DRAM bit supply by about 13%, all else equal. Of course, there is some increase in overall available DRAM bits from technology migration. This interview was mostly CEO Mehrotra talking too much and providing little useful information, but those data points on HBM share of total bits were worth the time. I have mixed feelings about what is coming for Micron. The comment from their CFO three weeks ago about gross margins not being up but a couple hundred basis points in the September-November quarter is negative because it says non-AI DRAM and NAND demand is still weak. In the positive column, the ramp of HBM happening now is drawing bits out of other segments at a higher rate. Micron seems to have adopted a strategy of holding inventory now to meet demand over the next twelve months rather than adding incremental wafers. This inventory will be a bridge over the shortage of bit production as HBM ramps. If they are right about HBM demand holding, and we get some recovery in non-AI memory demand, Micron will be in for a strong fiscal year 2025, ending in August of 2025. However, the weakness they are indicating is persisting outside of HBM is a negative, and it raises the risk of missing out on revenue before the market tips back into oversupply.
- S. Hughes (cyclical long MU)