1.31.25
Samsung is the market leader in both DRAM and NAND and their memory division is just one part of a massive conglomerate. I look through their earnings presentation and conference call transcripts for insight into the state of the memory market. This is not an analysis of their earnings, prospects, etc.
Earnings Presentation
- · Memory revenue in the fourth quarter was a new record, an achievement reached because of higher DRAM ASPs from HBM and high density DDR5 sales.
- · Memory demand from the mobile and PC sectors remains soft.
- · The company is accelerating their technology migration pace in the first calendar quarter of 2025. The stated reason is to address high-density/high-spec demand. While this is one way to put it, a more complete answer is to reduce cost per bit more quickly and shift their wafer mix away from legacy products where the Chinese are gaining market share. They said this directly in their outlook for the full year 2025.
- · In last quarter’s earnings release, the company said the supply demand balance for mobile memory was being affected by increasing bit supply from China. This is an important enough point for me to restate it this quarter, even though the company did not comment on it.
- · They also said last quarter that memory business fundamentals were improving. No such statement was made this quarter.
Earnings Call
These are the highlight quotes regarding DRAM and NAND:
- · CapEx in the fourth quarter of 2024 was KRW 17.8 trillion of which 16 trillion was DS investment. Memory is 80% of total DS revenue. They increased CapEx into memory both quarter-over-quarter and year-over-year.
- · Blended DRAM ASP per bit increased sequentially.
- · Server SSD shipments were limited compared to expectations. Demand from mobile and PC segments was weak.
- · In the first quarter of 2025, they believe mobile and PC customers will continue to lower inventory. This means they are forecasting continued demand weakness in these segments. Recall last quarter management thought memory business fundamentals were improving.
- · They are accelerating their transition to the 1-beta node in DRAM and to the V8 node in NAND. This is an attempt to distance themselves more from Chinese indigenous insurgent supply.
- · They expect DRAM demand to start to recover from the second quarter. I believe this is optimism and I don’t think it will happen. In NAND, they too are baffled by the market weakness in the face of record low capital investment. Between that and production cuts, they believe the NAND market will see a supply-demand impact in the second half of 2025. This is greater wishful thinking than their DRAM outlook.
- · In the fourth quarter, inventory adjustments by mobile and PC clients were more significant than previously expected, which dampened already weak demand. Demand from the server segment was “relatively solid.”
- · In the fourth quarter, DRAM ASPs increased around 20%. ASPs were up mid-single-digits % in Q3, high teens % in Q2, up 20% in Q1 and up very-low-teens % in Q4-23). The strength of the DRAM price increase this quarter is a real surprise to me. HBM sales were up 1.9x. They didn’t say explicitly this was quarter-over-quarter, but I believe it is.
- · DRAM bit growth in Q4 was down low-teens percent. DRAM bit growth was up mid-single-digits % in Q2, it was down mid-teens % in Q1 and was up mid-30% in Q4-23.
- · Over the last two years (eight full quarters, beginning with Q1-23), this is a total increase in DRAM ASPs of about 75%. Samsung’s DRAM pricing has been rising for the last six quarters and have increased by about 115% off of the bottom.
- · Over the last two years (eight full quarters, beginning with Q1-23), this is a total increase in DRAM bit output of about 25%. Over the six quarters that Samsung’s DRAM prices have been rising, their bit output has grown by 20%.
- · In the fourth quarter, NAND ASPs declined mid-single digits %. NAND ASPs were up low-single digits % in Q3, and up low 20% range in Q2 and in Q1. NAND bit growth was down low-single-digits % sequentially in Q4.
- · Over the last two years (eight full quarters, beginning with Q1-23), this is a total increase in NAND ASPs of about 50%. Samsung’s NAND ASPs rose for five consecutive quarters before dropping in Q4.
- · Over the last two years (eight full quarters, beginning with Q1-23), this is a total increase in NAND bit output of about 35%. Samsung’s bit output has been about flat for the last year.
- · Their HBM sales in the fourth quarter were slightly below their initial forecast. The reasons given were geopolitical issues and the pending release of their HBM3E product. Several of their major customers are waiting for their HBM3E product. The transition from 8-stack to 12-stack products in the second quarter of 2025 will happen faster than previously forecasted. Their HBM4 product will be based on the 1c DRAM process.
- · In the first quarter of 2025, they expect a high-single digit sequential decline in DRAM bit growth. They believe conventional (non-AI) memory product ASPs will decline in Q1. DRAM ASPs will see a “slight decline” quarter-over-quarter.
- · With continuing inventory adjustments in NAND, they expect bit shipments to decline by a low-teens percentage quarter-over-quarter. They didn’t comment on their forecast for what NAND ASPs will do in Q1.
- · Citing “external market research organizations,” they believe prices will decline in the first quarter for server DDR5 DRAM and SSDs, as well as for mobile and PC. They think memory demand will “start showing signs of recovery” in the second quarter.
- · The company is prepared to accept some bit share loss in the short term as they transition process nodes to more advanced technologies. Because of this, their production bit growth will likely be significantly limited for both DRAM and NAND in 2025.
- · When asked about the effects of Chinese suppliers on memory markets, the company said they are speeding their ramp-up of non-legacy products, to move away from DDR4 and LPDDR4 products faster, because that is where the Chinese are competing today. DDR4 and LPDDR4 were low-30% of sales in 2024. They are planning to scale this down to a single-digit % share of their business in 2025.
- · In the earnings call for Q3 of 2024, the executives said that the memory market had seen “completion” of inventory adjustments. Three months ago, the company characterized memory inventory levels as “normal.” Now they say those adjustments are continuing through at least the first quarter of 2025. Nobody knows anything.
- · Also in the third quarter conference call, Samsung said they thought the company would ship at least 10% more DRAM bits in Q4. Their actual DRAM bit shipments were DOWN 11%. The executives wildly misjudged the market, even in a quarter they were in the midst of. Their forecast for NAND bit shipments in Q4 was only off by a couple percent, which is within the noise.
Summary
A quarter ago Samsung’s management confidently stated that inventory levels at customers were approaching normal levels and that memory market health would improve in the fourth quarter as a result. This prediction was off by a wide margin. Instead of DRAM bit shipments rising 10% in the quarter, they declined 11%. I lead with this comment because it casts a shadow over everything they say now about what will happen next. While bit shipments were down, the DRAM market showed surprising strength for Samsung with ASPs up 20% in the quarter. That marks the sixth consecutive quarter of rising DRAM prices. Upturns typically last eight quarters. In contrast, the upturn in NAND appears to be over. After rising for the last five quarters, NAND ASPs for Samsung declined mid-single digits in Q4. The company’s response to the weaker memory market conditions they are seeing is to accelerate their technology node transitions in both DRAM and NAND. This is their central defense mechanism against the Chinese companies that are encroaching on both memory markets and disrupting the supply-demand balance. Samsung says they expect to experience a decline in bit output because of the faster node transitions and they are okay with a temporary loss in market share. As for the markets, they believe customer inventories will be depleted soon such that pricing will start to recover in Q2 for DRAM. They forecast weaker bit shipments in Q1 in both DRAM and NAND and for a “slight decline” in DRAM ASPs quarter-over-quarter. Management believes the DRAM market will start to recover in the second quarter of 2025 but NAND will be at least the second half before any recovery is seen. I am short this forecast. I believe indigenous Chinese supply has fundamentally upended both memory markets and has cut this upturn short, especially for NAND. When the upcycle is ending, management teams always think it is just some extra inventory, an “air pocket,” etc. The wild card in this cycle is AI memory demand. While the non-AI markets are oversupplied and will continue to be so, it is possible that AI demand will be so strong that memory companies can maintain profitability through 2025. Will this happen? I don’t know any more than the management teams at these companies do about that. I’m just more honest about history and what it tells us about today.
– S. Hughes (no MU position)