4.30.25
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 sales declined from KRW 23.0T in FQ4-24 to KRW 19.1T in FQ1-25. That is a 17% sequential decline.
- · Demand for NAND improved in the quarter. The company said this period was a “perceived” bottoming of the NAND market. This is a soft, hopeful prediction that NAND ASPs will improve from here.
- · Earnings were reduced because of “ASP erosion.” Which type of memory was not stated, which I read to mean both NAND and DRAM saw prices decline.
- · HBM sales dropped because of export controls.
- · HBM3E 12H is ramping to meet initial demand.
- · Migration to the 8th generation vertical NAND is being accelerated, to “enhance cost competitiveness.” This is a continuation of the statement from last quarter that the company will move more of their NAND wafer starts to the leading edge to lower cost faster. This is their best defense against the encroachment of YMTC into the NAND market. The risk is faster node migration increases bit supply and thus reduces prices. But having lower cost that your competitors is the best advantage in a commodity market like memory.
Earnings Call
These are the highlight quotes regarding DRAM and NAND:
- · Memory CapEx investment declined sequentially, “reflecting [Samsung’s] more flexible approach to better align with supply and demand.” Historically, Samsung has used supply pressure to bully their competitors. It is notable they are not saying they will do this to the upstart Chinese companies.
- · They expect AI server memory demand to continue growing. Mobile and PC demand “should benefit” from the on-device sale trend. I assume they mean on-device AI.
- · In the first quarter, demand for AI servers “stayed solid.” The recovery in memory demand from the mobile and PC segments “became more visible than initially expected.” This means demand from those segments came back faster than they thought it would. Last quarter the opposite was true. Management misjudged the timing of the recovery in demand from these segments, but not the prediction that it was coming. The server SSD market remained weak. The company cited delays in some data center projects for this weakness, yet these “delays” didn’t affect AI memory demand.
- · They exceeded their guidance in DRAM bit shipments by expanding shipments focused on servers. However, export controls led to a decrease in quarter-on-quarter HBM sales, as well as market anticipation for upcoming HBM3E products.
- · DRAM bit growth in Q1 was low single digits percent. This was much better than the high single digit declinethey had forecasted at the beginning of the quarter. DRAM demand returned faster than anticipated. Recall I was short the demand returning at all, so I was doubly wrong here. The last four quarters, oldest to newest, have seen DRAM bit shipments from Samsung up mid-single digits %, up 10%, down 11%, and up low single digits.
- · Over the last two-plus years (nine full quarters, beginning with Q1-23), this is a total increase in DRAM bit output of about 27%. Over the six quarters that Samsung’s DRAM prices have been rising, their bit output has grown by 20%.
- · NAND bit shipments decreased by about 10% sequentially. This is better than the company had forecasted for the quarter. They said the reason for this was additional purchases by customers who believe that pricing has bottomed and will start to increase.
- · Samsung’s NAND bit shipments have been down now for five consecutive quarters. The cumulative decline over this period has been approximately 17%.
- · Going back to the beginning of calendar 2023, the company’s NAND bit shipments are up a cumulative ~22%. That is over a total of nine quarters.
- · Memory inventory in the mobile and PC segments has normalized earlier than anticipated. This is an important development.
- · The company has seen preemptive purchasing after the pause in reciprocal tariffs. Demand for AI servers remains “robust.” They are seeing “stimulated demand” in the domestic Chinese market. This may be government stimulus. I wonder if some of this is also from a slowing of output from CXMT and YMTC, leading to a perceived increase in demand to Samsung that is really a shift away from indigenous Chinese supply.
- · They gave the same prediction that I have heard from all the memory makers: that the end of Windows 10 support and on-device AI features will stimulate mobile and PC demand. I am skeptical of this.
- · They are expanding their transition to the 1-beta DRAM node.
- · In the first quarter, AI server demand remained “solid,” and there is a “growing perception that prices are pretty much bottomed out for both PC and Mobile applications.” Some of this may be companies pre-purchasing ahead of possible tariffs. Management did not specify if this bottoming in PC and mobile pricing is DRAM, NAND, or both. I think it is both.
- · Server DRAM bit shipments for the company increased sequentially by a mid-teens percentage.
- · DRAM ASPs declined by a mid-single-digit percent quarter-over-quarter. This drop is the first pricing decline after six consecutive quarters of rising DRAM ASPs for Samsung. Over that time (the last drop was in Q2 of 2023,) the company’s ASPs rose 117% in total. The decline this quarter brings the total increase to ~106%.
- · The SSD market has seen weaker demand recently.
- · NAND ASPs were down “mid-10% level.” I think this means mid-teens percent down. Bits were also down by double digits sequentially. NAND prices for Samsung have experienced sharp swings in the last two years. After bottoming in Q2 of 2023, NAND ASPs rose for five consecutive quarters, peaking at a rise of ~110% off the bottom. Since that apex in Q3-24, NAND prices have dropped 5% in Q4-24 and then another 15% in Q1-25. Usually, when memory prices top and the start down, the bottom isn’t seen for five or six quarters. It will be an historical departure if pricing really is bottoming now.
- · They expect a “gradual contribution” to their top line from HBM3E starting in the second quarter of 2025. HBM “bottomed out” in the first quarter. I think this is because of tariff restrictions. The company expects HBM sales to step up each quarter going forward. They expect revenue from custom HBM4 projects to start contributing to revenue in 2026.
- · They believe customer DRAM inventories will normalize in the second quarter. The “relatively lower ASP” in the mobile and PC segments will start seeing recovery in the second quarter. They are forecasting DRAM bit shipments to rise low 10% range sequentially in the second quarter.
- · For NAND in the second quarter, they expect the price declines to slow down for some products and to be flat for others. They are forecasting NAND bit growth to be up mid-teens-percent sequentially in Q2.
Summary
I was hard on Samsung’s management team last quarter for being wrong about the non-AI memory markets and when inventory would normalize. They thought inventory across the industry of non-AI DRAM and NAND would reach normal levels in late Q4 of 2024 or in Q1. This quarter, they are saying inventory health has been reached, or will be quite soon. They are forecasting higher bit shipments in both DRAM and NAND in Q2. It may be my turn to apologize. I dumped all over their forecasts that PC and mobile DRAM demand – and the dependent pricing – will improve from the second quarter and beyond. I must be losing my mind because I am starting to believe this is true. The brevity of the upturn of the memory types outside of AI uses is what opened my mind to this. The emergence of YMTC and CXMT are the cause of the upturn only being twelve months long instead of the normal two years. But Chinese CapEx has shrunk and his forecasted to be lower in 2025. If Chinese indigenous memory supply growth really does slow, and inventories at customers and manufacturers draws down to healthy levels, we could see ASPs in mobile and PC segments improve. What I am most interested in is Samsung’s HBM growth. They said precious little in this call to quantify that growth. All they said is their HBM3E revenue bottomed in Q1 and will grow every quarter from here forward. They didn’t say by how much. My view is Samsung’s executives see Hynix, their national rival, taking the mantle of the largest DRAM company in the world, as a personal humiliation. Hynix did this by dominating in HBM, a product segment where Samsung is in a distant last place among the Big Three DRAM makers. I am certain Samsung’s DRAM division has been at DEFCON 1 for at least the last eighteen months. They have continued to fail to qualify at the major AI customers. But it is just a matter of time. Samsung has been throwing everything they have at making HBM work since probably the middle of 2023. The failures in that time have just intensified their resolve. By now, they are getting traction. I believe a flood of HBM will come out of Samsung in the second half of calendar 2025 as that spinning wheel catches, propelling their massive manufacturing machine forward with high-volume HBM production.
– S. Hughes (short MU)