10.29.25
Investor Presentation
Total revenue for the company in the third quarter was KRW 24.45T. That is approximately $17.0B. This is a 10% sequential increase. Ten percent is impressive, except it followed a 26% sequential increase last quarter. It’s still impressive. Within this higher revenue, DRAM outgrew NAND, rising from 77% to 78% of the total. NAND declined from 21% to 20%. Therefore, the small fraction of other products Hynix sells remained about flat. Within DRAM, the distribution of revenue by application was stable from Q2 to Q3. Server and mobile grew their share by a little, at the expense of HBM. Estimating from the bar chart provided by the company, HBM (and other graphics sales) are a little more than of the total DRAM revenue for the company. That works out to about $5.57B in HBM revenue this quarter, compared to $2B for Micron in a similar (started one month earlier) quarter. The company’s total DRAM sales in the quarter were $13.26B. Non-HBM DRAM outgrew HBM in revenue this quarter, consistent with what is being seen across the industry. The non-AI DRAM market has moved back into undersupply as bits are drawn over into HBM. NAND revenue was distributed across the three segments similarly to the prior quarter, with a small increase in the share to SSD at the expense of the lowest margin segment: USBs, cards, and other consumer applications.
Gross margin this quarter bounced back up to 57%, which was the level reached in Q1. Q2 saw a decline in gross margin of 300 bps, because there was a surge in sales to the lower margin PC, mobile, and consumer segments. Gross margin for the last eleven quarters, oldest to newest, are (32%), (16%), 1%, 20%, 39%, 46%, 52%, 52%, 57%, 54% and 57%. That is nine consecutive quarters of positive gross margin, with Q2 of this year being the only sequential decline. That drop was caused by a mix shift. The softening of the DRAM market seen late in 2024 and into early calendar 2025 is only visible in Hynix’s results as a one-quarter flattening of gross margin. Their growth in high-margin HBM share balanced out the softness elsewhere. We are now two-and-a-half years into the upturn for Hynix. Micron’s gross margin in the closest equivalent quarter to Hynix’s Q3 was 44.7%. Capital expenditures this quarter were $3.63B US. CapEx in the second calendar quarter of 2025 was $3.10B US. Micron’s CapEx in their most recent quarter was $5.66B. Micron’s CapEx the last four quarters has averaged $3.96B per quarter.
After two flat quarters, Hynix’s blended DRAM prices rose mid-single digits sequentially. This was lower than I thought it would be given the double digit rise in prices at Micron and all the positive news about the memory market. Bit shipments in DRAM were up high-single digits sequentially. Rising prices and rising bit shipments means the overall DRAM market is healthy. Company commentary indicates it is getting stronger. Going back ten quarters to the low point of the last cycle for Hynix, their blended DRAM prices are up more than 160% off the bottom. That is, the company’s blended DRAM ASP is 2.6x the level it was two-and-a-half years ago. That is a 47% CAGR. I would like to go back further, but before the turn up in early 2023, pricing was so bad that Hynix stopped reporting quarter-over-quarter ASP changes for a full year. Over that same period, the company has grown at a 36% CAGR. That is off the low base of a brutal downturn. It is also about 2x the long-term bit growth rate of DRAM demand. Hynix’s DRAM bit supply has more than doubled in 2.5 years (up 120%).
NAND prices at the company rose for the first time in four quarters, increasing by 13%. Bit shipments were down mid-single digits percent sequentially. This is to be expected after they rose by more than 70% last quarter. The rise in pricing with a small decline in shipments is likely more about the shift in mix to more SSDs and less consumer bits than about the strength of the market. My view is the NAND market is recovering. The NAND demand hole was deep. Hynix is shipping about the same volume of bits that they were two years ago. Despite the recent market softness, the blended NAND ASPs for the company are still up 90% off their bottom eight quarters ago. They were up nearly 250% before the pullback in the fall of 2024. What we have learned from this is that last quarter was not an anomaly. It was the start of a trend upward for the NAND market.
Halloween is almost here so it’s time for companies to start talking about their views for 2026 as well as where they see 2025 landing. Hynix believe DRAM bit demand growth will come in up high teens percent over 2024. They believe the market will further strengthen in 2026, with bit demand growth rising more than 20% year-over-year. This is useful not because I believe it is accurate but because it gives insight into the company’s investment plans for the next several quarters. They will add bit output to try and match this view of demand growth, but not overshoot it. Given that they already think the market is undersupplied, they are likely targeting their DRAM bit output growth to be in the mid-20% range. For NAND, they believe 2025 will show mid-teens % bit demand growth. They also believe 2026 will be a stronger year for NAND, forecasting high teens percent growth over 2025.
Their commentary on the server market outlook got technical in a way I haven’t seen before. It could be they are seeking to tell the story of why they believe this upturn will defy the pattern of past memory cycles. They called the change in both DRAM and NAND demand as a “structural shift.” Macro uncertainties continue to weigh on PC and smartphone shipments, leading to expectations for moderate growth. For the fourth quarter, guidance for DRAM sales bit growth is for a low-single digits % increase. For NAND, they are also forecasting a low-single digits percent increase in bit shipments. Just a quarter ago, management seemed to be leaving the NAND market for dead. This quarter is one that contains the first part of the post-holiday slowdown, traditionally a time for bit shipments to decline. But with the reduction in importance of the device market as server (AI and other) growth dominates, this effect should be muted. They know the next quarter better than time beyond that, but quarter-to-quarter has some lumpiness. They said they have completed supply discussions with key HBM customers, but did not comment beyond that. HBM4 shipments will begin in the fourth quarter of this year (2025). In NAND, 321L products will be introduced “promptly.” NAND investment is in tech migration. The company’s net margin was 52%. That is stunning. More than half of their sales make it all the way to the bottom line.
Analyst Call
Here are highlights from the earnings conference call:
Prepared Remarks:
· DRAM bit shipments exceeded guidance. They were forecasted to grow in the low-to-mid single digits range, and they came in high-single digits. This was driven by growing sales of HBM3E 12 high products and DDR5 server memory to support AI demand.
· ASP growth was strong for conventional DRAM products.
· NAND bit shipments in the quarter were down mid-single digits but were coming off a high base in the prior quarter, when they had risen 71%.
· The ASP increase in NAND of low-teens percent was from a recovery in pricing and a higher mix of SSDs.
· Global demand for AI infrastructure is not only fueling HBM but also demand for DRAM for general purpose servers and enterprise SSDs.
· In the coming months, AI data center demand will continue to create robust demand across a wide range of memory products, including conventional DRAM and NAND.
· PC and smartphone markets are expected to show moderate growth.
Analyst Q&A:
· DRAM bit demand growth is expected to rise from mid-teens percent this year (2025) to high teens percent in 2026.
· In the fourth quarter, considering the normalized levels of inventory (management did not specify whose inventory but this implies customer inventory), both DRAM and NAND bit shipments are expected to increase by low-single digits quarter-over-quarter.
· HBM4 shipments will begin in Q4 of this year and will expand in 2026.
· The company is ramping 1-c DRAM and they will accelerate migration in 2026 in response to surging demand.
· The company said they have “secured customer demand across all DRAM and NAND products, including HBM, through next year.” This doesn’t mean take-or-pay. It doesn’t even mean all their supply is sold out, but it is meant to sound like that is the case.
· AI memory demand is significantly exceeding expectations. In response, the company is opening their M15X cleanroom ahead of schedule to add capacity. In both DRAM and NAND, the company will accelerate node transitions. Their 2026 CapEx will be higher than 2025 levels.
· The company does not believe HBM supply will catch up with demand in a short period of time. They believe their HBM supply will be tight into 2027.
· In response to a question about the sustainability of this cycle, the company believes AI demand is more broad-based than past cycles, and hence it is more durable. I think this will turn out to be wrong. When AI investment enters a phase of digesting all the capacity that has been added, the demand will drop across all sources.
· Their overall DRAM shipments will grow at a high 10% level next year. I think that means high teens percent. It is not a coincidence that high teens is their forecast for the growth in bit demand in 2026 as well.
· Management believes that supply is structurally incapable of catching up with demand and that this condition will lead to a long drawn out memory super cycle.
· The company said in its answer to one question that “not only HBM but DRAM and NAND capacity has essentially been sold out.” This is still not saying a take-or-pay contract has been signed, but it is stronger language than what was said in the prepared remarks.
· The company’s CapEx in 2026 will “far outpace the level of this year [2025]”
· DDR5 customer inventories are “extremely low.”
· Management quoted that the HBM market will grow at a 30% CAGR for the next five years. In a business where predictions more than a few months out are more often wrong that right, this is a bold statement.
Summary
It is all systems go in the memory markets for Hynix. HBM demand has been strong for more than two years. Now, the conventional DRAM market has slid into undersupply and pricing is rising there. The company has reached a stunning level of profitability. Their gross margin was 57% this quarter. Because their revenue is so large relative to their operating expenses, their net margin came in at 52%. The company’s blended DRAM ASPs have risen by more than 150% since their upturn began two-and-a-half years ago. This quarter also showed that the NAND market improvement seen in Q2 was not an anomaly. Rather, it was the beginning of a trend. Both DRAM and NAND markets are improving. Management believes the good times will not end any time soon. They have raised their near term and 2026 capital expenditures in each of the last two quarters. Still, they believe AI memory demand is so strong that supply can’t catch up with it until at least 2027. They think both the DRAM and NAND markets have experienced a “structural shift.” Hynix believes this time is different. Right now, demand for memory for AI (and to some extent conventional) data centers appears to have no limit. If my short position ends up a loser, it will be that bottomless demand that does me in. However, if this time really isn’t different, then DRAM supply will overtake demand soon. I have been predicting that would happen by the end of 2025. It appears I’m going to be wrong about that. Hopefully, I’m just late and the oversupply hits in the first half of 2026. It is AI demand that will decide the match. If it does falter, that 3:1 trade ratio between conventional DRAM and HBM will reverse and spray a torrent of bits into that market.
– S. Hughes (short MU)