10.24.24
Investor Presentation
Q3 was another record revenue quarter for Hynix, surpassing the quarterly profit peak of the 2018 super-cycle. In the third quarter, DRAM bit shipments decreased slightly from the prior quarter (previous ten quarters, newest to oldest, were up low 20%, down mid-teens %, up low single digits %, up around 20%, up mid-30%, down around 20%, flat, down mid-single digits %, up 10% and down high-single digit %.) Over the last eleven quarters, Hynix’s DRAM bit growth has been almost flat. I estimate it to be up a cumulative 2% over this 2.75 year period. Said another way, Hynix’s DRAM bit supply is the same today as it was at the beginning of calendar 2022, when the last upturn was ending. This is an incredible statement on both the severity of the 2022-23 downturn and on the trade ratio of HBM from conventional DRAM. Bit growth from node migration has been almost totally offset by the reduction in bit output caused by converting wafer starts to DDR5 and HBM from DDR4. The DRAM market has softened more than Hynix anticipated in their last call when they forecasted low single digits DRAM bit growth in Q3.
For NAND, bit growth was down mid-teens % sequentially. This follows the previous ten quarters (most recent to oldest) down low single digits %, flat, down low-single digits percent, up mid-single digits percent, up around 50%, down mid-teens %, down high single digits %, down low teens %, up high single digits % and up high teens %.) Over the last two-and-a-half years, Hynix’s NAND bits are up a total of ~15%. Normal bit growth is at least 10% annually for DRAM and mid-teens percent for NAND. As with DRAM, the NAND market in Q3 weakened more than expected. In their last call, Hynix forecasted NAND bit shipments in Q3 to be down mid-single digits.
In Q3, DRAM ASPs again rose mid-teens % sequentially (same as last quarter). This follows the prior six quarters (newest to oldest) of up mid-teens %, up 20%+, up high teens %, up around 10%, up high single digits and down high teens %. This marks six consecutive quarters of rising DRAM prices. Cumulatively, Hynix’s blended DRAM ASPs have risen 125% off the bottom of the last downturn. Upturns typically last eight to nine quarters. If history is an accurate guide to this cycle, DRAM pricing will peak in the first or second calendar quarters of 2025. But this DRAM cycle is different in that all the growth in demand is from HBM and DDR5, fueled by AI applications. The PC, mobile, and consumer segments have all seen weak markets, which continues today. The rise in pricing has been a combination of premium ASPs for HBM and from bits being drawn out of other segments and into AI applications. All cycles have their differences, but the sluggishness in most DRAM segments has meant that no new capacity has been added so far. All of the investment has been in node transitions and in shifting wafers to make DDR5 and HBM. Because of the high trade ratio of these products, the effect has been to reduce worldwide DRAM bit output. If AI demand is as strong in 2025 as it is forecasted to be, this DRAM upturn may last an unusually long time. In NAND, ASPs rose mid-teens % quarter-over-quarter. This follows (newest to oldest) a rise of mid-to-high teens %, up 30%, up 40%, down slightly, and down 10%. This is now four quarters in a row of rising NAND prices, for a cumulative increase of approximately 140% off the bottom, which was in Q3 of 2023. Again, if the upturn lasts the historical duration of two years, NAND pricing will peak in the third quarter of 2025. For this to come to pass, investment will need to happen soon.
Gross margin in the third quarter increased 600 bps sequentially, to 52%. 52%! Gross margin for the last seven quarters, oldest to newest, are (32%), (16%), 1%, 20%, 39%, 46%, and 52%. That is a total rise of 8400 basis points off the bottom, in six quarters. Granted, I have not made any adjustments here for any inventory write-downs the company has done. Micron’s gross margin bottomed in their fiscal Q4 of 2023, about a quarter later than Hynix’s. That is the period from June to August of 2023. Hynix reached their gross margin nadir in the January to March period of 2023, bottoming at negative 32%. Last quarter, Micron appeared to be 1.5 to 2 quarters behind Hynix in gross margin. However, the most recent quarters for both companies show the curve of gross margin over time flattening for both at the same time. The offset in gross margin between Hynix and Micron is because Hynix has a much higher percent of HBM DRAM bits. The fact that both curves are flattening simultaneously indicates the shape of the rise has mostly been from overall DRAM ASPs increasing. The contribution of rising margins coming from increasing richness of HBM in product mix looks to be similar for both. In other words, the two companies are ramping HBM at similar rates, just Hynix is well ahead of Micron.
In the fourth quarter, the company forecasts their DRAM bit shipments to grow mid-single digits % sequentially, and they expect NAND to grow low-teens % sequentially. For the full calendar year 2024, the company sees overall DRAM market demand bit growth in the mid-to-high teens % year-over-year. They are forecasting high-teens % bit demand growth in calendar 2025. More than one quarter out, the companies don’t know much more than the rest of us do. For NAND, they see bit demand growth in the mid-teens % year-over-year for full year 2024 and they see the same for full year 2025. I think the numbers don’t matter much but the relative sentiment is useful. Hynix sees the NAND market will continue to be weak next year while DRAM demand in 2025 will improve some over 2024. They are probably forecasting some recovery in the PC, mobile and consumer segments. Their specific comments on the PC and mobile segments are low-to-mid single digits % unit growth in both, driven by AI features and the enterprise PC refresh cycle. I am short that AI features are going to boost either PC or mobile unit demand. It is going to be bits being drawn away into AI server applications that keep the other memory segments healthy.
The company is focusing their capital expenditures “on products that the company has a clear competitive edge.” This means HBM. They will also accelerate transition to leading-edge products to further reduce production on legacy nodes where demand continues to be sluggish. In the third quarter, DRAM revenue reached a 30% share of their total. They are forecasting this to be 40% of total DRAM revenue in the fourth quarter. Hynix’s HBM share is now above their overall revenue share in DRAM, and their lead is widening. With Micron planning to match their overall DRAM share in HBM before the end of calendar 2025, Hynix will be able to keep this outsized share only until Samsung gets caught up in their HBM products.
Analyst Call
Here are highlights from the earnings conference call:
Prepared Remarks:
· Mobile and PC memory demand was weak in the third quarter, which was the reason behind the slight reduction in DRAM bit shipments.
· Demand is “solid” for eSSDs but shipment declined due to weakness from PC and mobile customers.
· The record quarterly profit was driven by price increases and increased volumes of higher value products such as HBM and eSSDs. I think the strength being seen in eSSD demand today is coming from AI data center hardware demand.
· A “gradual recovery” in memory demand from PCs is expected in 2025, based on AI-feature PCs and the end of Windows 10 support.
· Hynix is similarly pinning their hopes for higher mobile demand in the next year on AI smart phones motivating customers to upgrade.
· Production of DDR4 and LPDDR4 will be reduced earlier than previously planned as demand in those end markets is weaker than expected.
· NAND bit growth will be solely from tech node migration. The rate of investment will be determined by end market demand. These two comments are saying that demand for memory outside of the AI segment is poor, those markets continue to be weak, and they will not invest in them accordingly.
Analyst Q&A:
· Chinese memory suppliers entering legacy markets are delaying the recovery in the health of those markets.
· HBM pricing and volumes are set by contract for 2025. As with Micron, I am doubtful that these contracts will hold if customers decide to pull back on their capital expenditures.
· Blended DRAM ASPs are expected to continue to increase as their HBM share grows, even if other DRAM segments see price declines.
· Lower than expected DRAM sales in Q3 was from weakness in the PC market leading to lower DDR4 sales.
· Similar to Micron, Hynix will hold and carrying inventory on some products with weak demand rather than sell at lower prices. This is an important development because it means that both Micron and Hynix are restricting DRAM and NAND supply in order to support prices. Combined, the two companies are half the DRAM market and 40% of the NAND market.
· An analyst directly asked if HBM will be oversupplied in 2025. Management responded with the “long term contracts” answer. They believe it is premature to talk about a slowdown in AI memory demand. They believe the difficulty involved in manufacturing HBM products will limit supply growth. They expect HBM demand will be higher than supply in the coming year. This is effectively betting against Samsung figuring out how to make competitive HBM products in high volume, which is a bet I won’t make.
· They understand memory inventory levels at PC and mobile customers have been flat over the last quarter.
· They expect industry-wide inventory levels of legacy memory products to normalize in the first half of 2025. This isn’t much of a prediction; saying the inventory will reach healthy levels sometime in the next nine months. Any prediction more than one quarter out has large error bars around it.
· The majority of HBM sales in the second half of 2025 will be 12-high packages.
· Their NAND business and investment therein is focused on profitability rather than growth. They will maintain a conservative investment posture as long as the overall market remains weak, and inventories are elevated. They are also moving bits into higher-margin products such as enterprise SSDs.
· They are committed to maintaining NAND market health and will allow lower bit growth to do so. They may lose bit market share but expect to keep or grow their NAND revenue share.
· HBM3e products require 2x to 3x the number of wafers for the same number of bits compared to DDR4 and DDR5 products. This is a lower range than what Micron management provided.
· Their long-term CAGR for DRAM bit growth is mid-to-high teens % annually. This is surprisingly high and is buoyed by their belief in HBM strength being a secular phenomenon.
· When asked about Chinese memory entrants, management said they lag behind in technology and quality. The Chinese can compete in areas where price is important, and performance/quality are not.
· Their capital investment for 2024 is slightly higher than what they thought it would be at the start of the year, because HBM demand has been higher than previously forecasted. This means AI demand has been higher than expected by more than the other segments have been weaker than expected. 2025 investment plan is not complete, but they think it will be “slightly higher” than 2024, to support HBM and DDR5 demand.
Summary
The company’s gross margin rose above 50% this quarter. These are incredible results and are driven directly or indirectly by the strength of AI demand. HBM is 30% of Hynix’s total DRAM revenue in this quarter and it will be 40% of the total in the fourth quarter. The only strength they are seeing in NAND is eSSDs, which I believe are also coming from AI data center demand. By far the most important statement in these results is that the company, like Micron, said they will hold inventory rather than sell at lower prices. Thus, Hynix is adopting the same strategy that Micron is. Rather than add new wafer capacity or transition technology nodes faster, the company will instead hold back inventory to support pricing. Between Micron and Hynix, this is 40% of the NAND market and 50% of the DRAM market. If Samsung also adopts this strategy, it will likely have the effect of prolonging the duration of this upturn. Of course, this is all predicated on demand for HBM and AI-related DDR5 continuing. Hynix stated several times that their HBM supply is sold out through 2025 at contractually-agreed volumes and prices, giving them stability and predictability in revenue and profitability. I doubt that if AI demand reverses that customers will honor these contracts with Hynix and Micron. While AI memory demand is firing on all cylinders, other segments (PC, mobile, etc.) are staying weaker for longer than was expected. Bit shipments this quarter were below what Hynix projected three months ago because of extended softness in non-AI demand. The company is pinning some hope on AI-enabled PCs and smartphones spurring unit demand. I doubt this use case will prove to be compelling to customers. Overall, this was a great report for Micron investors, because Hynix is committed to holding inventory to support memory prices. Hynix is making minimal investment in NAND and DRAM capital expenditures are to migrate technology nodes faster and in HBM manufacturing capacity.
– S. Hughes (cyclical long MU)