7.24.25
Investor Presentation
In the second quarter of 2025, Hynix’s revenue surged well beyond what the company had expected. Both DRAM and NAND bit shipments airmailed forecasts (see below). Total revenue grew 26% sequentially, to KRW 22.23T. As a percentage of revenue, DRAM shrunk in the quarter. However, this was only because of the massive growth in NAND shipments. DRAM revenue in the second quarter was almost as large as the whole company’s revenue in the first quarter. In absolute sales, DRAM was up 21% in the quarter and NAND rose by 47%. The memory companies have all been forecasting customer inventories to bottom out in the first half of 2025 and that has clearly happened. What they did not foresee was the strength in demand from those customers as they replenish their stocks. I see a couple of possible reasons for this unexpected surge in demand. One – the more likely of the two – is buying ahead of potential tariffs. And two, extra purchases at lower prices to protect against rising ASPs. The latter seems more likely in DRAM than NAND given the decline in NAND prices in the last two quarters. The NAND market is oversupplied.
The surge in bit shipments this quarter led to an historic increase in total sales. Revenue increased so much that the DRAM portion of Hynix’s business this quarter was almost as large as the whole company was last quarter. Following the surge of 500 bps last quarter, gross margin declined in Q2, from 57% to 54%. Gross margin for the last ten quarters, oldest to newest, are (32%), (16%), 1%, 20%, 39%, 46%, 52%, 52%, 57% and 54%. That is eight consecutive quarters of positive gross margin, with Q2 being the first sequential decline in this cycle. I think the decline in gross margin was the result of product mix shifting to more mobile, PC, and consumer products. These mix shifts were not large, but was enough to bring down gross margin. NAND becoming a larger fraction of the total also compressed company margins. Prior to this quarter, gross margin had expanded for eight quarters in a row. Because of the size of the swing in product mix, I don’t think the compression of GM this quarter is an indication of the market turning down. But the nine quarters since the nadir of the downturn is traditionally when memory upturns end. Does the emergence of AI as a new demand segment mean this time is different?
DRAM pricing was up slightly in the quarter, a 1% sequential increase. This follows flat pricing last quarter. Below this number, the company’s mix shifted to a higher fraction of non-AI memory, which was some or all of the reason for the lower gross margin. Normally higher bit shipments are accompanied by rising prices. In this quarter, DRAM bit shipments grew by a massive 25% while pricing held about flat. To me, this indicates strong DRAM demand, but not far above supply. Hynix was able to hold pricing even while shipping many more bits into the market. Looking back further, since the bottom of the market nine quarters ago, Hynix’s DRAM ASPs have risen by 150% and bit shipments have doubled. While the growth in bits has not been smoothly up and to the right, Hynix has well outgrown the overall DRAM market. The rise in DRAM prices has been steady for Hynix. I think this is because their growth in HBM has been so strong and that has washed out the variation seen in non-AI DRAM prices over the last two years. As good as the DRAM market continues to look, blended prices have now been flat for two full quarters. Historically, this would be the top of the DRAM market. With the strength of AI demand, maybe the top will extend out longer. My short position assumes this time isn’t that different.
NAND bit shipments surged by an incredible 70%+ in the quarter. This is miles above the low 20% increase the company forecasted last quarter. Management under-called the strength in demand recovery that came with customer inventory restocking. I wonder how much of this is buying ahead of potential tariffs. Like DRAM, the change in pricing did not match this large rise in bit shipments. NAND pricing in the quarter dropped by approximately 9% sequentially. That is not as large a drop as the 20% seen last quarter, but it is still the third quarter in a row of dropping prices. Hynix’s NAND ASPs are down 30% in the last nine months. Bit shipments over the same period are up 30%, on the strength of the most recent quarter. I don’t know what to make of this. From this quarter, Hynix regained the market share it appeared they had lost to YMTC. Their bit shipments are up more than 70% in the last nine quarters, outgrowing the overall market. But it this the anomaly of one outsized quarter? The trend of declining prices in NAND indicates the market is still oversupplied.
Company management sees a “low likelihood of a sharp demand correction in 2H” because customer inventories have reached healthy levels. A single bullet point covers their market outlook for the PC and mobile spaces. It says AI function growth will drive higher demand for memory in devices. Regular readers know how I feel about this. For those who don’t, I am short this story. Big tech company CapEx seems to know no limit on AI investment. That will continue to drive demand in the data center segment. Hynix also sees non-AI server demand strengthening because of new processors and the replacement cycle. This is the data center version of the PC replacement cycle story that I don’t believe. They continue to leave the NAND market for dead.
Following the massive bit shipment growth in Q2, the company is forecasting DRAM bit shipments to grow by low-to-mid single digits % in Q3. NAND bit growth will be “limited” sequentially. They did say it will be an increase. I think the surge in bit growth in Q2 was a return to normal demand following two to three quarters of weakness as customers worked down excess inventory. The company will increase their capital expenditures in 2025 from the levels previously announced. This is to insure they meet customer demand for HBM in 2026. The CapEx section has four bullet points and the last two concern the comments on increased investment. One way to interpret this is management is cautious about spooking fears of oversupply in the market. They also couched the increased investment as for HBM capacity. That has been how all of the Big Three DRAM makers have characterized their investments this cycle; as either in “leading edge” or in “HBM and other AI products.” My short thesis is that the over investment this cycle will come from leading edge and HBM capacity. On that topic, Hynix restated their plan to double HBM sales year-over-year in 2025 over 2024. They provided the world’s first HBM4 samples to customers in March. The company is bringing low power products for the server segment. Their NAND comments indicate that market is not healthy. The prepared remarks made no reference to their views on bit growth for the whole of 2025 for either DRAM or NAND.
Analyst Call
Here are highlights from the earnings conference call:
Prepared Remarks:
· The combination of customer buying ahead of tariff uncertainty and continued strong demand for AI memory led to higher than expected demand and pricing in the second quarter. The memory market also improved faster than the company expected in the first quarter. Thus, the memory market was above management’s forecasts in the first half of 2025.
· The relatively small rise in DRAM ASPs was indeed from a higher mix of consumer-oriented memory.
· They believe a downward correction in customer demand in the second half of the year is unlikely, given healthy customer inventory levels today.
· Unlike DRAM, the NAND market has yet to see an AI-driven recover in demand.
Analyst Q&A:
· Management declined to provide an update on volume and pricing negotiations for HBM products in 2026, beyond vague comments.
· The large increase in demand in the second quarter was from a shift among customers, from conservatism in inventory levels to desire for higher levels because of tariff uncertainty. EOL of DDR4 and LPDDR4 products also stimulated demand.
· Despite demand coming in ahead of forecasts in 1H 2025, management believes the risk of high variation in supply and demand going forward is unlikely. Management never sees demand slowdowns or oversupply coming.
· The company will use their fabs in China to supply legacy memory products. This strategy is to avoid US regulations blocking advanced technology going into China.
· The company’s inventories have now reached normal levels.
· Higher costs associated with making HBM4 will be passed on to customers through higher pricing with the aim of maintaining margins.
· The current spike in DDR4 products is a temporary demand concentration.
· Management believes the emergence of HBM gives the memory manufacturers higher leverage than they had previously over customers. This time is different. They believe Hynix’s customer focus gives them a competitive advantage over other DRAM manufacturers. I doubt this. The Big Three remaining DRAM makers have survived because they are all highly customer focused.
Summary
Bit shipments in the quarter were much higher than expected for both memory types. How much of this buying was customers getting ahead of coming tariff restrictions? Given the magnitude of it, I think a lot of this was buying ahead. What was unusual was that pricing didn’t rise along with this high bit shipment growth. Customers may have collectively said they will buy much more than expected but only if pricing doesn’t rise as a result. With Hynix and the other memory makers desiring to lower their inventory, this would have been an acceptable proposition. Hynix characterizes inventory levels overall in the industry to be at healthy levels. Overall, the memory market in the first half of 2025 was stronger than management expected at the start of the year, and it strengthened over the period. The higher bit shipments were heavily skewed to non-AI memory products, resulting in lower gross margin in the quarter. GM is still in the mid-50% range, more than 1500 bps above Micron’s, reflecting higher HBM percent in Hynix’s overall sales mix as well as a better HBM cost structure. It has now been nine quarters since the company’s gross margin started to expand. Blended DRAM prices have been flat for the last six months, as have DRAM bit shipments for the company. NAND prices over that period have been in steady decline. During a typical downturn, these characteristics would mark the top. But HBM demand is new and seemingly insatiable. To support this demand, the company announced this quarter that 2025 capital expenditures will be higher than previously announced. They restated their plan to double HBM revenue in 2025 over 2024. This means the higher investment announcement is mostly to support 2026 shipments. The DRAM market looks to me to be in overall balance. Capacity investment has been higher in 2025 than 2024 and Hynix is continuing their rapid ramp of HBM. I think this combination will lead to oversupply in DRAM in the back half of this year. If HBM demand is still well above supply, this prediction will prove to be wrong.
– Smooth Hughes (short MU)