4.24.25
Investor Presentation
For the first quarter of 2025, Hynix saw an 11% sequential decline in revenue, to KRW 17.64T. Within this revenue, DRAM was down slightly. Almost all of the drop came from weakness in NAND. Within the NAND sales, the decline was mostly in SSDs, with USBs, cards, and other low-end applications taking up a larger fraction of their total. Mobile held steady as a percent of total NAND sales, so the same percent piece of a smaller pie. Most impressive in this quarter’s results is the gross margin. This expanded 500 bps, to 57%. Gross margin for the last nine quarters, oldest to newest, are (32%), (16%), 1%, 20%, 39%, 46%, 52%, 52%, and 57%. Hynix’s gross margin has now been positive for eight quarters in a row. The increase in gross margin this quarter was in large part, this is because NAND dropped significantly as a percent of total revenue. The company is also continuing to grow their HBM in absolute volume and sales. Operating margin only increases 100 bps, to 42%. The company characterized this quarter as a “market correction.” Because of the higher margins this quarter, Hynix’s net profit still increased 1%, in spite of lower sales.
In the first quarter, DRAM bit shipments were down nearly 10% and ASPs were flat. This is a little better than what they forecasted for the quarter (low-teens decline.) That tells me they were surprised to the upside by the strength of the DRAM market this quarter. With the HBM market still going strong, this overall DRAM performance means that non-AI products saw price declines and even lower bit shipments than the nearly-double-digit drop seen in total DRAM. He previous twelve quarters, newest to oldest, were up mid-single-digits, down slightly, 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 two years, Hynix’s DRAM bit shipments have increased by a total of 60%. I picked two years ago as a starting point because that is when the HBM/AI upturn began in DRAM, for Hynix. That growth in two years is a CAGR of approximately 27%, well above the mid-to-high teens average predicted for the overall DRAM market.
In Q1, DRAM ASPs for Hynix were flat. This follows seven consecutive quarters where Hynix’s DRAM prices rose by a total of 150% off the bottom. Recall that the bottom for the company was the first quarter of 2023. After that period, demand for HBM took off. For Hynix, their ASP changes the last nine quarters (newest to oldest) have been flat, up 10%, up mid-teens %, up mid-teens %, up 20%+, up high teens %, up around 10%, up high single digits and down high teens %. In a “normal” upturn, I would have high confidence that this quarter, with flat pricing, is the top for the DRAM cycle. However, AI demand continues unabated, so Hynix’s HBM business will remain strong for at least one mor quarter. Just as important, the non-AI DRAM market is showing signs of reaching a healthy overall level of inventory. Three months ago, I didn’t believe this would happen. Also, the most recent round of results from WFE makers show that capital investment by CXMT in China has slowed, which will help the supply situation in non-AI DRAM. Put these factors together and I am less confident that the overall DRAM market will go into oversupply in 2025. I still think it will (more than 50% chance) but it is a lower probability than I believed a quarter ago.
NAND bit shipments dropped by almost 20% in the first quarter, in-line with their guidance. This follows a drop of ~5% last quarter, ~15% two quarters ago, and flat bit growth for the six months before that. Using Q3 of 2023 as a starting point – six quarters ago – Hynix’s NAND bit shipments are down by a full third. Starting with the first quarter of 2022, three full years ago, the company’s bit shipments are down by a quarter. Hynix is losing market share to YMTC.
You would think all this NAND supply coming off the market would lead to much better pricing. You would be wrong. In the first quarter of 2025, Hynix’s NAND ASPs dropped by nearly 20%. That follows a drop of 1-2% in the prior quarter. I predicted that fourth quarter would mark the top of the NAND cycle, because pricing usually looks like a sine wave. Once the ASP curve flattens, it almost always goes down. This quarter bears that pattern out, at least so far. The company does not forecast pricing, but they do forecast bits. Following the nearly 20% drop in bit shipments in Q1, they are guiding to a low-20% increase in bit shipments in Q2. If this rise is from a much stronger demand environment, my prediction of better pricing is at risk. However, if it is a reflection of dropping prices to move NAND inventory, then the downturn will continue. Pulling out to a longer view, the upturn in NAND for Hynix lasted five quarters. Pricing turned up for the company around Q3 of 2023 and rose from there for four quarters, into Q4 of 2024. Sometime in Q4, NAND prices for Hynix started to decline again. In that short five quarters of rising prices, Hynix’s ASP per bit rose almost 150%. Now, they are down to a rise of 85% to 90% above the bottom in 2023. The established NAND makers are all pulling back on wafer starts, but also accelerating their rate of node migration. These are contradictory forces. Hynix will grow their bit shipments around 20% in Q2-25. Unless the market has gotten significantly stronger (because of depleted inventory, for example,) this additional supply will continue downward pressure on prices.
Hynix’s management is sticking with the story of “structural change in the memory biz.” Famous last words if there ever were any. Their market outlook is for the PC and mobile markets to see higher demand because of a PC refresh cycle from the end of Win10 support and AI features in PCs and on mobile devices. Recent readers of this column may recall that I believe this is wishful thinking on the part of the memory makers. They forecast servers to continue to be strong because of AI. While they would never say it, the company is leaving the NNAD market for dead in 2025. Their exact quote is “Market recovery in ’25 expected more from supply factors.” The translation of this is, demand is weak and will stay weak. We hope all the producers will pull back on wafer starts and let supply decline to match demand. This will happen, eventually, but the existence of YMTC puts the timing in jeopardy.
The company forecasted a low-teens % rise in DRAM bit shipments in Q2. Their forecast for NAND bit shipments is even higher; more than 20%. Their comments on CapEx can be summarized as, we are not going to ramp new facilities until market conditions support it. They will “prudently operate new facilities.” In HBM, the company is sticking with their forecast that 2025 demand growth will be 2x over 2024. I think this means revenue. The transition to HBM3E 12-high cubes from 8-high cubes will reach crossover in the second calendar quarter of 2025. That is right around now. Hynix shipped the world’s first HBM4 12-high samples in March of 2025. This product will reach mass production later this year. The company provided no update to their forecast from last quarter’s results that 2025 would see bit demand rise mid-to-high teens % year-over-year. Similarly, no update to the prior forecast that NAND bit demand will grow low-teens% year-over-year in 2025.
Analyst Call
Here are highlights from the earnings conference call:
Prepared Remarks:
· The memory market improved faster than expected in the first quarter. Chinese subsidies into consumer markets helped this, as did inventory restocking.
· Sales of PCs and smartphones were higher than expected.
· NAND saw some spot market price increases due to production cuts in the industry.
· Because of tariff uncertainty, their forecast of memory market demand recovery in 2H of 2025 is at risk.
· Server demand is increasing because of continued investment in AI applications, leading to expected limited volatility in this demand.
Analyst Q&A:
· Globally, the tariff effects on customers have been limited so far. They are maintaining their demand levels for now. Some are pulling in short-term supply.
· Hynix’s current revenue share from the US is around 60%, because of AI-related sales. But direct memory exports to the U.S. are not that high.
· They expect a 50% CAGR in HBM demand from 2024 to 2028. This is fun to read, but is just a guess.
Summary
The highlight from these results is that Hynix’s gross margin rose from 52% last quarter to 57% this quarter. That is really high, especially considering Micron’s gross margin is yet to crest 40%. Looking deeper, a big reason for this rise in gross margin is Hynix’s NAND sales weakened so much. With NAND a smaller percentage of total revenue, and HBM financials continuing to improve (great pricing, better yields, higher volume) gross margin saw a large boost. Still, overall DRAM shipments were down 10% and ASPs were flat. The first part – lower shipments – can be attributed to seasonal weakness. The second part – the flattening of the DRAM pricing curve – typically portends the top of the cycle. Yet, with the ongoing strength of HBM demand and reports of healthy inventory levels being reached in PC and mobile DRAM across the inventory, I am more cautious in my prediction of a pending downturn. For DRAM, I continue to believe a downturn will start before the end of calendar 2025. My percent conviction is lower than it was ninety days ago. In NAND, I think that market is in a downturn, and it will stay that way for at least a couple more quarters. Overall, this report didn’t give nearly as much new information as we got last quarter, especially about the HBM market for Hynix. Non-AI DRAM got healthier than expected this quarter but is still not strong. HBM and other leading edge DRAM demand is still great. NAND is deteriorating rapidly. The forecast for higher bit shipments in both DRAM and NAND in Q2 is the one aspect of this report that gives me pause. Is this just coming off seasonal recovery, or have healthy inventory levels really been reached in non-AI DRAM? That is my biggest open question.
– S. Hughes (short MU)