Goldman Sachs Global Semiconductor Conference May 2024


This was a fireside chat hosted by Toshia Hari of GS, with Sumit Sadana, the CBO.

  • · PC unit growth and smartphone unit growth in the low single digit percent for calendar 2024.
  • · They have continued to see more normal inventory levels. They expect non-AI data center server demand to be better in the back half of calendar 2024 compared to the front half.
  • · They continue to expect low 30% range for GM in FQ4-24.
  • · He reiterated record revenue expected in FY-25 with much better profitability.
  • · The company expects pricing to continue to improve as calendar 2024 unfolds.
  • · Mid-teens is their through cycle growth expectation for DRAM and low 20% for NAND. These CAGRs depend on starting point, they are averages. Within this, they expect data center to grow faster than this average with PC and smart phones growing more slowly than the average.
  • · For NAND, tech transitions are adequate to support the needed bit growth. They don’t need to add wafers to meet the bit demand growth for NAND.
  • · For DRAM, because of the large trade ratio to convert capacity to DDR5 and HBM, wafer growth will be needed to meet the mid-teens bit growth rate that is expected.
  • · Last year, HBM was <2% of industry bits. This year (I assume he is referring to calendar year) that will be less than 4%. They forecast a 50% CAGR currently for HBM bits. Demand continues to be strong.
  • · Sadana reiterated that FY-24 will have several hundred million of HBM revenue and FY-25 will have “billions” of HBM revenue. Sometime in calendar 2025, they expect to match their low-20% DRAM bit market share in HBM. I read that to mean it will be near the end of the year that low-20% of Micron’s DRAM bits will be HBM, so late FQ1-26 or early FQ2-26.
  • · At least in this cycle, Micron expects HBM gross margin to be accretive to overall DRAM gross margin. This was part of an answer to asking if HBM will just become another DRAM SKU and will be commoditized. Sadana answered the question, but not in a satisfactory way related to HBM being commoditized. I think that is the future for HBM, that it will just be another mostly undifferentiated product that all three DRAM makers produce. However, I don’t think that stage will be reached until late in this up-cycle. This flooding of HBM product volume into the market will be the lead-in to the next downturn. That is my prediction.
  • · Sadana said that HBM volume and pricing is set for 2024 and is mostly locked for calendar 2025. He said that these Long Term Agreements were firmer than they have been in the past, but stopped short of saying they are take-or-pay agreements. I also am suspicious of how locked the pricing is.


I was disappointed that Micron didn’t provide any information on the third fiscal quarter, which ends today. I thought they would update guidance, but Sadana stuck with longer term trends. HBM is sucking the oxygen out of the memory market and Sadana pointed out that the trade ratio of bits needed for DDR5 and HBM is squeezing the supply of DRAM bits for other segments. I think this is going to continue to support higher ASPs as the year goes on. Micron is feverishly (along with the rest of the industry) shifting bits over into HBM, to meet the high demand and take advantage of better pricing. I predict this will continue into at least early calendar 2025. I was disappointed that Sadana stuck with the prediction that FQ4-24 gross margin for the company will be in the low-30% range. I think it will be mid-to-high 30% range, supported by faster-than-expected increases in pricing and more bits shifted to HBM than are in the baseline plan. The GS analyst asked a central question, one that predicts the future for DRAM. I think the AI hardware market will reach a saturation sometime in the first half of 2025. This will lead to a decline in demand for HBM, and then oversupply in DRAM. I believe today the market is underestimating how fast DRAM ASPs and gross margins are going to rise in the second half of calendar 2024. I also think markets are not sufficiently appreciating that all the enthusiasm for HBM and AI is inflating a bubble. I think there will be a pause in demand for AI silicon and that will be what leads us into the next DRAM downturn.

-S. Hughes (cyclical long MU)