Memory was 50% of revenue in Q1 of 2022, a doubling in percentage terms sequentially. Because revenue was lower (see tool release change) this large increase in percent of memory sales is smaller in total spending, a rise of €935M to €1,143M from quarter to quarter. In 2022, logic revenue from system sales to be up 20% year-on-year, and memory is forecasted to be up 25%. This would put memory sales at €5.08B in 2022, almost €500M higher than in 2018, the previous record year. Order flow from memory customers is primarily for capacity additions, as utilizations are already maxed out. ASML doesn’t provide a break-out of DRAM and NAND within memory. What we can see is total system revenue from memory customers. Total memory customer sales were higher in the last three quarters for ASML than they have been in three years, when memory revenue peaked in the first three quarters of calendar 2018. The trend of rising memory bookings continued this quarter, with €2.37B in Q1, up from €1.62B in the previous quarter. These figures are scary, but one should keep in mind they are also convoluted by EUV being added to the mix. These systems are more expensive than any other and have the longest lead time, which will make bookings larger than previously seen in euro terms, but not necessarily in added capacity. Still, the strong trend up in WFE sales into memory is concerning for overcapacity, though ASML’s CEO doesn’t see irrational investment in memory right now.
Semiconductor system sales for the first quarter were 65% Foundry/Logic, 21% DRAM, and 14% NAND. In Q4 of 2021, these percentages were 60%, 25%, 15%, respectively. Management said last quarter they believe memory will be flattish in 2022. But, if we look at the trends in Applied’s sales between DRAM and NAND, DRAM has grown for the last five quarters while NAND has shrunk each quarter over that same time. Overall flat sales into memory doesn’t mean either DRAM or NAND couldn’t be oversupplied. The focus of the call, both prepared remarks and analyst Q&A, was on supply chain, so we didn’t learn much about the DRAM or NAND markets. The supply chain constraints will help keep these markets in balance by preventing memory makers from adding all the WFE capacity they would otherwise like to.
Of KLA’s sales, 26% were for DRAM use and 11% were for NAND. In the June quarter, DRAM is forecasted to be 66% of memory and NAND will be 34%. Memory is expected to be 44% of total revenue in the June quarter. It was 37% of revenue in the March quarter. DRAM revenue is growing, though in last quarter’s call management said they believe NAND will outgrow DRAM for 2022 overall. Putting these two items together one can concluded we will see NAND outgrow DRAM in the third and fourth quarters. Technology migrations and the transition to EUV have been drivers in the strength in equipment sales to DRAM customers. KLA believes total WFE demand in 2022 will exceed $100B, led by Foundry/Logic.
In the March quarter, system revenue broke down as follows (previous quarter in parenthesis): NVM = 39% (35), Foundry = 21% (31), DRAM = 27% (23), Logic/Other = 13% (11). Stronger memory sales in the quarter were led by DRAM, which set a revenue record in the quarter. Like Applied, Lam is constrained by their supply chain and a shortage of components. Also like Applied, Lam didn’t provide any update on their views of the memory markets. Last quarter they said the memory market was “balanced”.
I looked back at the last five years of WFE revenue from the big four vendors. I want to see how total dollar spending on DRAM and NAND today compares to the last cycle. ASML doesn’t break their memory revenue out between DRAM and NAND, which is the company I’d most like to have this from. Second best is KLAC as they don’t have higher market share in one or the other segment, as Lam and Applied do. Also, the only company that does break out DRAM and NAND and has done so going back to at least 2017 is Applied. What I can see is that total WFE sales to DRAM (excluding ASML) have risen in each of the last five quarters. NAND has been the mirror image of this, declining each quarter since peaking in Q1 of 2021. So far, the DRAM market has absorbed this added capacity without tipping into oversupply. ASML forecasts total memory sales to be 25% higher than the previous peak in 2018. I think a good assumption is this will be driven by DRAM because it is more lithography intensive than NAND. My view of the DRAM market from this analysis is that supply and demand are currently close to balance. Micron’s margins have been flat for three quarters and their guidance is that this will continue for at least one more period. It is highly unusual for memory markets to be steady for more than a couple of quarters. Historically, they are always swinging one way or the other, into oversupply or into undersupply. The magnitude of WFE sales to memory that ASML is forecasting makes me continue to lean in the direction that DRAM will become oversupplied either late in 2022 or early in 2023. But supply chain constraints are holding back memory demand as well as the addition of more wafer capacity. I am continuing to hold my Micron shares, waiting to see if ASPs turn up later in 2022.
-S. Hughes (long MU)