Hello, I'm Pazz.
Recently, memory stocks have fallen sharply, and it seems many people are having a difficult time psychologically.
Those who have read my book (published in May) would know that I was someone who had the view that memory prices would peak from the second half of this year and then decline due to the periodic memory cycle, but my thinking has changed somewhat recently.
In my view, the memory market shortage itself will likely continue until at least mid-2028, and even that is the "minimum" estimate. I also think that according to CEO Choi's remarks, the shortage could continue until 2030. If that's the case, then Samsung and SK Hynix are indeed very cheap right now. Even if the market peaks by mid-2028, advance pricing would be at most one year, so it's still too early for stock prices to decline.
First, looking at the supply side, as you know, this year's memory demand forecasts have increased several times compared to last year due to Agentic AI. However, memory companies' bit growth supply is in the ten percent range for this year and next year respectively. Even if major new factories are completed in 2028, it's difficult to exceed twenty percent annual bit growth. The point where current capacity would double would likely need to reach close to 2030. Considering that this year's memory demand forecast has increased at least 2x to several multiples compared to last year's expectations, it seems almost certain that AI demand simply cannot be met for years to come.
Even if big tech companies reduce their CAPEX forecasts due to funding constraints, the big picture seems unlikely to change.
And as the proportion of HBM increases, HBM consumes more than twice the capacity of existing legacy DRAM to produce the same amount of bits, so production volumes decrease even further.
All these signals are pointing to a memory ultra super cycle. Of course, it will eventually return to cyclical business, but in my view, a downturn in the memory market within 2-3 years seems unlikely.
And not long ago, Meta unveiled D4 memory recycling technology. Existing server modules were in DIMM format and could be recycled. In fact, almost all of them were taken apart individually and recycled. In other words, once memory chips entered the world, they could be recycled and reused in any form.
However, in the HBM era, HBM memory is attached directly next to the GPU, so recycling is impossible. It's very difficult to remove, and even if removed, they all become defective, making recycling virtually impossible.
In other words, once HBM memory is used, it becomes a bit that disappears from the world. In 3-5 years, when current GPU servers fade out, it becomes a factor that further reduces memory bit supply to the market.
In any case, from my perspective, the memory market itself still appears to have no major issues.
Therefore, if I maintain an appropriate cash allocation and gradually average down each time prices fall, I think the memory companies' stock prices will recover and could challenge all-time highs. To be honest, in the second half (late Q3 to Q4), I believe they could break through all-time highs and create new all-time highs, not just challenge them.
Therefore, for those who have invested in Samsung and SK Hynix, don't worry too much. If you stay focused on your work, I think good days will come again soon.
