I wonder if supply chain constraints currently represent an inherent scepticism of manufacturers that demand will last. It feels to me like a side effect of the self-dealing / incestuous financing that is going on is that manufacturers are not willing to bet on it all materialising and therefore are not scaling up capacity nearly as much as they would if they thought it was certain.
It's funny you mention that because this exact same issue has come up in war in Iran and the closure of the Strait of Hormuz.
What a producer will do now is hedge their risk. They do this by forward-selling their product, in this case oil from an oil well. The peak for spot prices, depending on the particular flavor of oil, peaked at >$170/barrel while future prices (which are the prices we see) didn't really get above $100 while long-dated futures were significantly lower, rarely exceeding $70/barrel by much.
Now this is a problem because there's no incentive to invest in extra production when you can only sell it at $70/barrel, despite the spot price being up to $100/barrel higher. So new wells and new production in general is down despite the supply disruptions and high prices.
I imagine RAM is in a similar position. It takes a long time to bring on more supply and suppliers don't expect those prices to last. Likely a large amount of new supply will probably cause the next crash in the boom-bust cycle.
But there's another problem with all this. A lot of existing manufacturers are switching from DDR5 (and soon DDR6) to HBM because it's way more profitable and there's seemingly insatiable demand for high end GPUs. I expect that in the short-to-medium term we're both going to see no increase in supply and the DDR prices are going to keep rising to the point where in a year we'll probably look back on current prices fondly
All I can say is I'm happy I upgraded my laptop to 64GB (for $200) and my PC to 64GB (also for $200) last year.
.
> A lot of existing manufacturers are switching from DDR5
Nobody is "switching" old process node fabs to new processes, they just aren't building new fabs for old processes.
AI and high density tasks will use memory fabricated with newer processes, freeing the old supply for more elastic, price-sensitive tasks.
The vast majority of memory that has price impacts for most consumers isn't in personal computers, it's in embedded applications that were also impacted by the overall squeeze.
9-Aug-2026: SK Hynix to invest $38 billion building new memory chip plants as demand soars [1]
"In the AI era, technological competitiveness alone is not enough and the ability to supply the required volume at the exact moment customers need it is the ultimate competitive advantage," SK Hynix said in a press release. "We reached this investment decision after a thorough review of market demand."
They will do so in steps, tho. One cleanroom at a time so they don't eat a big loss if this crazy demand for a commodity goes away overnight.
They also took quite some time to decide to ramp up, as this comes a year or more late after they announced they were fully booked for 2026 and 2027.
And, their stock sold off on the news, so investors are not having faith either on the demand to justify capex expansion.
Honestly, it sucks to be a ram company, you are either buying into the hype and then spending years losing money, or you don't and leave a lot of money on the table while competition ramps up and eats your lunch (and then spends years losing money)
It's exactly this - the memory market has gone through exactly this boom/bust since I can remember - there was a great RAM price stabilization around Windows 95 (price stayed the same or went up per megabyte over years) and then a glut of oversupply.
Building new fabs is a very different bet from selling into a temporary shortage: you're committing billions based on what demand will look like several years from now
Manufacturers know demand will last to the extent that big companies are willing to commit to purchases over a given period of time, but supply today is still constrained by capacity built for the pre LLM era. The 500 billion put in by Samsung/SKHynix will increase modern DRAM by roughly 50% but it wont be fully online until 2030.
It might be fully online by 2030 if the current demand lasts. IMHO the demand is unsustainable & build on circular financing deals and hidden debt with no way to be ever profitable.
The memory cartel would be crazy to actually invest into new capacity at this point.
TSMC in the US is a bit of a special case .... wouldn't surprise me at all if they were virtue signalling to the Trump admin more than anything with what they are doing there.
How so? They are diversifying their production chain so that all their eggs aren't in one basket. Taiwan might be threatened one day but their US, Japan, and German plants will much less likely be in the same danger.
That seems wrong to me - the idea of the "Silicon shield of Taiwain" is that alle the critical cutting edge production is all physically in Taiwan & rigged to be destroyed if Taiwan is ever attacked.
If they do diversify, they are compromising a key policy protecting their homeland, it makes no sense!
