Laptop maker Framework announces another immediate memory price hike, says additional increase expected within a month — encourages buyers to bring their own memory and check PCPartPicker for better deals

Tanakoi said:

Nope. You forget that higher prices not only act to reduce demand, but also to stimulate supply. Current memory prices are getting chipmakers to allocate more capacity to RAM production lines. Prices will likely rise a bit further over the next 3-6 months, but will then begin descending.

There are no more RAM production lines, they’re already maxed out

VizzieTheViz said:

Apple has never been shy about charging an arm and a leg for upgrades, I don’t see them saying “an arm and a leg is well and good but you can keep your kidneys” any time soon.

Best option is not to buy memory at these prices unless you need it for work and can pass the cost on to your customers.

I doubt they’d be affected this soon. They price them up and adjust pricing of models to account for cost over time. Plus they have contracts with everyone for static pricing. Plus it’s come at a good time for them between large production cycles anyway.

Tanakoi said:

Nope. You forget that higher prices not only act to reduce demand, but also to stimulate supply. Current memory prices are getting chipmakers to allocate more capacity to RAM production lines. Prices will likely rise a bit further over the next 3-6 months, but will then begin descending.

Logic and DRAM production lines are completely dfifferent. You can't jsut re-purpose logic-manufacturing equipment to make DRAM. You have to replace most of the fab equipment. Things like the transports etc. that zip wafers around the line can of course be re-used, but most of the equipment that does the etching, depositions, lithography etc. would have to be swapped out for ones designed for making DRAM. (NOTE: HBM, which is where most of the DRAM capacity is being switched to produce, does use logic chips on the bottom layer of the stack, so those can of course be built in logic fabs, and currently are, but HBM DRAM is less dense and has a higher failure rate than traditional DRAM, therefore to make 8GB of HBM requires the DRAM manufaturing capacity of at least 2x (if not 3x) of DRAM, so switching low-margin DRAM manufacturing for high-margin HBM means a much greater than 1:1 loss in DRAM capacity, thus further excerbating the DRAM shortage) .

Therefore in all practical purposes the world is at 100% production capacity for making DRAM. It will require 2-3 years to substabtially increase manufacturing capacity of RAM by building new lines (3-4 years) or converting logic lines (2-3 years) to DRAM.

You have to remember that the fab equipment to make this sort of stuff isn't just sitting on shelves waiting to be sold and quickly installed by the RAM manufacturers. Every one of these items is 'special order' equipment. You place an order with the vendor of the equipment, and they then go and make it for you, which, depending on how big the queue already is with other manufacteres orders, is likely 18months at minimum, more like2-3 years. If you are really lucky someone who ordered equipment has cancelled their order (likely went bankrupt and left the business entierly) so you may be able to pick up a device quickly, but you need many such devices to build a production line, not just one or 2.

eldakka1 said:

Logic and DRAM production lines are completely dfifferent. You can't jsut re-purpose logic-manufacturing equipment to make DRAM.

Completely different? No, of course not. Logic requires more CD control and more metal layers, so repurposing a DRAM line for logic is indeed quite difficult, but the other direction is substantially easier. Your figure of "2-3 years" is quite inflated ... 9-12 months is much closer to the mark. Actually, it could be theoretically done with just recalibration -- no retooling at all -- though the manufacturer would get killed on efficiency and yields. In a normal market, that wouldn't make sense, but if DRAM prices rise much further, it might be economic.

And there's another factor you're forgetting: memory planned for introduction in 2027 are now being accelerated. At the margins that now exist in the industry, manufacturers can afford to pay for a lot of overtime and expedited delivery charges.

Tanakoi said:

Completely different? No, of course not. Logic requires more CD control and more metal layers, so repurposing a DRAM line for logic is indeed quite difficult, but the other direction is substantially easier. Your figure of "2-3 years" is quite inflated ... 9-12 months is much closer to the mark. Actually, it could be theoretically done with just recalibration -- no retooling at all -- though the manufacturer would get killed on efficiency and yields. In a normal market, that wouldn't make sense, but if DRAM prices rise much further, it might be economic.

And there's another factor you're forgetting: memory planned for introduction in 2027 are now being accelerated. At the margins that now exist in the industry, manufacturers can afford to pay for a lot of overtime and expedited delivery charges.

Aside from it’s not the same you need different equipment. Also they’re not being accelerated they’re expected to come online in 2027/28 as the current demand is unlikely to last through to then and they were burned the last time this happened

LordVile said:

Aside from it’s not the same you need different equipment.

Name one. Be specific.

LordVile said:

Also they’re not being accelerated they’re expected to come online in 2027/28.

Nope.

SK Hynix, M15X (Cheongju): This new DRAM fab starts production in the first half of 2026, and a an 8x capacity increase in their advanced 1c (6th-gen) at Icheon by Q3 2026.
Samsung: plans to boost 1nm-class DRAM capacity to 200,000 wafers/month by Q2 2026 and to repurpose NAND facilities (Pyeongtaek P4) for DRAM by Q3 2026.

Tanakoi said:

Name one. Be specific.


Nope.

SK Hynix, M15X (Cheongju): This new DRAM fab starts production in the first half of 2026, and a an 8x capacity increase in their advanced 1c (6th-gen) at Icheon by Q3 2026.
Samsung: plans to boost 1nm-class DRAM capacity to 200,000 wafers/month by Q2 2026 and to repurpose NAND facilities (Pyeongtaek P4) for DRAM by Q3 2026.

First, some of this work started 12 months ago on increasing DRAM fab capacity. Therefore 2-3 years starting 12 months ago means 1-2 years from today, which gets us into late 2026/2027. If they start from today it'd take 2-3 (closer to 4) to build and outfit a new fab. If there is spare clean-room space in existing fabs to allow adding lines to them in the same building, then knock a year, 18 months off, as you don't have the requirements for building a new building. But there isn't a lot of that 'reserve' cleanroom space, DRAM is so slim margin that having unused cleanroom would be a huge financial loss.

2ndly, NAND is not logic. Converting NAND lines to DRAM is simpler than converting logic, as NAND is more similar to DRAM than it is to logic. But, again, it still takes 1-2 years to do that work, and again, some will have predicted at least some of the increase and started work on some expansion/conversion 12 months ago, which again makes it in the ballpark of 2 years if they are bringing them online next year.

The DRAM manufacturers don't just sit on their arses until some major event happens and then start working on plans then. Most manufacturing businesses spend a lot of time and effort on predicting future requirements, whether that's an increas of decrease, and plan accordingly. While the massive change in demand would not have been predicted, say, 2 years ago, there is usually a broad increase in demand over time (just like inflaction) that many would have planned for. So they likely would have been planning small increase, 10-15%, in fab capacity 2 years ago. So now they are 'rushing' those into production sooner. The question is how much are they going to react to the current surge? As has been mentioned by others, DRAM is in a perpetual boom-bust cycle, undersupply then oversupply. So bringing forward existing plans to modestly expand capacity is a no-brainer. Maybe even doing some conversions of NAND to DRAM. But large capacity increases are more problematic, It costs billions and takes years to set up production lines, and if the demand recedes, then those who invested most into excessively expanding capacity will be left holding the bag. Potential bailouts if not outright bankrupcies are possible, as has happened before in the DRAM industry. In the 90's the register used to joke about the seven dramurai, the 7 Japanese memory makers. Now, most of them have gone, bankrupt, sold off, merged. Memory is a cut-throat, slim margin business (which is why they are eyeing off high-margin HBM at the expense of low-margin DRAM).