Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Are there any memory industry insiders on this forum? Are any of the memory suppliers working on significantly expanding production capacity?

I realise that fabs take a significant amount of time and money to build so if yes (to my question above) it wouldn’t be a quick fix but I’m wondering if new fabs planned or under construction are now on a much bigger scale than would maybe have been planned a year or two ago in order to better match supply with this explosion in demand.

I’ve read many posts from people longing for the AI bubble to burst so that some sanity returns to the memory market but I’m wondering whether there might be an alternative light at the end of this tunnel - the supply-side adapting to this new world order.

Edit: Sorry, I missed previous posts asking the same question. I’m still very interested in any detailed answers though in terms of when sufficient extra capacity might come on line to ease the current situation.
Honestly even if 10 new fabs magically appeared tomorrow I dont think we’d suddenly have a memory surplus. The AI giants would just eat the additional supply and we’d probably end up right back where we started.


The thing about AI is that there really isnt much of a point where “more memory” stops being useful in the same way it does with normal consumer hardware. Bigger models, longer context windows, larger batches, more concurrent inference, bigger KV caches etc all create ways to use more memory.


Obviously there are diminishing returns eventually. AI doesnt have an infinite appetite for memory. But as memory becomes available and cheaper it can also make workloads that werent previously economical or practical worth doing.


So I dont think the question is really “when will enough fabs come online to fix the shortage?”


Its more like


can the memory industry increase supply faster than the AI industry can figure out new ways to consume it?


And honestly I’m not convinced it can.
 
This is going to be so much fun!
Indeed...

For those not obsessed with shiny and new, not obsessed with annual upgrades, not obsessed with spending money, this presents a wonderful opportunity.

An opportunity for those with serviceable iPhones to sit back and enjoy the show.

No need to upgrade a perfectly good phone, no need to feed the machine. Watch and enjoy how the movie plays out, who wins, who loses and what this behemoth of a company does over the next year or two.

A wonderful opportunity to get your money's worth out of the iPhone and other Apple gear you currently own.
 
  • Like
Reactions: eifelbube
Honestly even if 10 new fabs magically appeared tomorrow I dont think we’d suddenly have a memory surplus. The AI giants would just eat the additional supply and we’d probably end up right back where we started.


The thing about AI is that there really isnt much of a point where “more memory” stops being useful in the same way it does with normal consumer hardware. Bigger models, longer context windows, larger batches, more concurrent inference, bigger KV caches etc all create ways to use more memory.


Obviously there are diminishing returns eventually. AI doesnt have an infinite appetite for memory. But as memory becomes available and cheaper it can also make workloads that werent previously economical or practical worth doing.


So I dont think the question is really “when will enough fabs come online to fix the shortage?”


Its more like


can the memory industry increase supply faster than the AI industry can figure out new ways to consume it?


And honestly I’m not convinced it can.
The endless burning of cash will come to an end at some point. Investors are going to want a return on their investment that they're simply not going to find with most AI solutions that are being pushed right now. Like the dotcom boom, there will be big winners, but there will be major losers, and the appetite for risk will moderate eventually.
 
  • Like
Reactions: apple4everOG
They could literally purchase an iPhone 17 Pro and it'll be exactly the same as the iPhone 18 Pro, except that the 17 Pro is available right now. There might be a 1mm thicker body for a 1mm thicker battery and then an A20 chip on the new phone... Whoopty-doo...Not worth waiting 4 months and paying $300 extra for it.

- The same mediocre cameras.
- The same mediocre aluminum design.
- The same lithium battery --- and not Silicon Carbon.
- The same screen with no fingerprint reader.
- The same low-watt charging capabilities.

The real show this September will be the foldable iPhone, but most people won't budget for a phone at that price point anyways.
lEaSe It ThEn LoL
 
This is starting to get downright stupid.

When are regulators going to step in and get this DRAM oligopoly under control?
When we either get...
a) Bureaucrats that know anything about technology
b) Lobbyists to pay off the right people
 
The memory shortage is expected to worsen in 2027. Therefore, if I was a decision maker at Apple, I would consider moving the launch of the 20th anniversary iPhone to Spring 2028. Heck, you could make an argument for a June 29th 2028 launch.
 
  • Like
Reactions: apple4everOG
Which DRAM manufacturers are adding new fab capacity?
For number of fab plants in the pipeline, here's what I found with a combination of Google searches and ChatGPT queries. [Yes, I know it's ironic to be using ChatGPT on a thread discussing a key negative consequence of ChatGPT and its ilk 😉]

TLDR is that quite a bit of this capacity starts hitting during 2027-28, which is why a substantial RAM price correction beginning in 2028 is plausible, and also what most of the industry analysts seem to be predicting. Specifically, 2028 is the most likely year for a price drop, but we'll probably have to wait till 2029 for a return of normal consumer RAM pricing.*

Fab plants already in the pipeline prior to the start of the RAM crisis (let's call that Q4 2025):

Samsung (2), SK hynix (5), Micron (6), CXMT (0), Winbond (0), and Nanya (1). Seven of these have expected first production dates of 2029 or earlier.

Fab plants added to the pipeline after the start of the RAM crisis:

Samsung (3), SK hynix (2), Micron (1), CXMT (2), Winbond (1), and Nanya (0). Three of these have expected first production dates of 2029 or earlier.

And the same as above, except where the numbers in parentheses are the estimated number of 300 mm DRAM wafers per year. You can get about 3,000 GB of consumer RAM per wafer, but note that those wafer numbers don't tell you how much is allocated for consumer RAM versus server RAM, HBM, etc.

Fab plants already in the pipeline prior to the start of the RAM crisis:

Samsung (≈4.8–7.2 million/year), SK hynix (≈10.6 million/year), Micron (≈3.6 million/year, estimated), CXMT (0), Winbond (0), and Nanya (≈0.54 million/year).

Fab plants added to the pipeline after the start of the RAM crisis:

Samsung (≥1.2 million/year + capacity TBD for two southwestern fabs), SK hynix (capacity TBD for two southwestern fabs), Micron (≈0.6 million/year), CXMT (≈3.6 million/year incremental, approximately), Winbond (≈0.6 million/year), and Nanya (0).
********************

*From https://www.notebookcheck.net/DRAM-crisis-Analysts-expect-drastic-price-drop-in-2028.1337992.0.html
"According to the latest estimate from Bernstein analysts, an improvement could be in sight by 2028, but only after prices have risen further next year.

The price per Gb of DRAM is expected to rise by 43.2 percent to $2.23 in 2027, but then fall by 52.9 percent to $1.05 in 2028. As expected, prices in 2028 will still be twice as high as they were before the DRAM crisis, but at least one-third lower than they are today. A similar trend is expected for SSDs. While 1 GB of NAND flash memory costs an average of $0.30 today, the price is expected to rise only slightly to $0.32 next year, and eventually fall by a remarkable 68.8 percent to $0.10 in 2028, meaning prices are expected to remain 25 percent higher than before the crisis."
 
Last edited:
Indeed...

For those not obsessed with shiny and new, not obsessed with annual upgrades, not obsessed with spending money, this presents a wonderful opportunity.

An opportunity for those with serviceable iPhones to sit back and enjoy the show.

No need to upgrade a perfectly good phone, no need to feed the machine. Watch and enjoy how the movie plays out, who wins, who loses and what this behemoth of a company does over the next year or two.

A wonderful opportunity to get your money's worth out of the iPhone and other Apple gear you currently own.
Except I need to upgrade because I have no Apple Silicon Macs. They are all 6+ years old too. This was gonna be the year. And the slop hit me in the face instead.
 
For number of fab plants in the pipeline, here's what I found. TLDR is that quite a bit of this capacity starts hitting during 2027-28, which is why a substantial RAM price correction beginning in 2028 is plausible, and also what most of the industry analysts seem to be predicting. Specifically, 2028 is the most likely year for a price drop, but we'll probably have to wait till 2029 for a return of normal consumer RAM pricing.*

Fab plants already in the pipeline prior to the start of the RAM crisis (let's call that Q4 2025):

Samsung (2), SK hynix (5), Micron (6), CXMT (0), Winbond (0), and Nanya (1). Seven of these have expected first production dates of 2029 or earlier.

Fab plants added to the pipeline after the start of the RAM crisis:

Samsung (3), SK hynix (2), Micron (1), CXMT (2), Winbond (1), and Nanya (0). Three of these have expected first production dates of 2029 or earlier.

And the same as above, except where the numbers in parentheses are the estimated number of 300 mm DRAM wafers per year. You can get about 3,000 GB of consumer RAM per wafer, but note that those wafer numbers don't tell you how much is allocated for consumer RAM versus server RAM, HBM, etc.

Fab plants already in the pipeline prior to the start of the RAM crisis:

Samsung (≈4.8–7.2 million/year), SK hynix (≈10.6 million/year), Micron (≈3.6 million/year, estimated), CXMT (0), Winbond (0), and Nanya (≈0.54 million/year).

Fab plants added to the pipeline after the start of the RAM crisis:

Samsung (≥1.2 million/year + capacity TBD for two southwestern fabs), SK hynix (capacity TBD for two southwestern fabs), Micron (≈0.6 million/year), CXMT (≈3.6 million/year incremental, approximately), Winbond (≈0.6 million/year), and Nanya (0).
********************

*From https://www.notebookcheck.net/DRAM-crisis-Analysts-expect-drastic-price-drop-in-2028.1337992.0.html
"According to the latest estimate from Bernstein analysts, an improvement could be in sight by 2028, but only after prices have risen further next year.

The price per Gb of DRAM is expected to rise by 43.2 percent to $2.23 in 2027, but then fall by 52.9 percent to $1.05 in 2028. As expected, prices in 2028 will still be twice as high as they were before the DRAM crisis, but at least one-third lower than they are today. A similar trend is expected for SSDs. While 1 GB of NAND flash memory costs an average of $0.30 today, the price is expected to rise only slightly to $0.32 next year, and eventually fall by a remarkable 68.8 percent to $0.10 in 2028, meaning prices are expected to remain 25 percent higher than before the crisis."
Thank you. Something does not add up. From the notebookcheck article:

For example, a Crucial Pro 2 x 16 GB DDR5-6,000 kit now costs $549, up from $89 charged for the same kit last October.

So that is a 600% increase! And that does not correlate with the percentages quoted in the article.

Also the article talks about Gbits of DRAM and GBytes of SSD so I am not sure they have their units correct.

So for DRAM if a Gbit was 50 cents before the "crisis", that means 8GB DRAM was $32 and what was Apple charging for a 8GB upgrade on the MBA? $200 or so.

I hope my arithmetic is correct.
 
Except I need to upgrade because I have no Apple Silicon Macs. They are all 6+ years old too. This was gonna be the year. And the slop hit me in the face instead.

Sounds very similar to what I and a few others were talking about earlier. The RAM shortage has altered everyone's cost to performance equation.

I’ve been through various shortages before. The crypto boom that ate up GPU supply until the market eventually adapted is the first example that comes to mind. I didn't know enough to mine or really get involved at the time. I was too busy trying to become a network engineer, and later I started learning Swift at the same time. But I did get to see what the economics looked like firsthand. Some people around me were spending $10K, $15K, even $20K+ putting together mining farms. I didn't go that far, but I did get to experiment with it.

For myself, when it comes to experimentation, I almost never buy the best thing immediately. I like starting with something that gives me enough capability to develop some firsthand experience and figure out what I actually need. Then, when I have the foundation and the budget, I can make a much better purchase.

For experimentation, I usually try to keep myself around $2,000 max. I'm not necessarily buying today's technology for that purpose. I'm more interested in something that is 2 or 3 years old, because if the experiment doesn't work out, I haven't lost nearly as much.

The memory situation has completely changed that calculation for me. I'm now looking at hardware that is considerably older than I normally would for experimentation. I ended up with an MBP 16" M1 Max 32GB/1TB and an M4 Mac mini 24GB/256GB, which fit right in my $2k budget. The 256GB storage on the M4 I'll upgrade with additional storage, I bought a 1TB SSD that I plan on installing when Apple ships it within the 7-8 weeks I have to wait. I got the M1 Max because I travel a lot, and the M4 mini is going to be my home machine.

What's wild to me is how competitive older Apple Silicon hardware can still be against what you can realistically afford today once you put enough memory into the equation.

A year ago, two years ago, or even four years ago, the budget that gets you into 24GB, 32GB, or higher-memory configurations could put you into considerably newer hardware. Now, if memory capacity is important to your workload, you can end up making a pretty uncomfortable choice: buy newer hardware with 16GB, or spend substantially more to get the memory capacity you actually wanted.

And that matters because RAM isn't just a benchmark number. If your workload actually needs the capacity, running out of memory can mean more compression, swapping to storage, closing applications, reducing how much you can multitask, or simply accepting longer workloads.

Not everyone can just find another $500 or $1,000 because the newest configuration they need happens to cost that much more. Some people have a job starting, school starting, bills to pay, or simply a fixed technology budget that will not allow the time for simply another paycheck or two in order to afford something else.

So I think that's the part that's getting lost in the discussion. A memory shortage doesn't just affect people who want the absolute highest end machines. It changes what ordinary people can realistically afford at a given performance and memory level.

That's why I'm finding myself looking further back into Apple's lineup than I normally would. Not because I suddenly think old hardware is better, but because the price/performance calculation has changed enough that hardware that would normally be considered "old" can make a lot more sense when the alternative is paying a large premium for the memory capacity you actually need.

It's a pretty strange time in the hardware market.
 
Thank you. Something does not add up. From the notebookcheck article:



So that is a 600% increase! And that does not correlate with the percentages quoted in the article.

Also the article talks about Gbits of DRAM and GBytes of SSD so I am not sure they have their units correct.

So for DRAM if a Gbit was 50 cents before the "crisis", that means 8GB DRAM was $32 and what was Apple charging for a 8GB upgrade on the MBA? $200 or so.

I hope my arithmetic is correct.
The 600% increase was for a specific retail item. You can always find specific items that have had shocking price increases.

By contrast, the Bernstein analysis they quoted was for quantity OEM pricing (i.e., what computer makers like Dell and Apple pay, though it doesn't tell you what specific contract prices those brands have negotiated) on DRAM and NAND.

So there's no inconsistency, because they're referring to different (though related) things.

"The price per Gb of DRAM": If you look in the screenshot of the Bernstein estimate they linked, DRAM is listed in Gb, while NAND is listed in GB, which is why the Notebookcheck author defaulted to those specific units. But the Notebookcheck article is giving percent increases, and percent increases are independent of the units chosen. That's because percent increases are just ratios, and when you calculate a ratio, the units cancel out. So the units don't matter.

Thus, since the notebook check article was giving percent increases, what it should have referenced is not "the price per Gb of DRAM", but simply "the price of DRAM".

Yes, $0.50*8*8 = $32

But Apple wasn't charging $200 for 8 GB of DRAM pre-crisis. They were charging $100, i.e., $12.50/GB.

Currently, they're charging $25/GB.

So based on Bernstein's listed ASP's (Average Sales Price), we have:

pre-crisis: $12.50/($0.50*8) ≈ 3 => 3-fold markup
currently: $25/($1.85*8) ≈ 1.7 => 70% markup

But we don't know what Apple's markup actually is, since:

(a) Apple specs very high-end RAM, which may cost more than the ASP; and

(b) Apple may be able to negotiate better prices than average for that particular high-end RAM product, which would push Apple's costs in the opposite direction.

According to https://appleinsider.com/articles/2...e-is-not-immune-to-massive-ssd-cost-increases"In 2012, Apple bought 23% of the world's NAND supply, paying around $0.67/GB. NAND prices fluctuate greatly, but Apple has almost always paid less than market rates — such as in 2018, when it was reportedly paying an incredibly discounted $0.25 per gigabyte."

And for NAND:
pre-crisis: $??/($0.08) ≈ ?? => ??-fold markup [I don't have Apple's pre-crisis $/GB for NAND; maybe someone can supply it.]
currently: $0.50/($0.36) ≈ 1.4 => 40% markup
 
Last edited:
Sounds very similar to what I and a few others were talking about earlier. The RAM shortage has altered everyone's cost to performance equation.

I’ve been through various shortages before. The crypto boom that ate up GPU supply until the market eventually adapted is the first example that comes to mind. I didn't know enough to mine or really get involved at the time. I was too busy trying to become a network engineer, and later I started learning Swift at the same time. But I did get to see what the economics looked like firsthand. Some people around me were spending $10K, $15K, even $20K+ putting together mining farms. I didn't go that far, but I did get to experiment with it.

For myself, when it comes to experimentation, I almost never buy the best thing immediately. I like starting with something that gives me enough capability to develop some firsthand experience and figure out what I actually need. Then, when I have the foundation and the budget, I can make a much better purchase.

For experimentation, I usually try to keep myself around $2,000 max. I'm not necessarily buying today's technology for that purpose. I'm more interested in something that is 2 or 3 years old, because if the experiment doesn't work out, I haven't lost nearly as much.

The memory situation has completely changed that calculation for me. I'm now looking at hardware that is considerably older than I normally would for experimentation. I ended up with an MBP 16" M1 Max 32GB/1TB and an M4 Mac mini 24GB/256GB, which fit right in my $2k budget. The 256GB storage on the M4 I'll upgrade with additional storage, I bought a 1TB SSD that I plan on installing when Apple ships it within the 7-8 weeks I have to wait. I got the M1 Max because I travel a lot, and the M4 mini is going to be my home machine.

What's wild to me is how competitive older Apple Silicon hardware can still be against what you can realistically afford today once you put enough memory into the equation.

A year ago, two years ago, or even four years ago, the budget that gets you into 24GB, 32GB, or higher-memory configurations could put you into considerably newer hardware. Now, if memory capacity is important to your workload, you can end up making a pretty uncomfortable choice: buy newer hardware with 16GB, or spend substantially more to get the memory capacity you actually wanted.

And that matters because RAM isn't just a benchmark number. If your workload actually needs the capacity, running out of memory can mean more compression, swapping to storage, closing applications, reducing how much you can multitask, or simply accepting longer workloads.

Not everyone can just find another $500 or $1,000 because the newest configuration they need happens to cost that much more. Some people have a job starting, school starting, bills to pay, or simply a fixed technology budget that will not allow the time for simply another paycheck or two in order to afford something else.

So I think that's the part that's getting lost in the discussion. A memory shortage doesn't just affect people who want the absolute highest end machines. It changes what ordinary people can realistically afford at a given performance and memory level.

That's why I'm finding myself looking further back into Apple's lineup than I normally would. Not because I suddenly think old hardware is better, but because the price/performance calculation has changed enough that hardware that would normally be considered "old" can make a lot more sense when the alternative is paying a large premium for the memory capacity you actually need.

It's a pretty strange time in the hardware market.
My biggest regret right now is not spending the extra cash on a 64GB M4 Max Studio last year instead of settling for 36GB. I was planning a 2-3 year upgrade cycle that's now looking like 6+ and that extra RAM would have really come in handy over the years as I'm already hitting 28GB or so with my normal daily tasks.

At least I have a desktop PC I can upgrade from 32GB if I can ever stomach the cost but I'm stuck with the 36GB on my Mac.
 
Sounds very similar to what I and a few others were talking about earlier. The RAM shortage has altered everyone's cost to performance equation.

<snip for brevity>

Not everyone can just find another $500 or $1,000 because the newest configuration they need happens to cost that much more. Some people have a job starting, school starting, bills to pay, or simply a fixed technology budget that will not allow the time for simply another paycheck or two in order to afford something else.

So I think that's the part that's getting lost in the discussion. A memory shortage doesn't just affect people who want the absolute highest end machines. It changes what ordinary people can realistically afford at a given performance and memory level.

Yes you are absolutely right. While for myself I want to get a Mac Studio (was planning last year but couldn't quite afford it yet, and now that I can, prices gone up and can't afford it again, and was also waiting a bit for the M5), my wife just needs a MacBook Air. Yet that went up a couple of hundred dollars too.

It's going to affect everything, and not just price, but availability! It's so extremely frustrating.

That's why I'm finding myself looking further back into Apple's lineup than I normally would. Not because I suddenly think old hardware is better, but because the price/performance calculation has changed enough that hardware that would normally be considered "old" can make a lot more sense when the alternative is paying a large premium for the memory capacity you actually need.

It's a pretty strange time in the hardware market.

Yes that's true, I was looking back as well, but even that is getting hard to find! Because the pressure on the new then becomes pressure on the used, as everyone is starting to have to make that calculation for themselves.
 
  • Like
Reactions: Sosab123
I don’t think so. Read the news. It took less than nine months to stand up Apple's new Advanced Manufacturing Center (AMC) in Houston, Texas.

World of difference between a state of the art chip fab and an assembly line that the AMC is. You’re stuck buying lithographic machines from one company in the world to stock your chip fab.

I'd make an even stronger statement. Apple's AMC isn't it even an assembly line. It is a training center. Nothing is made there, other than prototype and demo products associated with its training function.

So the fact that they stood it up in under nine months doesn't tell you anything about how long it would take to bring an actual manufacturing plant, of any sort, online.
 
Last edited:
I'd make an even stronger statement. Apple's AMC isn't it even an assembly line. It is a training center. Nothing is made there, other than prototype and demo products associated with its training function.

So the fact that they stood it up in under nine months doesn't tell you anything about how long it would take to bring an actual manufacturing plant, of any sort, online.

Isn't it also the facility where Apple plans to manufacture Mac minis...?
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.