Question Is the cost of RAM going up everywhere?

[DZero]

ASUS entering the DRAM market next year according to this rumor:

ASUS Rumored To Enter DRAM Market Next Year To Tackle Memory Shortages

ASUS is rumored to enter the DRAM manufacturing segment by 2026, which should ensure a stable memory supply for its PC lineups.

wccftech.com

”ASUS plans to set up dedicated DRAM production lines by the end of Q2 2026 if memory prices and supply don't return to normal”

Could this be real? Does ASUS have DRAM production capabilities, or do they perhaps intend to buy it from somewhere? 🤔

Maybe Taiwan wants to step up since they realize that this is turning critical that only 2 (ok, 3) countries have all the fab process.

Don't be surprised if Japan enters too. At least Japan and China taking cover for DDR4 and low tier DDR5, stabilizing the prices.

Long live the Free Market.

 

[DrMrLordX]

Maybe Taiwan wants to step up since they realize that this is turning critical that only 2 (ok, 3) countries have all the fab process.

Don't be surprised if Japan enters too. At least Japan and China taking cover for DDR4 and low tier DDR5, stabilizing the prices.

It would be nice, but it's hard to tell if the Asus announcement is related to production of actual DRAM chips or if it just refers to production of DIMMs or what it is that they really mean. Spitting out an entirely new fab for DRAM should take longer than a year.

 

[Abwx]

No need of a new fab, they can rely on TSMC, UMC or even Global foundries, not counting Samsung and eventualy SMIC.

 

[DrMrLordX]

No need of a new fab, they can rely on TSMC, UMC or even Global foundries, not counting Samsung and eventualy SMIC.

If they have the wafers and proper processes available, sure. But doesn't DRAM require specialized 10nm processes?

 

[Abwx]

If they have the wafers and proper processes available, sure. But doesn't DRAM require specialized 10nm processes?

That s likely planar tech stuck on DUV, nothing specialised here and much more manufacturable than finfet even by second tier foundries, FI most advanced Samsung DDR5, LPDDR5 and LPDDR6 RAMs use a 12nm process, guess that for 5600-6000 T/s even 14-16nm is good enough.

 

[DrMrLordX]

FI most advanced Samsung DDR5, LPDDR5 and LPDDR6 RAMs use a 12nm process, guess that for 5600-6000 T/s even 14-16nm is good enough.

Is it really a 12nm process? I had thought they were using processes labeled 10x and 10y etc. Of course what's in a name . . . also seems like GF should be a decent candidate for potential DRAM production. Maybe they could uh cut a deal with Micron?

 

[marees]

Is it really a 12nm process? I had thought they were using processes labeled 10x and 10y etc. Of course what's in a name . . . also seems like GF should be a decent candidate for potential DRAM production. Maybe they could uh cut a deal with Micron?

GF does something known as embedded memory (on 28nm)


GlobalFoundries and Microchip are collaborating on the production of the third generation embedded SuperFlash technology non-volatile memory flash on GF’s 28 nm process.

The flash memory chips are designed for use in smart cards, microcontrollers and internet of things (IoT) devices.

https://electronics360.globalspec.com/article/20258/gf-begins-28-nm-embedded-flash-memory-production

Flash​

In production since 2008, GF’s high-performance embedded Flash memory has the endurance and scalability that make it a reliable memory option for a variety of applications in automotive, industrial, IoT and SmartCard markets.

• Highly reliable and proven technology qualified for 28, 40, 55 and 130nm process nodes with the lowest mask count 28nm in the industry
• Qualified to Auto Grade 1 on 40, 55 and 130nm, and also available in conjunction with power management 130 BCD and 55 BCDLite® technologies
• Integrated alongside RF and ULL/ULP blocks
• Design, test and package turnkey services

Embedded Memory | GlobalFoundries

Bring outstanding reliability and proven manufacturability to all your embedded applications with memory from GF Your customers’ demand for memory […]

gf.com

 

[DrMrLordX]

GF does something known as embedded memory (on 28nm)

Glad to see they're keeping the 28nm fabs busy. Might be time for them to expand their product portfolio with 12nm+ or 12FDX . . .

 

[DZero]

Glad to see they're keeping the 28nm fabs busy. Might be time for them to expand their product portfolio with 12nm+ or 12FDX . . .

They had the chance but quited hard... @NostaSeronx had info, was too technical, but was on spot at least in the nm process. They have the chance to retake the market with 12nm.

 

[marees]

Didn't know this. Groq's compute on memory LPUs are fabbed at GlobalFoundries 14nm !!

Groq’s latest generation of LPUs is, in fact, fabbed at a 14nm node at GlobalFoundries. While they are transitioning to newer nodes at Samsung...

The $20 Billion Admission: Why NVIDIA Just Bought Into the ASIC Revolution with Groq

We got news that Nvidia was »acquiring« (more acquihire) the chip ASIC startup Groq for around $20B. What does this mean for Nvidia and the industry?

www.uncoveralpha.com

 

[DrMrLordX]

Didn't know this. Groq's compute on memory LPUs are fabbed at GlobalFoundries 14nm !!

Wow, interesting. Sad that they'll be losing the business soon.

 

[jamesdsimone]

Used PC2133 and PC2400 DDR4 ECC memory is going for about 60USD for a 16Gb module on ebay. Ouch.

 

[manly]

About that Asus report...

https://www.tomshardware.com/pc-components/dram/no-asus-isnt-going-into-memory-manufacturing-taiwanese-tech-giant-issues-statement-smashing-rumor

Maybe Taiwan wants to step up since they realize that this is turning critical that only 2 (ok, 3) countries have all the fab process.

Don't be surprised if Japan enters too. At least Japan and China taking cover for DDR4 and low tier DDR5, stabilizing the prices.

Long live the Free Market.

What does the T in TSMC stand for?

 

[DrMrLordX]

About that Asus report...

Shucks. That would have been interesting to see.

 

[Fjodor2001]

Well, with DDR profit margins like we’re seeing now some new player(s) will likely enter the arena. Be it ASUS and/or someone else.

The question is what companies are the most likely candidates.

 

[marees]

Well, with DDR profit margins like we’re seeing now some new player(s) will likely enter the arena. Be it ASUS and/or someone else.

The question is what companies are the most likely candidates.

Maybe joint partnership between Japanese & Taiwanese corporations??

 

[DZero]

Shucks. That would have been interesting to see.

Still the Chinese could be around.

 

[DAPUNISHER]

LOL

 

[Steltek]

Well, with DDR profit margins like we’re seeing now some new player(s) will likely enter the arena. Be it ASUS and/or someone else.

The question is what companies are the most likely candidates.

The problem with that theory is that it would require any such company initially entering the DDR memory chip market to source the manufacturing equipment. Then, they have to spend 3-5 years building and validating a fab to actually manufacture it in. Making DRAM chips isn't the same as just making printed circuit boards.

And, if memory is still in short supply at that time, both Microsoft and Google are absolutely desperate for DRAM supply contracts and would likely eat up any additional supply as quickly as it comes online. We, as end users, wouldn't see any improvement in that situation.

Plus, that scenario doesn't even cover what will happen to such a new memory manufacturer when the market eventually corrects and goes into a historical oversupply situation. It always does this, at some point.

Such companies typically don't maintain money loosing operations that cut into the bottom line.

I just don't see any of that happening, especially with the market being so close to shifting to DDR 6. Especially when the AI companies are going to want the entire supply of that when it is released, putting us back in the same situation yet again.

 

[Fjodor2001]

And, if memory is still in short supply at that time, both Microsoft and Google are absolutely desperate for DRAM supply contracts and would likely eat up any additional supply as quickly as it comes online. We, as end users, wouldn't see any improvement in that situation.

Not sure I follow the logic. Sounds like you don’t think additional supply would result in lower prices, all else equal.

But it will of course be better than if no supply is added. Microsoft and Google would want the RAM you mentioned regardless, so if no supply is added the situation will be even worse, because then they’ll purchase it from somewhere else driving price higher.

 

[DZero]

The problem with that theory is that it would require any such company initially entering the DDR memory chip market to source the manufacturing equipment. Then, they have to spend 3-5 years building and validating a fab to actually manufacture it in. Making DRAM chips isn't the same as just making printed circuit boards.

And, if memory is still in short supply at that time, both Microsoft and Google are absolutely desperate for DRAM supply contracts and would likely eat up any additional supply as quickly as it comes online. We, as end users, wouldn't see any improvement in that situation.

Plus, that scenario doesn't even cover what will happen to such a new memory manufacturer when the market eventually corrects and goes into a historical oversupply situation. It always does this, at some point.

Such companies typically don't maintain money loosing operations that cut into the bottom line.

I just don't see any of that happening, especially with the market being so close to shifting to DDR 6. Especially when the AI companies are going to want the entire supply of that when it is released, putting us back in the same situation yet again.

Well... DDR6 shift might get delayed and focused on servers, client side would wait for more time.

 

[dangerman1337]

Well... DDR6 shift might get delayed and focused on servers, client side would wait for more time.

Considering Zen 7 (hell wouldn't be surprised of a Zen 7+ on A14 SPR) is on AM5 and Hammer Lake is 2H of 2029/1H of 2030 on LGA 1954 which uses DDR5 I think Client side DDR6 is 2H of 2030 at the absolute earliest. And even then we'll probably see Zen 7(+) and Hammer Lake be in production for a long time (and for the love of god AMD & Intel, don't EOL Zen 7 X3D or Hammer Lake eLLC until very late DDR6 lifecycle). DDR6 Client side platforms will be very expensive and probably not worth upgrading to until when DDR7 comes to happen.

At this point end of this decade will probably be the best time to get a PC if DDR5 gets really cheap and the hyperscalers move ASAP onto DDR6 and then DDR6 Clientside gets justifiable when DDR7 is going to be on the horizon.

 

[maddie]

Wouldn't it be tragicomic if the current AI mania regresses the main PC market to where it was decades ago? With AI & related CaPex the main US economic growth driver, is there a solution to not being screwed?

As a trivial consequence, think of the game developers thinking Ram was going to be available in large volumes by release date.

 

[DrMrLordX]

With AI & related CaPex the main US economic growth driver, is there a solution to not being screwed?

If the demand is there, eventually investors will take the risk on expanding production. There would need to be some effective bifurcation between AI and consumer supply channels. The problem we're having right now is that DRAM, NAND, and HBM suppliers are all essentially one and the same.

 

[DZero]

Wouldn't it be tragicomic if the current AI mania regresses the main PC market to where it was decades ago? With AI & related CaPex the main US economic growth driver, is there a solution to not being screwed?

As a trivial consequence, think of the game developers thinking Ram was going to be available in large volumes by release date.

Well.. China would enter and supply half of the world, then screw HARD the US and the West by blocking the rare earths supply for their own production.