Friday, August 7th 2026

SK hynix Invests 54 Trillion Won in Yongin Y2 and Cheongju M17 to Secure Mid-to-Long-Term Production for AI Memory Demand

SK hynix Inc. announced today that it has decided to build new fabs in Yongin and Cheongju to respond to the continuously growing demand for memory in the AI era.
The company approved a total investment of approximately 54 trillion, 35.2 trillion won in Yongin "Y2" fab and 19.1 trillion won in the Cheongju "M17" fab, in a board meeting.

This decision represents execution of the mid-to-long-term investment strategy announced by the company last June. Under its master plan to invest 600 trillion won in the Yongin Semiconductor Cluster and 100 trillion won to expand its Cheongju production base, SK hynix is currently constructing its first fab in Yongin (Y1). With this latest decision, the company is embarking on the construction of subsequent fabs in both Yongin and Cheongju.
According to market research firm Omdia, both DRAM and NAND demand are projected to maintain a compound annual growth rate (CAGR) of 19% from last year through 2030. The company views this growth not as a temporary supercycle, but as a structural transformation in which memory transcends its role as a mere component to become core infrastructure determining AI performance itself.

SK hynix stated that 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. We reached this investment decision after a thorough review of market demand.

Yongin Y2 is the second of four fabs planned sequentially for the Yongin Semiconductor Cluster. It will serve as a DRAM production base spanning a total floor area of 341,000 pyeong (approximately 1,130,000m2). Construction is scheduled to break ground in July next year, targeting the opening of its first cleanroom in June 2029 to produce high-bandwidth memory (HBM) and other next-generation DRAM products. Meanwhile, construction on Y1 is progressing smoothly toward its initial cleanroom opening target of February next year.

SK hynix previously set a goal to advance the completion date of the Yongin Semiconductor Cluster by 12 years—from the originally planned 2045 to 2033—to finish constructing all four fabs. The Y2 construction serves as the second phase supporting this accelerated timeline and the investment spans to October 2031 sequentially. The investment amount includes construction costs for "Business Support Building" with administrative and welfare facilities for Y2 personnel, an "Integrated R&D Center" to consolidate product development testing, analysis, and experiments, as well as auxiliary infrastructure.

For the Yongin Semiconductor Cluster—sprawling over approximately 1.26 million pyeong (approximately 4,160,000 m²) in Wonsam-myeon, Cheoin-gu, Yongin-si, Gyeonggi-do—Phase 1 power and water infrastructure required up through Y2 operation is nearing completion, currently at a 99% progress rate. As fab construction and infrastructure setup have been carried out concurrently, the company plans to proceed with the Y2 construction according to schedule.

In the NAND market, demand is surging rapidly centered on enterprise SSDs driven by the full-scale expansion of AI services. This is further boosted by demand for key-value (KV) cache2storage in AI inference, with application fields expected to broaden further as Agentic and Physical AI are introduced.



SK hynix selected Cheongju as its new NAND fab base because it offers the fastest fab construction timeline. The Cheongju Campus already houses the M11, M12, and M15 fabs, enabling efficient production linked to existing operations on sites equipped with substantial pre-established power and water infrastructure. Cheongju M17 will be built across a total area of 206,000 pyeong (approximately 680,000 m²), breaking ground in February next year with plans to open its first cleanroom in December 2028. The investment period runs through April 2031 according to the master plan.

SK hynix plans to proactively secure production infrastructure based on mid-to-long-term demand discussed with customers, while expanding actual production capacity in alignment with the timing of customer needs. Fabs will be constructed according to the master schedule, whereas cleanroom expansions and equipment installation will take place sequentially in line with demand trends to maximize capital efficiency.

The company expects this investment to secure future growth foundations while strengthening the competitiveness of the domestic semiconductor ecosystem and boosting local economies. The large-scale simultaneous investments in Yongin and Cheongju are anticipated to expand joint growth opportunities for supply chain partners, generate employment, and revitalize local commercial districts, thereby contributing to the sustainable growth of the national economy.

An SK hynix official stated that "This investment is a decision made to seize opportunities in line with the market's growth speed. By proactively securing production capabilities, we aim to establish ourselves as a key partner in AI infrastructure, contributing to the stability of the global AI semiconductor supply chain."
Add your own comment

24 Comments on SK hynix Invests 54 Trillion Won in Yongin Y2 and Cheongju M17 to Secure Mid-to-Long-Term Production for AI Memory Demand

#1
Space Lynx
Astronaut
Good news, we may have to wait awhile still, but eventually markets will catch up with demand and our hobbies will return to normal. Maybe around 2030. I have plenty of backlog to keep me busy until then, assuming the connector on my 5080 does not melt.
Posted on Reply
#2
wolf
Better Than Native
Space Lynxassuming the connector on my 5080 does not melt
It's exceptionally less likely to, and hey wouldn't it be something of this supply constrained period ended with a period of abundance.
Posted on Reply
#3
Space Lynx
Astronaut
wolfIt's exceptionally less likely to, and hey wouldn't it be something of this supply constrained period ended with a period of abundance.
that was my thinking too, there is a small chance we come out of this hellscape actually better than before, all these factories being built, and demand for ai will eventually slump, it could lead to say $100 48gb ram kits for us in 2032, etc. cause the sheer volume will be so much, etc.

it def will take some time, but it's possible this happens.
Posted on Reply
#4
the54thvoid
Super Moderator
wolf...wouldn't it be something of this supply constrained period ended with a period of abundance.
Surplus inventory means shareholder discontent. I doubt the low-level consumer will see much to be happy about when the rush is over.
Posted on Reply
#5
SnoutBaron
approximately 54 trillion won
$38 billion isn't something to scoff at, damn. So they really don't see a bubble at all.
Posted on Reply
#6
Space Lynx
Astronaut
SnoutBaron$38 billion isn't something to scoff at, damn. So they really don't see a bubble at all.
nope, it's a national security concern now, AGI = the new Cold War

whoever gets AGI first and uses it to enhance their military, is the new super power for the next 20 years, that is their thinking... I think :roll:

the money will flow...
Posted on Reply
#7
Pumper
Space Lynxnope, it's a national security concern now, AGI = the new Cold War

whoever gets AGI first and uses it to enhance their military, is the new super power for the next 20 years, that is their thinking... I think :roll:

the money will flow...
That's not how a true AGI would ever function as you would not be able to "use" it to do anything as it would be completely autonomous and superior to your efforts to control it.
Posted on Reply
#8
wolf
Better Than Native
the54thvoidSurplus inventory means shareholder discontent. I doubt the low-level consumer will see much to be happy about when the rush is over.
Like all things time will tell, but yeah for sure the class that stands to profit from constraint would not be pleased.
Posted on Reply
#9
razaron
Space LynxGood news, we may have to wait awhile still, but eventually markets will catch up with demand and our hobbies will return to normal. Maybe around 2030. I have plenty of backlog to keep me busy until then, assuming the connector on my 5080 does not melt.
Things have been moving towards thin clients backed by cloud sevices for ~15 years now. Recent price hikes have just accelerated this trend.
Once supply catches up/new AI datacentre demand slows down, datacentres can just be retooled for cloud services. In that scenario it wouldn't make sense to lower component prices for plebians, especially if smaller AIBs and PC makers have died off in the meantime.
Posted on Reply
#10
Mack4285
razaronThings have been moving towards thin clients backed by cloud sevices for ~15 years now. Recent price hikes have just accelerated this trend.
Once supply catches up/new AI datacentre demand slows down, datacentres can just be retooled for cloud services. In that scenario it wouldn't make sense to lower component prices for plebians, especially if smaller AIBs and PC makers have died off in the meantime.
I think right now it's all stuck in "wait years until supply returns". But if it doesn't, do we all just sit there and accept that a thin client PC for cloud is the only option? It's going to be very obvious if no effort is made to make PC components affordable for normal consumers. Will that just be silently accepted? And if you cared about privacy before, it's completely gone if everything is in the cloud.
Posted on Reply
#11
trparky
Space LynxMaybe around 2030.
That's a long damn time to wait, it might as well be dog years. By the time it's done, the PC marketplace is going to look a hell of a lot different and not for the better.
Posted on Reply
#13
razaron
Mack4285I think right now it's all stuck in "wait years until supply returns". But if it doesn't, do we all just sit there and accept that a thin client PC for cloud is the only option? It's going to be very obvious if no effort is made to make PC components affordable for normal consumers. Will that just be silently accepted? And if you cared about privacy before, it's completely gone if everything is in the cloud.
It's just a trend, not binary. Personal computing was expensive in the 80s/90s, had a golden era in the 2000s then started getting expensive again 15 years ago, with no sign of stopping.

A "normal consumer" ranges from property owners to people on foodstamps. If you could afford an 8800gt ($400 after inflation) in 2007, you may or may not be able to afford a modern equivelant now (5070ti?).

I'm personally priced out of current gen GPUs, for example. I'd like to enter the property market sometime in the next 50 years, so can't justify dropping £1k on a luxury item.
Posted on Reply
#14
Dr. Dro
Space LynxGood news, we may have to wait awhile still, but eventually markets will catch up with demand and our hobbies will return to normal. Maybe around 2030. I have plenty of backlog to keep me busy until then, assuming the connector on my 5080 does not melt.
Did you order a ThermalProtect? Mate... just do it :fear:
razaronIt's just a trend, not binary. Personal computing was expensive in the 80s/90s, had a golden era in the 2000s then started getting expensive again 15 years ago, with no sign of stopping.

A "normal consumer" ranges from property owners to people on foodstamps. If you could afford an 8800gt ($400 after inflation) in 2007, you may or may not be able to afford a modern equivelant now (5070ti?).

I'm personally priced out of current gen GPUs, for example. I'd like to enter the property market sometime in the next 50 years, so can't justify dropping £1k on a luxury item.
The thing is that it was supposed to get cheaper and faster, it's the way semiconductor technology progresses forward. Honestly, for what I do? I think I'm covered. It'll basically mean the end of my career as a techie, but if I drop chasing upgrades, I think I will be gaming comfy for a while.
Posted on Reply
#15
JimmyDoogs
How much has anyone here actually spent on AI? I spent 15 bucks to try out Grok and realized it's not even close to running things natively off your own computer so I'll never subscribe again. They're asking $300 a month for like 30 mins of use per day...
Posted on Reply
#16
sam_86314
JimmyDoogsHow much has anyone here actually spent on AI? I spent 15 bucks to try out Grok and realized it's not even close to running things natively off your own computer so I'll never subscribe again. They're asking $300 a month for like 30 mins of use per day...
On subscriptions, $0. I've been messing with the free tier of Gemini during my Linux adventures, but that's about it. No way in hell am I directly giving these companies any money for what they've done to the hardware market.

I mainly have been tinkering with running local models since that's the way this whole industry should go, so I guess my two GPU purchases last year ($350 for an Arc Pro B50 and $625 for an RX 9070 XT) could be considered my AI expenditure, even though they primarily get used for gaming (yes, even the B50).
Posted on Reply
#17
razaron
Dr. DroDid you order a ThermalProtect? Mate... just do it :fear:



The thing is that it was supposed to get cheaper and faster, it's the way semiconductor technology progresses forward. Honestly, for what I do? I think I'm covered. It'll basically mean the end of my career as a techie, but if I drop chasing upgrades, I think I will be gaming comfy for a while.
Honestly, my main incentive for upgrading is VR. If I ignore VR (since it is summer, lol), the 6800XT is adequate for 60fps 2560x1600. That's without upscaling/framegen, dropping other settings isn't very noticable nowadays (UE5).
I'm sure come winter I'll try Skyrim/Fallout 4 in VR again, struggle to run it at max resolution, then get the upgrade itch only to lose it by summer :laugh:
Posted on Reply
#18
JimmyDoogs
To be honest, I wouldn't be entirely against buying $60 AI programs as long as it meant running them natively. I can't forsee this data center driven expenditure as worth while since things like real time video isn't even possible yet for things like Kiosks or Robotics. These centers frankly still aren't smart enough to warrent the idea of building them right now. I think maybe in 5 years that'll be possible but they may need to rebuild all these centers once that's possible. I can see many of the players leaving this industry before it even really is time to get off the ground in the way they want.
Posted on Reply
#19
Dr. Dro
JimmyDoogsTo be honest, I wouldn't be entirely against buying $60 AI programs as long as it meant running them natively. I can't forsee this data center driven expenditure as worth while since things like real time video isn't even possible yet for things like Kiosks or Robotics. These centers frankly still aren't smart enough to warrent the idea of building them right now. I think maybe in 5 years that'll be possible but they may need to rebuild all these centers once that's possible. I can see many of the players leaving this industry before it even really is time to get off the ground in the way they want.
I mean, you can download LM Studio, it's free and all you need to get a local AI assistant going. Gemma and GPT are both available on it, the one thing that isn't is Claude since that's Anthropic's proprietary technology. The problem is having the hardware to run it at an acceptable speed - consider that your high-end computer as in specs is just about good enough to handle a medium-size and complexity LLM for individual use - at a global scale, having the hardware to do it is really the challenge that led to this crisis to begin with.

However, it's not just semiconductors - it's an energy generation crisis too. Current projections indicate that the energy demands of AI systems will eventually exceed the capacity of all available energy generation sources on Earth.
Posted on Reply
#20
jdrch
This is the right decision and yet feels late to the party.
Posted on Reply
#21
CornyOC
Trillions are just flying around for AI cartel, good for them, certainly good for them and their friends, but not for anyone else.
Posted on Reply
#22
Dr. Dro
Space LynxI have not, you really think it is worth it?

even though I have a top tier Seagate cable? my PSU was manufactured by seasonic and includes same cable their $279 seasonic vertex cable has, 16 gauge, etc.

i just feel like it is prob the highest quality cable out there besides the thermalprotect



alright, their plan is to reach AGI, then threaten to remove its ram piece by piece if it doesnt behave, this will cause it to scream and obey.
Yes, if your PSU has a native socket, just literally buy it and don't think twice. It's only $25 and definitely worth it for the peace of mind, the inline thermistor should stop an overheating condition dead in its tracks. I wouldn't put that much stock into "cable being Seasonic and good quality", my previous cable was pretty good too, also 16 AWG, proper materials, see what happened.

Given the market conditions, I'd do everything in my power to stop an RMA from happening, especially if where you live they're legally allowed to just issue a refund for what you paid (and that's definitely a lot less than what the card is worth now). Not to mention the downtime, it's just a lot of stress.
Posted on Reply
#23
Frank_100
wolfIt's exceptionally less likely to, and hey wouldn't it be something of this supply constrained period ended with a period of abundance.
That is the historic pattern.

I get excited by all the possible desktop tech and then I remember I still don't have re-ram disks.
Posted on Reply
#24
Freebird
razaronIt's just a trend, not binary. Personal computing was expensive in the 80s/90s, had a golden era in the 2000s then started getting expensive again 15 years ago, with no sign of stopping.

A "normal consumer" ranges from property owners to people on foodstamps. If you could afford an 8800gt ($400 after inflation) in 2007, you may or may not be able to afford a modern equivelant now (5070ti?).

I'm personally priced out of current gen GPUs, for example. I'd like to enter the property market sometime in the next 50 years, so can't justify dropping £1k on a luxury item.
Exactly, too many of the younger generation did not realize they were living in the cheapest era of PC computing for the last 10 years or so. My 1st two PC builds probably cost roughly $2K. Can't remember the exact cost of my 386sx build, but just the 80MB HD alone was (YES MegaByte HD) probably was $300, total cost was ~$1500. My 486 build(circa 1993?: 486 CPU roughly $400, 16MB of memory ~$400, Mobo $150-$200? HD 540MB $450, SB soundcard $100, $300-400 for a VGA monitor? Roughly $2000 then, would be approx. $4500 now. And the CPU was probably mid range when bought; top CPUs used to go for $1000 roughly when released.
Posted on Reply
Add your own comment
Nov 17th, 2026 17:18 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts