SK hynix did not improvise its response to the AI memory boom. According to information published by The Elec, the South Korean manufacturer had already shared with key suppliers a plan to increase its DRAM wafer production capacity to around 1 million units per month between 2030 and 2031, nearly doubling its current capacity.
This data comes just days after a highly symbolic image at Computex 2026: Jensen Huang, CEO of NVIDIA, signing a SK hynix wafer with the message “Please make more.” The scene summarized in three words the pressure felt throughout the entire memory supply chain, but SK hynix’s industrial plan predates that moment. The company had been informing partners and suppliers for weeks about an ambitious expansion plan focused on its future Yongin semiconductor complex and the M15X plant in Cheongju.
From 550,000 to 1 million wafers per month
SK hynix’s current DRAM capacity is around 550,000 wafers per month, including approximately 200,000 from the Wuxi plant in China. The company’s target is to approach 1 million wafers per month by 2030 or 2031, a significant leap for a market strained by demand for HBM, AI servers, graphics accelerators, and advanced memory.
The expansion will mainly focus on the Yongin cluster in South Korea. The first phase of the factory will consist of two main structures and six cleanrooms. According to The Elec, SK hynix plans to introduce the first equipment in February 2027, three months earlier than initially scheduled, and to add capacity gradually: 60,000 wafers per month per cleanroom every six months.
If this timeline is maintained, only the first Yongin facility would reach about 360,000 DRAM wafers per month by the first half of 2030. To this figure, the expansion of M15X in Cheongju would be added, starting operations in the second half of this year with around 40,000 wafers per month and reaching nearly 80,000 in 2027.
| Facility / Phase | Estimated Capacity | Planned Schedule | Role in the Plan |
|---|---|---|---|
| Current SK hynix DRAM capacity | 550,000 wafers/month | Current situation | Starting point, including Wuxi |
| Wuxi plant, China | 200,000 wafers/month | Operational | Part of current capacity |
| Yongin phase 1, cleanroom 1 | 60,000 wafers/month | Equipment installation starting February 2027 | Initial expansion block |
| Yongin phase 1 complete | 360,000 wafers/month | By mid-2030 | Main capacity increase |
| M15X Cheongju | Starting with 40,000 wafers/month | Second half of 2026 | Short-term capacity boost |
| Cheongju expansion | Approximately 80,000 wafers/month | 2027 | Additional contribution to the plan |
| SK hynix total goal | Close to 1 million wafers/month | 2030–2031 | Approximate doubling of DRAM capacity |
AI forces a rethinking of memory capacity
The pressure from NVIDIA on memory manufacturers is not coincidental. Each new generation of AI accelerators demands more bandwidth, greater capacity, and a much more stable HBM supply chain. SK hynix has become one of the most relevant providers of high-bandwidth memory for AI platforms, strengthening its position in a market where available capacity directly influences major chipmakers’ product launches.
Huang’s message was not just a media gesture. It reflects an industrial reality: memory has become one of the critical resources for artificial intelligence. For years, the focus was on GPUs, accelerators, and computational capacity. Now, the conversation is shifting toward the availability of HBM, advanced DRAM, packaging, wafers, lithography equipment, power supplies, and cleanroom capacity.
The SK Group had already publicly anticipated this tension. Its chairman, Chey Tae-won, stated at Computex that the group plans to double total wafer capacity within five years to meet AI-driven demand. He also warned that memory shortages could last until 2030 if supply does not grow at the pace the market requires.
Yongin’s plan aligns with this perspective. It is not just about manufacturing more generic memory but about establishing an industrial foundation capable of supporting increased demand for advanced DRAM and HBM throughout the decade. The company has decided to focus the new expansions on DRAM, while in NAND technology, the priority appears to be technological improvements, such as more layers, rather than a capacity expansion of similar scale.
An ambitious expansion, but with risks
However, the schedule is not without uncertainties. The suppliers referenced by The Elec have received messages indicating that they should prepare for a rapid increase in orders, but they also view the plan cautiously. The memory sector is highly cyclical, and SK hynix has previously reduced equipment orders after presenting more optimistic investment guidance to its partners. For equipment, materials, and component suppliers, ramping up capacity before having concrete orders can strain their financial resources.
Additionally, filling a cleanroom every six months depends not only on investment willingness but also on coordinating critical machinery, supplies, permits, electricity, gases, skilled personnel, installation, calibration, and production ramp-up. Delays in a specific piece of equipment could impact the entire schedule.
There is also the risk that demand in some segments may not grow at the same pace. While AI is strongly boosting HBM and advanced DRAM, such a large-scale expansion requires several years of visibility. If the market enters a phase of overcapacity or customers adjust inventories, the balance between investment, price, and profitability could become complicated.
Nonetheless, a key difference from previous cycles is that memory demand no longer depends solely on PCs, mobile devices, or traditional servers. AI has created a much more intensive consumption vector, where each new training and inference platform requires more memory per system. This doesn’t eliminate cyclicality but does change the size of the opportunity.
Memory as a bottleneck of the new AI infrastructure
SK hynix’s decision illustrates how the AI race is shifting across the entire supply chain. It’s not enough to design more powerful accelerators; manufacturers of memory capable of mass production, wafer suppliers, EUV equipment, advanced materials, packaging, energy, and sufficient industrial capacity are all essential.
For major AI customers, the challenge is not just purchasing chips but ensuring supply for years. If memory becomes a bottleneck, it could delay data center deployments, drive up prices, and limit the availability of new hardware generations. That’s why companies like NVIDIA are publicly pressuring their partners, and why SK hynix is accelerating plans that might have once seemed overly aggressive.
This move also reinforces South Korea’s position as a strategic hub for global memory. Yongin won’t be just a manufacturing expansion but a key component of the country’s capacity to sustain AI demand, compete with other Asian manufacturers, and maintain a strong position in advanced semiconductors.
Huang’s signature on a SK hynix wafer set the scene. The industrial plan was already underway. The important difference is that NVIDIA’s call didn’t create the expansion but confirmed to the entire market that memory will be one of the decisive fields of AI infrastructure in the coming years.
Frequently Asked Questions
What are SK hynix’s plans for its DRAM production?
SK hynix has shared a plan with suppliers to increase its DRAM wafer capacity from approximately 550,000 per month to nearly 1 million between 2030 and 2031.
What role will the Yongin factory play?
Yongin will be the main hub of expansion. The first phase will include six cleanrooms and could add around 360,000 DRAM wafers per month by the first half of 2030.
Why is this plan so important for artificial intelligence?
Because AI systems increasingly rely on advanced memory, especially HBM and high-performance DRAM. If supply doesn’t grow sufficiently, memory could become a bottleneck for new servers and accelerators.
Was SK hynix’s plan a reaction to Jensen Huang’s message?
According to The Elec, no. The company had already communicated its plan to suppliers before Huang wrote “Please make more” on a SK hynix wafer during Computex 2026.
via: thelec