The AI memory shortage will persist through at least 2030, according to SK Group chairman Chey Tae-won, who announced the expansion plan at Computex in Taipei on June 2. Even as SK hynix doubles its memory wafer capacity, the semiconductor industry faces a structural undersupply driven by artificial intelligence workloads that shows no sign of slowing. This is not a temporary bottleneck—it is the new normal for the next five years.
Key Takeaways
- SK hynix will double DRAM wafer capacity to roughly 620,000 wafers monthly by late 2026, nearly doubling 2023 levels
- The new M15X fab in Cheongju starts production in H1 2026, ramping to 50,000 wafers monthly by Q4
- About 30% of SK hynix’s DRAM capacity is already devoted to HBM, reflecting AI’s dominance
- AI memory shortage will remain acute even as capacity expands, lasting through 2030
- Samsung plans to expand memory production by roughly 50% in 2026, competing directly with SK hynix
SK hynix’s Massive Capacity Bet on AI Memory
SK hynix is committing serious capital to address the AI memory shortage. The company announced a $15 billion investment to expand advanced memory manufacturing, with particular focus on high-bandwidth memory (HBM) and advanced DRAM. The new M15X fabrication plant in Cheongju will begin operations in the first half of 2026, starting at around 10,000 wafers per month and ramping to 50,000 wafers monthly by the fourth quarter. By late 2026, SK hynix’s total DRAM capacity could reach roughly 620,000 wafers per month—nearly double the mid-300,000 level the company logged in 2023.
This expansion is not theoretical. SK hynix is already allocating approximately 30% of its DRAM capacity this year to HBM production, a specialty memory type essential for AI accelerators and large language model inference. The shift shows where the real growth lies: not in consumer DRAM or laptop memory, but in the specialized silicon that powers data center AI workloads. The company’s willingness to dedicate a third of current output to HBM signals confidence in sustained demand from AI infrastructure spending.
Why Doubling Capacity Still Will Not Fix the AI Memory Shortage
Here is the uncomfortable truth: even a near-doubling of SK hynix’s capacity will not resolve the AI memory shortage through 2030. Analysts and original equipment manufacturers expect memory scarcity to remain tight even if SK hynix increases production sharply in 2026. The problem is demand growth outpacing supply growth by a structural margin. Every major cloud provider, chip maker, and AI infrastructure company is competing for the same constrained pool of advanced memory.
Samsung is also scaling aggressively, planning to expand memory production capacity by around 50% in 2026. Yet even with both SK hynix and Samsung ramping simultaneously, the market will remain undersupplied. This suggests that AI adoption is accelerating faster than the semiconductor industry’s ability to build fabs and fill them with equipment. The capital expenditure required to double capacity takes years to convert into actual wafer output, and by the time that output arrives, demand has already moved further ahead.
The AI-Driven Memory Market Through 2030
The AI memory shortage reflects a fundamental shift in semiconductor demand. Traditional DRAM and NAND flash markets grew predictably, following Moore’s Law and generational device refresh cycles. AI workloads shattered that pattern. Data centers are building new clusters constantly, each requiring massive amounts of HBM and specialized DRAM. Large language models consume memory in ways that consumer products never did.
What makes this different from past shortages is duration. The 2020–2021 chip shortage was a supply chain crisis—a temporary mismatch between unexpected demand and production capacity. The AI memory shortage is structural. It reflects genuine, sustained demand growth that will outlast any single expansion cycle. SK hynix’s five-year capacity plan acknowledges this reality: the company is not expecting the shortage to end in 2026 or 2027. It expects to be selling every wafer it produces through 2030 and beyond.
SK hynix vs. Samsung: A Race With No Finish Line
SK hynix and Samsung are locked in a capacity race, but neither company can win by outpacing demand alone. SK hynix’s projected DRAM input of about 600,000 wafers per month in the second half of 2026 would reach rough parity with Samsung. Yet parity is not victory—it is merely equilibrium in a market where both suppliers are perpetually undersupplied. The real competition is not between these two giants, but between them and the relentless growth of AI infrastructure spending.
Samsung’s 50% capacity expansion in 2026 shows the industry is not complacent. Both suppliers are betting billions that AI demand will justify the investment. What neither company can control is the timeline. Building a new fab takes years. Equipping it takes longer. By the time M15X reaches full production, the market may have already consumed the gains and demanded more.
What This Means for the Rest of the Industry
The AI memory shortage will shape semiconductor strategy through the end of the decade. Companies dependent on DRAM and HBM will face sustained pricing pressure and allocation constraints. Data center operators will compete fiercely for allocation from SK hynix, Samsung, and smaller suppliers like Micron. Chip designers will optimize for memory efficiency, knowing that memory bandwidth and capacity will remain bottlenecks.
For consumers, this shortage is largely invisible—it manifests as higher cloud computing costs and slower AI service rollouts, not as empty shelves in retail stores. But for enterprise and infrastructure buyers, the AI memory shortage is the defining supply chain story of the mid-2020s. Every data center expansion, every AI accelerator deployment, every large language model training run depends on securing memory that is structurally scarce.
Will the AI memory shortage ease before 2030?
Unlikely. Even with SK hynix and Samsung ramping aggressively, demand from AI infrastructure, cloud providers, and chip makers is expected to outpace supply through the end of the decade. The shortage reflects structural demand growth, not a temporary bottleneck.
How much is SK hynix spending on the capacity expansion?
SK hynix committed $15 billion to expand advanced memory manufacturing, particularly HBM and advanced DRAM production. This investment spans multiple facilities, including the new M15X fab in Cheongju.
What percentage of SK hynix’s output goes to HBM?
Approximately 30% of SK hynix’s DRAM capacity this year is already allocated to HBM, reflecting the dominance of AI workloads in the company’s product mix.
The AI memory shortage is not a crisis to be solved—it is a market condition to be managed. SK hynix’s expansion acknowledges this reality. By committing to double capacity and investing heavily in HBM production, the company is betting that AI demand will remain strong through 2030 and beyond. That bet reflects the industry consensus: artificial intelligence is not a temporary trend, but a structural shift in computing that will define semiconductor demand for years to come. Memory will remain scarce, expensive, and fiercely competed for.
Edited by the All Things Geek team.
Source: Tom's Hardware
