Since October, SK Hynix’s headquarters in Icheon, near Seoul, has seen a steady stream of visitors from PC makers and electronics suppliers arriving for urgent meetings with the South Korean memory chipmaker’s executives. Their top concern: a looming shortage of memory chips that could start to disrupt product launches as soon as next year.
“We are receiving requests for memory semiconductor supplies from many companies, and we are deeply concerned about how to handle all of them,” Chey Tae-won, chairman of the SK Group, said at a forum in Seoul last month.
OpenAI is among the companies seeking large supplies, Chey added, pointing to a deal signed last month to supply 900,000 wafers of high-bandwidth memory (HBM) per month for the US company’s Stargate infrastructure project.
Similar scenes unfolded at rival Samsung Electronics, the world’s largest memory chipmaker. Clients from around the world have come knocking in hopes of securing enough chips to ensure their product launches for 2026 remain on track.
“We continue to see strong demand in the memory market for both DRAM and NAND, driven by server applications,” said Kim Jae-june, executive vice president in charge of the memory business, during an earnings conference call in October, referring to dynamic random-access memory and NAND flash chips.
“Also for mobile and PC applications, with the industry providing priority supply to address artificial intelligence server demand, there are growing concerns of supply shortages and we have seen a rise in market prices,” he added.
Surging demand for AI data centers and related infrastructure has sparked a rush to snap up key memory components, namely DRAM and NAND flash, as well as other specialized types, such as NOR flash. Nearly all major memory manufacturers are already operating at or near full capacity, and production slots for 2026 are almost “sold out,” several executives told Nikkei Asia.
“Currently, small-to-medium players need to race against all the deep-pocketed internet and tech giants for memory supplies,” said a supplier executive with knowledge of the situation. “Most major memory chipmakers from Samsung and SK Hynix to Kioxia and Micron are prioritizing supplies for all the AI computing- and data center-related demand. Those placing orders from other segments will be a bit panicked about whether they can secure enough supplies,” the executive added.
“It is a bit like during COVID, in that even if you have the money, you can’t get the supplies,” said an executive with a Japanese component supplier. “SK Hynix, Samsung, Micron … their capacity for now and 2026 are fully booked already.”
The sudden supply constraints mark a sharp reversal from the past two years, when many memory chipmakers struggled through a severe downturn as consumer electronics companies and automakers worked through the large inventories they had built up. AI data centers require not only powerful processors but also significantly more memory than conventional ones. As a result, the sudden surge in AI infrastructure building has led to a rebound in memory demand far earlier than many had expected, and the resulting shortages are unlikely to be resolved anytime soon.
Major cloud service providers are expected to increase their spending on AI data center servers from USD 285 billion in 2024 to USD 468 billion in 2025 and USD 621 billion in 2026, according to estimates by Morgan Stanley.
“The order numbers we got from Nvidia are about double from 2025, and those from AMD are up 1.5 times by volume … Not to mention Amazon and Google’s own numbers are all up from this year,” an executive at one electronics component supplier told Nikkei Asia. “If the numbers are real, not only memory products face constraints, other types of accompanying components, such as printed circuit boards, can also face a shortage.”
Nvidia and AMD are two companies leading the push to build AI computing capacity, providing chips for companies such as Google, Amazon, Meta, and Microsoft. These latter players also develop their own AI chips and need memory products that meet their specific in-house requirements.
In the DRAM space, growth is mainly being driven by demand for HBM, which stacks multiple layers of DRAM to deliver faster connections, higher transfer rates and lower latency. DRAM production is dominated by Samsung, SK Hynix and Micron, with China’s ChangXin Memory Technologies emerging as a contender, while some smaller players, such as Nanya Technology and Winbond, are also looking to carve out a piece of the market.
An industry executive said the key reason behind the supply crunch is that memory chipmakers have prioritized making HBM chips for AI servers.
To meet the ferocious demand for AI servers, many major global suppliers are phasing out older-generation DDR4 products in favor of DDR5, even though many consumer and other electronic devices still require the less advanced grade of DRAM.
This combination of constrained output and continued demand has sent prices skyrocketing. The spot price of 16-gigabyte DDR4 DRAM, for example, has surged 840% year-on-year to about USD 30.3, making it more expensive than 16 GB DDR5, which is currently around USD 20 after rising 316% over the same period, according to DRAMeXchange and Bank of America Securities.
PC makers are the most vulnerable, as they “didn’t show any mercy when the market turned from extreme shortage to oversupply when [the Covid-19 pandemic] ended, so now it is natural for them not to get more supplies from the memory chipmakers,” the industry executive said.
Supply chain disruptions during the pandemic era, coupled with a boom in demand for personal computers and other electronics as working from home became the norm, led to severe shortages in a wide range of components. Investment in rapid capacity expansion resulted in supply gluts once more normal conditions returned.
“This time, chipmakers are very carefully increasing memory-chip production capacity, and they are focusing on very high-end memory chips for AI demand. If there is a shortage coming for [memory chips] for smartphones, PCs, or commodity [memory] chips, they raise the price,” Simon Woo, head of Korea research of BofA Securities, told Nikkei Asia.
Larger tech companies and top-tier electronics makers may have to pay more for memory chips, but they will not have to worry about their access to supplies, Woo said, while “white-box guys, small companies, could have trouble getting sufficient memory chips.”
Gokul Hariharan, co-head of Asia Pacific tech, media, and telecom equity research at JPMorgan, said that strong AI demand has pushed up overall memory consumption, even though a recovery in consumer electronics and automotive and industrial applications “is not that obvious.”
“Demand from AI is crowding out a lot of the supplies … so even supply chains that don’t have a direct linkage with AI, the pricing is starting to move up for a lot of commodity [memory products],” Hariharan said.
Constraints are also emerging in NAND flash memory, which is used in solid-state drives (SSDs) for data storage and is particularly important in inference, or the use of a trained AI model to produce content or carry out tasks.
Japan’s Kioxia Holdings, the world’s second largest NAND flash memory maker, expects global supply to remain tight amid a surge in AI data center construction, prompting the company to project record revenue and profit for the next quarter. Senior executives say the supply-demand imbalance is likely to persist through at least 2026 as hyperscalers, companies that offer cloud services on a large scale, accelerate their infrastructure investments.
Kioxia president and CEO Nobuo Hayasaka said during an earnings call that demand from AI-related data centers “continues to be strong,” leading to a sharp rise in orders for high-performance storage.
“AI data center inference demand is surging, and running inference workloads requires far more data storage than training. This is driving substantial new storage demand, which relies on NAND flash and hard drives,” said Wallace Gou, president of Silicon Motion, the world’s leading developer of NAND flash controller chips. “It’s a structural shift, with new demand emerging rapidly, and the entire industry must tackle it together.”
An executive with a memory chip distributor in Taiwan told Nikkei Asia that his company has been raising the bidding prices for DRAM, NAND flash, and SSD on a daily basis in light of the worsening supply constraints.
“The supplies are really limited even though we are a relatively big buyer from South Korean and American memory chipmakers,” the executive said. “Recently we had to introduce a daily quoting mechanism, as the prices in the morning session would be different from those in the afternoon.”
One American NAND flash memory chipmaker, for example, has raised its price more than 70% this year, the executive said. “And yet even if we pay 70% higher we still cannot get more from them, even a small quantity, due to their limited capacity.”
By late August and early September, demand for SSD among major cloud service providers had increased sharply, disrupting what had been a relatively balanced supply-demand situation, said Claire Wen, an analyst with Omdia. “At this point, we estimate that next year’s supply-demand bit gap will exceed 10%, compared with less than 2% for 2025.”
“Bit gap” refers to the difference between supply and demand as measured in bits of memory.
Those constraints will push up average NAND prices across the industry by around 15% to 20% in the fourth quarter of this year, with a further low- to mid-teens increase likely in the first half of next year, according to Wen. “Some deals’ prices could be even higher due to the shortage,” the analyst said.
“We can even foresee 2026 becoming the best year ever for all NAND makers,” said TrendForce analyst Bryan Ao. The industry’s average operating profit margin is normally around 15%, he said, “but in the fourth quarter, margins are already in the 15–20% range, and they are expected to rise further next year thanks to strong demand.”
The memory market is extremely volatile and prices can quickly plunge once supply exceeds demand. Over the past two years, memory chipmakers have struggled with the slow demand for memory apart from AI applications. Taiwan’s Nanya Technology, for example, posted net losses for more than ten consecutive quarters, while even major producers like Micron and Samsung suffered periods of profit decline.
That makes the current demand surge a welcome headache for many. Micron’s share price has jumped more than 150% over the past six months, while Samsung’s has risen more than 70% in the same period. Winbond returned to profitability in July-September after four quarters of losses and saw its utilization rate reach 100%, with president Chen Pei-ming saying the company expects a healthy outlook through 2026.
“AI infrastructure spending may eventually hit a point of overinvestment, but we are not there yet. This cycle still has room to run,” Chen said.
According to TrendForce, average price movements for conventional DRAM turned positive in the fourth quarter of 2024 and have continued climbing since. NAND flash prices, meanwhile, only began to recover from their downturn in the second quarter of 2025, but the upward trend is now accelerating.
But it is not all sunshine ahead. One major worry is that the boom, while good for memory makers, could push costs up to painful levels for device manufacturers and consumer electronics brands.
China’s top chipmaker, Semiconductor Manufacturing International Corporation (SMIC), has voiced concerns. Co-CEO Zhao Haijun said the impact is already visible among smartphone makers, who are increasingly worried about securing enough memory chips. “If smartphone makers can get the processors and other components but not enough memory, they still can’t ship their products,” he said.
“From another perspective, if memory prices rise too sharply, it’s unclear whether these smartphone makers can absorb the higher costs, and whether spending on other components will then be squeezed,” Zhao added.
Johnson Deng, co-CEO of Pegatron, a supplier to Apple, Microsoft, Google, and Asus, said DRAM supplies are already “hand-to-mouth.” “Our forecast suggests that our computing business this quarter and next year should have single-digit percentage growth, but memory chip supplies are the biggest variable,” Deng told an investors conference.
Asus co-CEO Samson Hu told investors that the company has around four months of memory chip inventories, therefore the impact for the current quarter is limited. “However, we would adjust product prices under appropriate conditions” to factor in such variables as changes in costs, the situation at retail and sales channel partners and end market demand, he said.
Ellie Wang, an analyst at TrendForce, is also of the view that the biggest concern is whether downstream device makers can absorb sharply rising component costs.
“What worries us more is that if memory prices climb too aggressively, consumer electronics makers will eventually have to raise their product prices to reflect the higher costs. That could hurt end-market demand.”
Memory makers themselves, meanwhile, are grappling with another big question: how long will this supercycle last?
Demand is picking up, prices are clearly rising and overall industry conditions are improving, said Miin Wu, founding chairman and CEO of Macronix International, a memory chipmaker and key supplier to Nintendo as well as clients in the automotive, space, and consumer electronics segments.
“However, what remains uncertain is how long this momentum will last and how the trend will evolve,” Wu said. “Whenever supply gets tight, some customers start placing orders with multiple suppliers to secure supplies, which can lead to double booking somewhere in the supply chain.”
This article first appeared on Nikkei Asia. It has been republished here as part of 36Kr’s ongoing partnership with Nikkei.