SK Hynix, one of three chipmakers that dominate the global memory market, says tripling production capacity by 2034 is the fastest the industry can physically build, leaving consumers with no near term price relief.
A 2TB consumer SSD that sold for around $100 in mid-2025 now runs closer to $300 on retailer shelves, and the price of a stick of desktop DRAM has moved in the same direction. The reason is not a temporary stocking glitch. SK Hynix, the Korean chipmaker that supplies the high-bandwidth memory stacked inside Nvidia's most expensive AI accelerators, says the industry's planned response to the squeeze is a buildout that will not finish until the early 2030s.
In an interview with Nikkei Asia, SK Group Chairman Chey Tae-won said the company intends to roughly triple its wafer production capacity by 2034, a target that pulls in a timeline previously expected to stretch over twenty years (Nikkei Asia interview with Chey Tae-won). The Register's coverage of the interview framed the pledge with a sarcastic tagline, "you can wait another 8 years, right?" (The Register, Jun 12 2026), and that question is now the operative one for buyers, system builders, and the procurement teams that price cloud capacity.
What Chey is describing is the cap, not the floor. Memory demand from AI training and inference has outrun every prior planning cycle, and SK Hynix, Samsung, and Micron, the three firms that together control most of the world's DRAM and NAND production, are publicly trying to build their way out of a shortage that has already pushed consumer prices up sharply. Consumer DRAM and SSD prices have more than tripled compared with a year earlier, per The Register's prior reporting (The Register, Jun 2 2026), and the same outlet warned in December that the climb was not close to ending (The Register, Dec 20, 2025).
SK Hynix is the world's largest supplier of HBM, the high-bandwidth memory stacks that sit next to a GPU's processor die and feed it data. AI accelerators consume HBM in volume; consumer SSDs use NAND, a different kind of flash memory that the same companies make in the same factories. The tripling pledge does not change the immediate mix much: the first phase of the new fabs is scheduled to come online as early as 2027, and the most aggressive timelines still leave the bulk of the added wafers arriving between 2028 and 2034. HBM lines for AI customers will be filled first, which means the consumer SSD shelf is unlikely to see meaningful relief this cycle.
Chey was explicit about the constraint. "There is currently no way to move faster than this," he told Nikkei, a line that has come to define how the industry is now talking about capacity. Independent analyst James Sanders of TechInsights has put the physical floor on a new leading-edge memory fab at roughly three years from groundbreaking to first wafers, and has described the AI demand surge, which he dates to 2022, as having hit the DRAM industry at the worst possible time. SK Hynix itself said roughly a week before the Chey interview that it planned to double production capacity within five years; the 3x by 2034 figure layers on top of that earlier doubling pledge, and Chey separately flagged Japan as an "excellent" candidate for additional overseas fabs, citing the country's existing semiconductor supply chain.
The supply picture has other fragilities. With only three firms controlling the bulk of advanced memory production, any disruption at one of them — a fire at a fab, a labor action, a tooling delay — can ripple through global pricing. A possible strike at Samsung, SK Hynix's larger Korean rival, would arrive just as the industry is publicly admitting it cannot build its way out of the shortage on the timeline consumers want. That concentration also shapes where the new fabs will land: Korea and Japan will get the bulk of the new construction, with the United States, India, and Europe all competing for pieces of the regional fab policy agenda. The geographic bet is not just industrial; it is a hedge against the next geopolitical shock.
For buyers, the practical read is that the historical DRAM boom-bust pattern is on pause. Analysts quoted in the Register's December coverage warned that a 2027 plateau is more likely than the kind of price crash that ended the 2017 to 2018 memory supercycle. Watch the first 2027 fabs: if SK Hynix hits that first-phase target, the constraint shifts from wafer shortages to where the new capacity is pointed. If it slips, the squeeze simply gets a longer name and a later expiration date.