SK hynix CEO Kwak Noh-Jung is basically pulling the emergency brake on anyone banking on a quick end to the memory supply crunch. In a Reuters interview, he forecasts that next year will be the worst year in the industry’s history from a supply perspective. He also predicts the imbalance does not magically resolve in 2028, the date many in the market had started mentally circling, when SK hynix, Micron, and Samsung were expected to complete major capacity expansion builds.

Instead, Kwak says customer demand keeps climbing while capacity still has limitations. His forecast goes even further than a 2028 turnaround: “We still forecast that customer demand will remain higher than our supply capacity even beyond 2030.” The industry got used to thinking in milestones. Now the milestone is a moving target that likely lasts well past the date investors already had on their spreadsheets.

So why does that matter beyond headlines? Because memory is not just another commodity cycle. It is the storage and bandwidth glue for AI workloads, PCs, and data centers, and supply timing can decide whether companies sell into scarcity pricing or get stuck with underutilized capacity. Kwak’s framing implies a longer period where buyers keep reaching for chips faster than factories can produce them. Even if new builds are completed, the ramp, the effective availability, and demand growth can stay ahead.

SK hynix is also backing that thesis with capital. The company began trading on Nasdaq last Friday, launching a US megalisting of shares that was seven times oversubscribed. SK hynix said it wants to use the resulting capital for at least two more construction projects in South Korea: a fabrication hub in Yongin and an advanced packaging facility in Cheongju. Kwak added that the company is considering other global sites for future wafer fabrication investment, including the US, Japan, and Southeast Asia, and said, “Nothing has been decided yet. We are evaluating which location can provide the greatest business advantage.” In other words, the company is not just promising capacity. It is funding the geographic flexibility executives often need when supply chains, incentives, and customer concentration shift.

But the market is not buying certainty with the same speed executives wish it did. Despite that supply optimism, SK hynix shares were up 13.3% at $168.85 on the Nasdaq Friday afternoon. Then, in Seoul on Monday, the stock plunged by 15%. The US-listed shares also dipped by 9.2% to $152.50 by Monday morning.

What explains the whiplash? Analysts cited by Reuters point to the risk that heavy capacity additions and AI-related spending could swing the cycle toward oversupply. South Korea has recently made multi-billion dollar investments in the AI chip industry, and that aligns with other capacity-expanding builds that the market had been tracking for 2027 and 2028. Jing Jie Yu, an equity analyst at Morningstar, said: “Our base case here is the fresh capacity in 2027 and 2028 coming up in earnest will improve supply dynamics, thereby leading to price erosion.” Another Morningstar perspective also injects macro pressure: Lorraine Tan said, “Despite accelerating artificial intelligence adoption, monetisation remains uncertain and profitability for key players, such as OpenAI, appears to be under pressure.” She added that funding is shifting toward debt or equity, raising concerns about whether current spending levels are sustainable.

Put together, the worry is second-order and uncomfortable. Yes, memory demand is being pulled in by AI adoption. But if the AI spend does not convert into durable returns, or if data center build-out slows, then the industry could overshoot on capacity. In that scenario, even a forecast that demand stays higher than supply through 2030 would not protect pricing if buyers start renegotiating, reducing orders, or changing mix. For executives at memory makers and their customer ecosystems, it means planning cannot just be about “more supply eventually.” It has to be about demand quality, pacing, and how quickly pricing power could change if expectations for AI monetization wobble.

The strategic stakes are straightforward: if SK hynix expects the supply crunch to last beyond 2030, investors and boards need to pressure-test capex timelines, capacity ramp assumptions, and financing costs against multiple demand paths. Meanwhile, competitors like Micron and Samsung are effectively forced into the same conversation, because the entire industry’s build schedules are interlocked. Kwak’s message is not that the crisis ends in 2028. It is that the crisis might never fit neatly into anyone’s comforting forecast window.