There are surely no shortage of readers who, while shopping for a new smartphone or PC, were startled by a price tag higher than expected. DRAM and NAND prices have surged over the past several quarters, and that surge is now spilling over into retail prices. Against this backdrop, Samsung and SK hynix both posted record earnings—yet Samsung's stock actually fell by about 10%. Behind this paradox, in which strong earnings are met with selling, lies a long-term target that memory makers have begun to talk about: "2034." The very structure of the industry—where more than three years pass between the groundbreaking of a new fab and the start of mass production—is pushing this AI boom onto a different timeline than the two previous boom-and-bust cycles.
Why Samsung's Stock Fell Even as Operating Profit Rose 19x
Lining up the earnings of Samsung, SK hynix, and Micron reveals that, while the magnitude of growth differs across all three companies, they share a common trait: the stock market's reaction was more extraordinary than the numbers themselves. In its guidance for Q2 2026, Samsung projected revenue of KRW170–172 trillion (approximately $111.9 billion) and operating profit of KRW89.3–89.5 trillion (approximately $58.4 billion). That operating profit figure represents an extraordinary roughly 19-fold increase year over year. And yet, after the announcement, Samsung's stock fell by about 10%. It's a seemingly contradictory reaction in which record earnings become a trigger for selling.
SK hynix likewise posted extraordinary results. Its Q2 2026 revenue came to KRW82.89 trillion, up 273% year over year—roughly a 3.7-fold increase in absolute terms. Operating profit reached KRW63.45 trillion, up 589% year over year. Micron, too, recorded revenue of $41.46 billion in its fiscal Q3 2026 (announced June 24), marking its fifth consecutive quarter of record highs. Compared to $9.30 billion in the same period a year earlier, that's roughly a 4.5-fold increase, and gross margin also hit a record 84.9%.
Samsung attributed its performance to "technological leadership in the memory market and rising average selling prices." In other words, rising average selling prices—premiums driven by scarcity—pushed up profits. What the stock market appears to be wary of is not the profit level itself, but the timeline of how long this scarcity will persist.
The Mechanism Behind Why New Fabs Can't Catch Up Even in 3 Years
Bringing a new memory semiconductor fab from groundbreaking to mass production takes more than three years. Building the cleanroom alone takes around a year, and on top of that, the delivery and installation of manufacturing equipment—including lithography tools—along with the process of raising yields to commercial levels, each add up to anywhere from several months to a full year. The plan SK hynix announced in June 2026 calls for tripling wafer-based production capacity by 2034. This announcement came at almost the same time as the current wave of earnings reports, effectively an admission by the company itself that roughly eight years will pass between the moment the starting gun for capacity expansion is fired and the point at which its full effect materializes.
Another reason supply struggles to keep pace is the conversion of production lines toward HBM (High Bandwidth Memory, a high-performance standard directly tied to AI-focused GPUs). HBM requires an additional process of stacking multiple silicon dies and connecting them via through-silicon vias, meaning ordinary DRAM lines cannot simply be repurposed as-is. All three companies—SK hynix, Micron, and Samsung—have nearly sold out their 2026 HBM production allocations, and SK hynix is reported to be supplying roughly two-thirds of the HBM4 used in NVIDIA's Rubin platform. The more existing lines are diverted to HBM, the tighter the supply of general-purpose DRAM and NAND used in PCs and smartphones becomes.
Taking this structure into account, IDC forecasts that memory shortages will ease at the earliest by 2028. And even then, the view is that price levels will never return to pre-2025 levels. This memory price surge is not a one-off spike in demand. The very length of time it takes for the current rush of new fab construction to translate into a genuine increase in supply is what makes this phenomenon structural.
Verifying the Difference from 2019 and 2022 in Numbers
The memory industry is well known as a commodity market that repeats cycles of boom and bust. Following the boom of 2017–2018, prices crashed in 2019; after peaking in 2021, prices collapsed again in 2022. In 2019, a backlash against hyperscalers' overordering caused DRAM prices to fall by roughly 60% over four quarters. In the latter half of 2022, Micron's stock price nearly halved, falling from about $98 to about $49. Both cycles followed the same pattern: prices collapsed the moment increased production caught up with demand.
In both the 2019 and 2022 cycles, memory makers kept insisting—right up until the moment they entered a capacity-expansion phase—that "this time, demand is structural and won't easily collapse." And yet, in both cases, prices collapsed once increased production caught up with demand. There is no guarantee that the current cycle won't meet the same fate. That said, there is a clear difference in scale. The global memory market as of Q2 2026 is estimated at approximately $254.4 billion, more than 2.5 times the roughly $99.0 billion size of the super-cycle period of 2017–2018. The fact that HBM production allocations for 2026 are nearly sold out also represents a depth of demand that didn't exist back in 2019.
Still, there are signs worth noting. TrendForce forecasts that DRAM contract prices will rise 13–18% quarter-over-quarter in Q3 2026, a clear deceleration from the roughly 60% rise recorded in Q2. NAND is likewise expected to rise 10–15%, with the rate of growth itself beginning to shrink. Looking closely at capacity-expansion plans and supply contract details, it remains undisclosed whether order-cancellation clauses exist in case AI demand slows.
What sets this cycle apart from the two previous ones is the sheer scale of the market. But that is no guarantee that a downturn won't come. The more the numbers are laid side by side, the more that difference stands out.
The Risk of Oversupply Lurking Behind Each Company's Capacity-Expansion Investments
As if betting that the supply shortage will persist for the long haul, Micron announced on July 9, 2026 a strategic investment of up to $3 billion aimed at strengthening the U.S. semiconductor supply chain. Of that, $500 million is a loan directed at GlobalWafers America's (Texas) 300mm wafer plant, bundled together with a 10-year supply agreement. Locking in a 10-year contract even for upstream raw-material wafer procurement reflects an investment premised on the assumption that shortages in the upstream raw-material supply chain will also persist for years. The South Korean government likewise announced on June 29, 2026 that President Lee Jae Myung, together with Samsung and SK hynix, unveiled an AI chip investment plan worth a combined $576 billion, of which KRW800 trillion (approximately $517.9 billion) to be invested by Samsung and SK hynix will go toward building four new fabs in southwestern South Korea. The fact that each company is piling on multi-year investments is itself a reflection of an industry-wide consensus that the supply shortage will be prolonged.
But this bet only holds if AI demand keeps growing as projected in these investment plans. According to IDC, global PC shipments in Q2 2026 fell 4.9% year over year to 68.2 million units, marking the first year-over-year decline in two years. According to Omdia's data, as of Q1 2026, memory accounted for roughly 60% of the component cost of smartphones priced at $400 or below, and shipments in that price segment are projected to fall 22% year over year for full-year 2026. By contrast, shipments of smartphones priced above $400 are projected to rise 5.7% year over year over the same period—memory price hikes are increasingly splitting the fortunes of the low-end and high-end segments.
In other words, the production capacity that memory makers are adding over multi-year timelines is designed on the assumption that AI server demand will keep growing, while behind the scenes, non-AI demand has already begun to shrink. If the growth in AI server demand fails to fully absorb the decline in non-AI demand by around 2028, when the new fabs come online, the added capacity will have nowhere to go. The financial staying power of AI startups, whose margins are being squeezed by rising costs, is one variable that will determine how much of this decline can be absorbed.
Ripples Reaching Japan Too: Kioxia and Price Pass-Through
Driven by tailwinds from NAND demand for AI data centers, Kioxia Holdings is projected to surpass ¥2 trillion in full-year revenue for the fiscal year ending March 2026 for the first time. Net profit is expected to reach a level roughly 17 times that of the previous fiscal year—Japanese semiconductor companies, too, are not standing outside this boom.
By contrast, SK hynix and Samsung have repeatedly shelved plans to build new fabs within Japan. As a result, Kioxia continues to hold what amounts to a near-monopoly position domestically in the NAND segment. That said, this advantage offers no protective barrier against price increases for domestic consumers. Gartner forecasts that PC prices will rise 17% and smartphone prices 13% by the end of 2026, and since these products share the same DRAM and NAND procurement costs, products sold in the Japanese market can hardly escape this wave either.
Gartner also projects that global PC shipments will fall 10.4% year over year and smartphone shipments will fall 8.4%, suggesting that price increases are starting to slow down the very replacement cycle itself. The focal point remains the same in every market, Japan included: exactly when supply will finally catch up with demand. IDC's projected easing of shortages around 2028 arrives earlier than SK hynix's stated target of full-scale capacity expansion by 2034—meaning that if the rush of new fabs coming online coincides with a slowdown in AI investment growth, the capacity-expansion investments made by companies including Kioxia will head toward payback in the form of settled prices. Conversely, if AI demand continues to grow at a pace that outstrips these investment plans, the current supply lag will simply lock in as a sustained high-price plateau.