Jeju Semiconductor Is Riding a Boom Built on Capacity It Doesn't Control

By the headline numbers, Jeju Semiconductor looks like one of the AI memory boom's quiet winners. The KOSDAQ-listed fabless memory designer reported first-quarter revenue of 180.4 billion won, up 273% from a year earlier, and operating profit of 67.1 billion won, up 1,713%. The reporting framed the surge as the AI tailwind "lifting legacy chips". That description is accurate as far as it goes — and it is the wrong lens for judging whether the stock deserves the attention it is getting.

The problem is that nearly every driver behind Jeju's explosion sits with another company. Its growth is a pass-through of price inflation in ordinary memory caused by the giants redirecting their wafers to AI products, and even the product itself is fabricated by someone else. The boom is real. Jeju's ownership of it is thin.

The earnings move is pricing, not power

The first thing to establish is why a fabless memory merchant's revenue roughly tripled in a single quarter. This is not a story of the company shipping dramatically more units or seizing share. It is a story of a market-wide shortage.

The memory industry has split into two pools. In the first, AI accelerators consume high-bandwidth memory in volumes that are crowding out everything else at Samsung, SK Hynix, and Micron. In the second sits the memory Jeju actually sells — low-power DRAM and NAND used in phones, IoT devices, and legacy systems. Because the big three keep pointing their capacity at the AI pool, the legacy pool has gone undersupplied, and prices have inflated violently. Gartner projects DRAM prices will rise 125% in 2026. Jeju rides that price, not a unit upswing.

The product mix shows the shift. In the first quarter, Jeju's DRAM sales reached 57.6 billion won, and DRAM's share of total product revenue jumped from 12.8% a year earlier to 32%. That is the two-market split working in Jeju's favor: a shortage created elsewhere is driving ASPs on the chips it sells.

Pool	What's emphasized	Where capacity goes
AI/high-bandwidth memory	HBM for accelerators	Samsung, SK Hynix, Micron direct wafers here
Legacy/low-power	LPDDR, commodity DRAM, NAND	Undersupplied, prices surging — Jeju's pool

The constraint is SK Hynix's, not Jeju's

The harder question is how much of this Jeju actually captures going forward. It is a fabless designer with roughly 134 employees and no fabs of its own. Its LPDDR4X is manufactured at SK Hynix's fabrication plants. Jeju buys the memory wafers, does the packaging and testing, and sells the finished parts under its own brand. This is a capacity-host arrangement: Jeju contributes design and distribution, and SK Hynix supplies the scarce commodity — the wafers.

That structure puts Jeju's fate in a partner's hands, and industry reporting notes that a major manufacturer allocating capacity to an external fabless partner is "highly unusual", particularly in a period of tight memory supply. The natural instinct during a shortage is to keep scarce wafers for one's own high-value products, not to hand them to a rival's brand. That SK Hynix has done so is the central anomaly underpinning Jeju's boom — and it is an anomaly SK Hynix can reverse.

Jeju's own ambitions make the dependency visible. It plans to move to LPDDR5 at SK Hynix fabs, with mass production targeted for 2030, likely at SK Hynix's Wuxi plant in China, a facility that cannot install EUV lithography and is therefore suited to legacy-node production. The company expects its LPDDR business to eventually reach annual revenue of one trillion won, against 302.2 billion won for all of 2025. That is an aspiration contingent on SK Hynix keeping a steady stream of wafers flowing to a small partner for years.

The bull case depends on a condition Jeju doesn't set

Valuation has already moved to price in sustained boom. At the end of June, Jeju carried an enterprise value of roughly $2.26 billion against trailing-twelve-month revenue of about $464 million — close to five times revenue, rich for a fabless merchant whose growth driver it does not control.

The bull case for the stock rests on two things, and neither is Jeju's to decide. First, the shortage must persist: SK Hynix expects the global memory shortage to run through 2030 with little risk of oversupply, so the industry backdrop is supportive if management's own forecast holds. Second, and more tenuous, SK Hynix must keep choosing to feed its capacity to Jeju's branded parts rather than redirecting it to its own higher-margin AI products as that demand keeps growing.

The distinction matters. In a genuine structural winner, the company controls the constraint and therefore captures the economics. Jeju does not. It is enjoying a period in which someone else's supply discipline and capacity decisions are creating pricing power for its products. For as long as SK Hynix keeps allocating wafers and the legacy pool stays tight, Jeju's numbers can keep surprising to the upside. The investment question is not whether memory stays scarce. It is whether the company that owns the scarce capacity keeps choosing to share it.