The central finding nevertheless holds up: the AI-memory cycle is producing a sharply uneven recovery in South Korea’s semiconductor industry. Samsung and SK hynix are converting constrained supplies of HBM, server DRAM, and enterprise SSDs into margins rarely seen in hardware manufacturing. Meanwhile, many Korean materials, parts, and equipment suppliers are gaining orders without receiving a proportional increase in pricing power or profitability.
Samsung confirmed on July 30 that its Device Solutions division generated 127.5 trillion won in second-quarter revenue and 89.2 trillion won in operating profit, an operating margin of roughly 70%. SK hynix, which reported a day earlier, recorded 79.3187 trillion won in revenue and 60.5426 trillion won in operating profit, a 76% margin and a company record. Both companies attributed the result to higher prices and a richer sales mix centered on AI-server memory.
For Windows PC buyers and enterprise IT teams, this is not distant balance-sheet news. Samsung explicitly said it expects continued supply constraints even after accounting for softer demand in mobile devices and PCs. The scarce capacity is being directed toward HBM4, server DDR5, enterprise SSDs, and other products with higher margins than mainstream client memory. That does not establish an immediate shortage of laptop RAM or consumer SSDs, but it does mean ordinary PC components are competing for wafer capacity, packaging capacity, and investment attention against far more profitable AI infrastructure products.
The headline numbers needed correction
The mismatch between the CHOSUNBIZ figures and the companies’ own earnings releases is straightforward. Samsung’s 89.2 trillion won Device Solutions profit appears as 8.92 trillion won in the report; SK hynix’s 60.5426 trillion won figure appears as 6.05426 trillion won. In both cases, the operating-margin percentages quoted in the article — 70% and 76% — align with the larger figures, not the smaller ones.
The report also says the Ministry of Data and Statistics and Korea Customs Service released their second-quarter trade statistics on August 20. The Korea Customs Service publication is dated August 11. That is a less consequential error, but it matters because the trade data are used to support the wider claim that the chip recovery is concentrating export gains among the country’s largest companies.
The corrected earnings data make the asymmetry more severe than the original article conveyed. Samsung’s semiconductor division alone produced operating income well above the annual revenue of many global component companies, while SK hynix’s quarterly operating profit exceeded its entire revenue base from only a few years ago. This is an AI-memory pricing event on an unusual scale, rather than a conventional cyclical rebound.
AP independently reported the record profits at both Korean memory leaders, while Samsung and SK hynix’s releases give the product-level explanation: high-performance memory for AI servers, including HBM and server DRAM, is selling into a market where demand remains ahead of available supply. SK hynix also said it had completed long-term supply agreements with roughly 10 key customers. Those contracts can give a memory maker better visibility, but they also make supply allocation less flexible for buyers outside the largest AI deployments.
Korea’s export data show concentration, but not a supplier profit ledger
The broader trade picture is real. The Korea Customs Service and Ministry of Data and Statistics reported that South Korea’s second-quarter exports rose 57.3% year over year to $275.5 billion. Exports by large enterprises climbed 81.8% to $206 billion, while exports by small and medium-sized companies increased 13.1% to $32.7 billion. The top 10 exporters accounted for 55.3% of total exports, up from 38.3% a year earlier.
The Korea Times, reporting the government data through Yonhap, also noted that electronics exports more than doubled to $160 billion and represented 58% of total exports. Semiconductors were plainly the decisive engine.
But that dataset has an important limitation that gets lost in the K-shaped polarization framing. It is trade data organized by company characteristics, not a semiconductor supply-chain income statement. It does not show that Samsung and SK hynix directly caused every smaller supplier’s margin pressure, nor does it isolate materials suppliers from unrelated exporters. The Customs Service also cautions that its company-characteristic classifications use data that are two years old, and that recent figures remain preliminary.
What the numbers do establish is narrower and still significant: the export boom is much more concentrated among large corporations than it was a year ago. They do not, by themselves, prove how profits are divided between a memory manufacturer and an individual photoresist, quartz, chemical, deposition, test, or packaging supplier.
That distinction matters for IT readers trying to interpret the cycle. A national export boom can coexist with a difficult operating environment for a vendor whose costs rise faster than contract prices. Revenue growth and order backlogs are not interchangeable with margin expansion.
Why HBM toolmakers are the exception
CHOSUNBIZ’s strongest evidence is the contrast among the suppliers themselves. Wonik IPS reported second-quarter revenue of 216.5 billion won and operating profit of 18.4 billion won, down 11% and 50% year over year, respectively. Yet its first-quarter order backlog had reached 400.4 billion won, according to SK Securities reporting carried separately by CHOSUNBIZ.
That is a useful warning against reading backlog as immediate profit. Semiconductor-equipment revenue is recognized around delivery, acceptance, and installation milestones; large orders can therefore arrive well before a vendor records sales. Wonik IPS attributed the uneven quarter to regional timing in revenue recognition, and its analyst forecast points to stronger third-quarter recognition. The company’s weak reported profit is a snapshot, not proof that its booked work has disappeared.
Still, the gap between orders and earnings is not only an accounting-timing issue. Equipment vendors must fund engineering, inventory, service capacity, and customer qualification work before all revenue arrives. Materials makers selling consumables face a different constraint: their products may be essential, but frequent volume purchases and qualification requirements can lock them into long-term contracts and make price increases slow to negotiate.
HANMI Semiconductor illustrates the opposite case. Maeil Business Newspaper reported that the company posted record second-quarter sales of 251.1 billion won and operating profit of 130.3 billion won, for a 51.9% operating margin. Its thermal-compression bonders are used in the HBM assembly process, where vertically stacked memory dies are joined under heat and pressure.
That tool occupies a genuine production bottleneck. HBM is not simply higher-priced DRAM; it depends on advanced packaging steps with demanding yields and tight process control. A supplier whose tool is hard to replace during a capacity buildout can command materially stronger pricing than a supplier offering a qualified but more substitutable chemical or consumable component.
The lesson is not that every equipment company is winning. It is that scarcity inside the supply chain is determining who captures the AI premium. HBM-related bottleneck tools can participate directly in the economics of constrained AI memory. Suppliers selling products that are easier to dual-source, or whose prices are set under older supply agreements, may see more units shipped with little improvement in operating leverage.
The PC market is outside the profit center, not outside the impact
Samsung’s own earnings outlook places PCs on the lower-priority side of its memory portfolio. The company expects server demand to remain robust and says supply constraints are likely to persist even with partial moderation in mobile and PC demand. Its stated plan is to concentrate on HBM4, DDR5, SOCAMM2, and other high-value products intended for next-generation AI platforms.
That is rational for the manufacturer and potentially awkward for device makers. A laptop vendor or enterprise desktop-refresh program does not buy HBM, but it competes indirectly for DRAM wafer output, NAND output, controller ecosystems, testing, and packaging resources. When a memory maker can earn dramatically more on AI-server products, maintaining low-margin client-memory supply becomes a strategic allocation decision rather than an automatic result of demand.
IT procurement teams should therefore avoid assuming that declining consumer-PC demand will necessarily translate into cheaper memory and storage. The relevant question is whether client-grade DDR5 and NAND supply grows faster than demand after manufacturers fulfill their more profitable server and HBM commitments. Samsung has signaled that overall memory supply will remain constrained; SK hynix has signaled that it is relying on long-term customer agreements while expanding capacity with discipline.
Neither company announced a specific reduction in client-memory supply, and no evidence in the current disclosures supports predicting a near-term price spike for Windows PCs. What the record does support is a more cautious budgeting posture for server upgrades, AI workstation deployments, and fleet refreshes that require higher RAM and SSD configurations.
A boom with fewer automatic spillovers
The Korean chip sector is enjoying a record period, but the evidence does not support treating all semiconductor companies as beneficiaries in equal measure. Samsung and SK hynix have the products, scale, customer contracts, and market position to convert AI demand into exceptional margins. HANMI Semiconductor appears to be benefiting because its TC bonders are tied directly to an HBM manufacturing bottleneck. Wonik IPS shows how a healthy order book can still coexist with falling quarterly profit.
The practical consequence is that the AI memory boom should be evaluated product by product and process step by process step. For the Windows and enterprise market, the companies accumulating the largest profits are also the ones deciding where scarce memory capacity goes — and their second-quarter results show exactly why AI servers will remain at the front of that queue.