Korean Memory Giant Posts Record Quarter on AI Demand, Ships HBM4 and Tightens the $500 Billion Nvidia Partnership
South Korea’s second-largest memory chip manufacturer posted record second-quarter earnings on July 28, with operating profit surging to 60.54 trillion won (about 41.6 billion U.S. dollars) and revenue jumping 257% year over year to 79.32 trillion won. The result, which beat internal expectations but missed the most bullish analyst consensus, cements the company’s position as the second-largest memory manufacturer in the world and as one of the most direct plays on the AI infrastructure build-out outside of Nvidia itself.
Net profit soared more than twelvefold year over year to 93.92 trillion won, supported by higher prices for both DRAM and NAND flash products, which together account for the bulk of the company’s shipments. Operating margin hit a record 76%, a level that even Apple’s hardware segment has approached only in its best quarters. The numbers underscore how AI infrastructure spending has changed the economics of commodity memory: a segment that historically traded as a cyclical commodity has, for two quarters running, looked more like a one-customer supplier.
HBM Becomes the Defining Product Line
High-bandwidth memory, the family of stacked DRAM chips that sits next to AI accelerators like Nvidia’s H100, H200 and B100 GPUs, is the single biggest driver of the result. The company confirmed it started shipping its next-generation HBM4 products during the second quarter and plans to significantly expand production in the second half of the year. In addition, HBM4E samples have already been supplied to a major customer, signaling continued progress in advanced AI memory technologies and a roughly one-year cadence between generations.
That cadence matters because Nvidia’s roadmap is now publicly anchored to the HBM4E and HBM5 generations for its 2027 and 2028 platforms. Each new generation of HBM roughly doubles effective bandwidth per accelerator and is required to keep GPUs fed with model weights at training-time throughput. With HBM4 shipping and HBM4E sampling, the Korean memory giant has effectively kept pace with the accelerator roadmap. Any slip in HBM4 yields would have shown up first in the gross margin line this quarter, and it did not.
Demand Forecast Through Year-End
Looking ahead, management forecasts DRAM bit demand to grow in the mid-20% range during 2026, with NAND demand projected to increase in the high-teens. For the third quarter, the company expects DRAM bit shipments to rise about 10% from the previous quarter and NAND shipments to increase by a low-single-digit percentage. That is consistent with continued tightness in high-bandwidth memory in particular, and with steady-state growth in traditional server DRAM and enterprise solid-state drives.
Management framed the environment as structural, not cyclical. The world’s second-largest memory chip manufacturer said the rapid adoption of AI across industries continues to generate structural demand for HBM, server DRAM, and enterprise SSDs, with growth expected to accelerate further as supply constraints gradually ease and major cloud service providers continue expanding investments in AI infrastructure. The “gradually ease” framing is the caveat: HBM4 production yields are still climbing, and the bottleneck has moved from assembly to advanced packaging.
The Wider AI Infrastructure Stack
The memory giant remains a critical supplier to Nvidia, whose AI accelerators rely heavily on HBM chips. The two companies recently strengthened their partnership through a long-term agreement to co-develop next-generation AI memory as part of a broader AI infrastructure initiative valued at more than $500 billion. That initiative includes NAND and enterprise SSD co-design for AI training clusters, advanced packaging work in South Korea and the United States, and direct integration between the memory giant’s stacking technology and Nvidia’s accelerator memory controllers.
The deal also signals tighter coupling between Nvidia and its memory supplier of choice, a posture that has implications for the wider memory market. Samsung Electronics, the world’s largest memory manufacturer, has been the secondary HBM supplier for Nvidia and has prioritized HBM3E while it works on HBM4 qualification. Micron has shipped limited HBM3E volumes but has yet to receive major HBM4 orders. The result is a memory market for AI accelerators that increasingly routes through a single Korean supplier for the highest-bandwidth tiers, with competitors positioned for the second-source role rather than the lead.
Market Reaction and Forward Risks
Despite the strong financial performance, investors continue monitoring whether the AI-driven semiconductor rally can be sustained. Shares of the Korean memory giant have experienced heightened volatility following the company’s recent Nasdaq listing, reflecting broader market uncertainty surrounding AI-related technology stocks. The stock gave back much of its pre-earnings gain in the 24 hours after the report as the headline profit number fell just short of the most aggressive analyst expectations, even though revenue and forward guidance were stronger than feared.
Macro risks remain significant. Chinese DRAM and NAND competitors are ramping new fab capacity, and a sustained slowdown in the broader AI infrastructure build-out would expose any over-investment in HBM. Currency volatility between the won and the dollar is another swing factor, as the company reports in won but generates roughly 90% of revenue in U.S. dollars or dollar-priced contracts. None of these risks dimmed the second-quarter result, but each will matter for the full-year 2026 story.
Why This Quarter Matters Beyond the Headline
For the wider semiconductor industry, the company’s quarter is the cleanest read on the AI infrastructure cycle available. Nvidia publishes its results quarterly but bundles AI accelerator sales with gaming, networking, and automotive. The memory giant is a near-pure-play on the AI infrastructure stack, with HBM attached to roughly two-thirds of all AI accelerators sold in the industry. When this company reports 76% operating margins and 257% revenue growth, the rest of the AI infrastructure supply chain has a high-confidence read on what the next quarter’s demand will look like.
The third-quarter guidance, with DRAM bit shipments up about 10% sequentially, also functions as a leading indicator for HBM prices, server DRAM contract pricing, and the volume envelope for the next generation of Nvidia, AMD, and custom-ASIC accelerators. That is why a memory report is being read across semiconductors this week as the most important earnings print of the cycle so far.