SK hynix reported record second-quarter results, but the headline numbers did not translate into the market response many had expected. Operating profit surged, yet still came in below market forecasts. Revenue also missed expectations, and the shares fell in after-hours trading. For investors, the quarter quickly became a case study in expectation gaps rather than headline strength.
That tension had already started to build before the release. Korea Investment & Securities, or KIS, had cut its profit estimate for SK hynix two weeks earlier, but the reaction after the results suggested sentiment was even more fragile than that adjustment implied. On the earnings call, management largely reiterated familiar talking points. The sharper signal came from Wall Street questions, which centered on demand durability, supply constraints, long-term agreements, HBM competitiveness and the pace of capacity expansion.
Revenue and profit hit fresh records, but the market wanted more
SK hynix said strong demand tied to expanding AI infrastructure and a still-tight supply environment continued through the second quarter, keeping pricing on an upward trend. DRAM and NAND both posted another round of meaningful price increases from the previous quarter, with AI-related products such as server DRAM and enterprise SSDs leading the growth.
Second-quarter revenue reached KRW 79.3 trillion, up 51% from the prior quarter and 257% from a year earlier, setting another company record after the previous quarter.
With supply capacity still constrained, the company expanded sales centered on HBM and AI server DRAM. Shipments rose by a high single-digit percentage from the prior quarter, in line with earlier guidance. Sales of server LPDDR products, including SOCAMM, increased sharply, while conventional DRAM price strength helped lift DRAM average selling prices by about 30% quarter on quarter.
On the NAND side, a low base from weaker first-quarter shipments and stronger enterprise SSD sales helped drive bit shipments up by a mid-double-digit percentage sequentially, also in line with guidance. Enterprise SSD revenue doubled from the prior quarter, and revenue from 30TB-and-above high-capacity enterprise SSDs rose by more than two times. With prices strengthening across products, DRAM and NAND both moving higher, and the cost structure improving, NAND ASP increased by a mid-50% range quarter on quarter.
Profitability reached new highs as well. Operating profit came in at KRW 60.5 trillion, up 61% from the previous quarter and 557% from a year earlier. Operating margin rose 5 percentage points sequentially to 76%, another record for the company. Depreciation and amortization totaled KRW 4 trillion in the quarter, EBITDA was KRW 64.6 trillion, and EBITDA margin stood at 81%.
Non-operating net income was KRW 62.2 trillion, including KRW 1.1 trillion in foreign-exchange gains from a rise in exchange rates and KRW 63.3 trillion in gains from disposals and valuation of investment assets. Pretax profit reached KRW 122.7 trillion, net profit came to KRW 93.9 trillion, and net margin was 118%.
As of the end of the second quarter, cash and cash equivalents, including short-term investments, stood at KRW 88 trillion, up KRW 33.6 trillion from the end of the prior quarter. Interest-bearing debt fell by KRW 0.7 trillion to KRW 18.6 trillion. Net cash expanded to KRW 69.4 trillion, and the debt-to-equity ratio improved by 5 percentage points from the prior quarter to 7%.
SK hynix says Agentic AI is widening memory demand beyond HBM
Management argued that AI is evolving into an Agentic form capable of carrying out complex tasks for users over extended periods. As AI spreads across search, coding, productivity tools and other services, the company said the scope of demand keeps broadening.
From a memory perspective, that means more than just HBM for AI servers. SK hynix said server DRAM demand is also rising as Agent-style services expand, while the role of high-performance enterprise SSDs is growing as customers need to process the output generated by AI more efficiently. The company framed this as a structural shift in which AI memory and traditional memory grow in tandem.
At the same time, improvements in AI models and software optimization are lowering compute cost per task. SK hynix said it does not view those efficiency gains as a drag on total infrastructure demand. Instead, the company expects them to lower the price and usage threshold for AI services, broadening the user base and widening adoption.
Management said major technology customers are still expanding infrastructure spending because AI service usage is rising and compute remains tight. Supported by growth in revenue and profit from AI services, their memory procurement also appears to be increasing. According to the company, key customers are still asking for more supply.
In PCs and mobile devices, memory shortages have caused temporary sales adjustments. SK hynix said those segments could gradually recover their growth momentum as supply tightness eases and AI services spread more quickly.
Under supply-constrained conditions, the company expects DRAM demand growth in the mid-20% range and NAND demand growth in the high-10% range. If supply bottlenecks ease later and suppressed demand is met, the market growth path could move higher still.
On supply, management said a material improvement in balance is unlikely in the short term. The reason, it said, is the rising complexity of advanced process technologies used in HBM and AI server memory, along with the construction time needed for new capacity. In its view, tight market conditions are likely to last for a considerable period.
About 10 customers have completed LTA negotiations
To shore up medium- and long-term supply stability, SK hynix is in talks with customers on multi-year contracts, or LTAs. So far, the company said it has completed LTA negotiations with about 10 customers, including core clients, and continues discussions with other major industry participants.
Management described these agreements as more than simple volume commitments. They are meant to support supply stability over the medium to long term and align the development of next-generation memory products with customers' technology roadmaps.
The pricing framework will vary by customer and product. SK hynix said it is discussing several mechanisms designed to better absorb price volatility, while also using financial tools such as deposits to support contract performance and improve visibility into customers' long-term demand plans. In turn, the company expects better visibility to improve investment and production efficiency.
During the Q&A, management said LTA terms are typically around five years, though actual conditions differ by customer and product. It did not disclose what share of sales LTAs will eventually cover, saying only that the level will be managed at an appropriate range based on market conditions and customer demand. The goal is to strengthen downside resilience while retaining flexibility when market conditions improve.
The company also said that through long-term relationships with major AI customers including Nvidia, it has already built a solid profit base in HBM. Looking ahead, it plans to use the demand visibility and operational flexibility provided by LTAs to balance stability and profitability.
Third-quarter outlook points to stronger DRAM shipments
For the third quarter, SK hynix expects DRAM shipments to rise by about 10% from the second quarter, with server products at the center of its response to demand. NAND bit shipments are expected to increase by a low single-digit percentage sequentially.
Management said growing AI model complexity is raising performance requirements for memory, expanding competition beyond standalone memory design into system architecture and packaging. SK hynix said it aims to drive innovation at the system level by combining its DRAM and NAND portfolio, including HBM, with joint development work alongside customers.
HBM4 entered mass production in Q2, while HBM4E samples went out in the first half
On the product roadmap, SK hynix said HBM4 has been optimized to meet customer data-processing speed requirements while delivering industry-leading energy efficiency and cost competitiveness. The company began mass production shipments in the second quarter and plans a full ramp in the second half.
For HBM4E, management said samples were sent to a major customer in the first half of the year. The product uses an optimized process whose maturity and production stability have already been verified. Based on that, SK hynix said development should continue smoothly. The company added that stable supply backed by strong yields and quality, along with cost competitiveness and leading performance, should help preserve its lead in HBM.
In response to questions about competitive pressure, management said HBM4 competitiveness depends not just on hitting performance targets, but also on delivering at scale with stable yields and consistent quality. Since the HBM2 generation, the company said, it has built strengths in time to market, product performance, mass-production yield, quality and customer trust that cannot be replicated quickly.
SK hynix said HBM4 yield and quality are now close to the level of the mature HBM3 generation. The priority now is to expand capacity steadily. HBM4E development is also progressing on schedule, with volume production targeted for 2027.
Its roadmap goes beyond those products. In addition to hybrid bonding, the company said it is developing cooling technology for future products such as HBM5. The technology, as described by management, integrates thermal components inside the package and is expected to reduce thermal resistance by more than 30%, improving system stability and operating efficiency in high-performance, high-density AI environments.
Conventional DRAM and NAND upgrades continue in parallel
In conventional DRAM, SK hynix said it fully began supplying SOCAMM2 products based on the 1c nm process in the second quarter. It plans to refine the product lineup in line with customer development schedules and prepare additional samples to broaden the customer base.
In NAND, the company is accelerating its transition to more advanced process technologies while strengthening its product mix around high-capacity, high-performance offerings to match market demand. It said 321-layer products became the largest portion of NAND output in the previous quarter, and it plans to raise their share of domestic production capacity to around 50% by year-end.
On demand, management said NAND is quickly becoming a core component in the AI memory hierarchy as the market shifts from training-heavy workloads toward inference-heavy ones. As a result, SSD-led NAND demand is rising quickly, and the company expects that trend to continue.
It also argued that the AI storage market cannot be served with a single technology. Requirements for latency, throughput, power, capacity and TCO differ by customer, and what matters is not a specific medium but whether the needed performance can be delivered reliably for each workload.
Management links softer DRAM ASP to mix and shipment timing
Asked why second-quarter DRAM ASP appeared to undershoot market expectations, SK hynix said it manages the sales mix between HBM and conventional DRAM based on customer demand and long-term product strategy. Some shipments of higher-value products were pushed into the second half, and that mix shift affected blended ASP in the quarter. The company said those factors should ease gradually later this year.
With HBM4 shipments ramping fully and 1c nm conventional DRAM shipments rising, SK hynix expects bit growth in the second half to exceed the first-half pace. Changes in customer demand and product mix, including a larger HBM4 contribution and a greater share of high-value products, should also support blended ASP. Higher shipments and a better mix, it said, should lift ASP and earnings in the second half.
Even so, management said its sales strategy is not centered on short-term price moves or near-term profit maximization. Instead, it weighs demand visibility, long-term customer relationships and supply-demand conditions in each product segment. That principle, it said, will remain unchanged.
2026 capex is expected at the high end of the KRW 40 trillion range
SK hynix said that in a market where supply-demand imbalance persists, stable delivery capability, meaning the ability to provide the required volume at the requested time, has become a core operating strength alongside technology leadership.
To respond to strong customer demand and long-term growth opportunities, the company plans to continue investing in capacity expansion in the near term. That includes pulling forward the production schedule for M15X and increasing investment to expand output more quickly. Yongin Fab 1 is scheduled to complete cleanroom opening in early 2027. Because of the faster timeline and larger investment scale, 2026 capital spending is expected to come in at the high end of the KRW 40 trillion range.
Over the longer term, SK hynix said it will prepare infrastructure in advance for future capacity based on discussions with customers and demand forecasts. It has already announced a new investment plan to strengthen advanced packaging capabilities, along with the new NAND production base M17. It also laid out a long-term plan for a new semiconductor cluster in South Korea to meet demand beyond Yongin.
Actual construction, equipment installation and capacity additions will be phased in after considering demand visibility and investment efficiency, management said. The company wants to avoid missing longer-term growth opportunities while preserving capex discipline, supply responsiveness and financial stability.
SK hynix says long-term expansion will not immediately create oversupply
In response to concerns that aggressive expansion could eventually lead to oversupply, the company said its medium- and long-term capacity strategy is built on structurally rising memory demand driven by AI expansion and continued discussions with core customers on longer-horizon demand.
Management said customer relationships are shifting from transactional arrangements to more strategic long-term partnerships. The growing push from customers to sign long-term agreements and build those partnerships is, in its view, evidence that AI ecosystem demand is sustainable.
SK hynix said its current expansion plans are based on market visibility gained through those partnerships. Actual capital deployment and output ramps will be phased in based on demand visibility and investment efficiency, with expansion aligned flexibly to confirmed customer demand. On that basis, management said the long-term investment plan should not immediately result in oversupply.
Nasdaq ADR listing and shareholder returns remain part of the message
On July 10, SK hynix listed its ADR on Nasdaq. The company described the deal as the largest issuance by a foreign company in a U.S. IPO. Management said the listing was not just about raising capital. It also reflected global market confidence in the company's technological competitiveness and growth potential, while broadening its links to the next-generation computing ecosystem.
SK hynix said it will use that foundation to deepen strategic cooperation with major customers and partners, pursue new business opportunities and continue contributing to semiconductor industry development and AI system growth through ongoing technology innovation.
With profit and cash-generation capacity reaching record levels, the company said its financial firepower has strengthened further. At the same time, structural growth opportunities in the AI era are widening, and the scale of investment needed to capture them is much larger than before.
Against that backdrop, management said capital allocation will prioritize growth opportunities with high profitability and strategic value, while also building a financial structure that can support stable operations through market volatility. The company said it intends to continue sharing the results of that approach with shareholders.
Even though future investment needs are expected to rise, SK hynix said its improved cash-generation ability is strong enough to support growth investment targets, maintain financial stability and still allow for a meaningful expansion of shareholder returns. It added that it is reviewing additional ways to execute shareholder return plans from multiple angles.
No new overseas expansion decision has been made
Asked about possible expansion outside South Korea, including into the United States or Japan, management said technology leadership alone is no longer enough in the AI era. Supplying the required quantity at the right time has also become a critical part of competitiveness. In a period of extreme tightness, the company said, providing the ecosystem with the memory products it needs is part of a supplier's responsibility.
Its medium- and long-term investment direction is to invest in line with AI memory demand while keeping capex tied to business feasibility and investment efficiency. Over time, SK hynix plans to secure additional manufacturing capacity through the best mix of maximizing existing sites and building new infrastructure when needed.
Within South Korea, the company said it will continue using Icheon and Yongin as core manufacturing hubs for next-generation DRAM and AI memory, while strengthening Cheongju in both NAND and advanced packaging. The large-scale investments already announced form part of that strategy and are meant to secure the manufacturing base and infrastructure required for future demand.
As for future production locations, management said it will not make decisions based simply on domestic-versus-overseas distinctions. Instead, it will consider factors such as power supply, water resources and labor, the supply chain and semiconductor ecosystem, and customer accessibility.
For now, however, the company said there are no decisions beyond the investments already announced. It added that it will continue securing production sites at the appropriate time to meet customer demand, while improving investment efficiency through the use of existing assets and the evaluation of new projects.

