The AI memory supercycle has entered a new phase. SK hynix shattered records with quarterly revenue surpassing 100 trillion won for the first time, and management declared on an earnings call that booming AI memory demand means "customer demand is twice our supply," with core HBM and DRAM capacity "effectively sold out through 2029" via long-term agreements.
On October 28, SK hynix delivered a historic set of results at its third-quarter earnings briefing: quarterly revenue exceeded 100 trillion won for the first time, and operating margin climbed to a record 80%. More importantly, management made clear on the call that, combining volumes locked in through long-term agreements (LTAs) and customer demand currently under negotiation, "our DRAM and HBM capacity for the next three years through 2029 is effectively sold out," while "customer purchase demand is more than double what we can supply today."
Meanwhile, beyond the stunning backward-looking results (revenue of 104.62 trillion won and operating profit of 83.33 trillion won), the market and investors are more focused on the sustainability of this AI feast. Facing concerns about AI monetization and the pace of capital expenditure, SK hynix management offered extremely strong forward guidance: the company's business model is shifting from traditional "cyclical swings" toward "high-visibility, deterministic growth" anchored by long-term agreements.
As the CEO summarized at the end of the call:
"We have posted record results for six consecutive quarters, but we believe what matters more is that the underlying business environment is changing. The customer trust we have built through technology leadership is translating into long-term agreements, giving us greater medium- to long-term demand visibility and a more solid foundation for sustained profitability. This means a new growth phase, different from the past, is becoming visible through concrete changes in our business structure."
SK hynix shares opened higher and extended gains throughout the day, rising more than 11% intraday after the results were released before closing up 10.89%. South Korea's KOSPI index closed up more than 6%.
Demand is twice current supply as capacity is effectively sold out for three years
The most encouraging incremental information from the call was undoubtedly SK hynix's multi-year order visibility.
Management disclosed that SK hynix has signed long-term supply agreements with 14 core customers, including major global cloud service providers (CSPs). The agreements are typically five-year terms, and the company is discussing extensions with some customers. Volumes locked in through LTAs already account for more than 50% of total capacity in 2027 and 2028, and the company plans to maintain that ratio over the medium to long term.
The LTA pricing structure includes a price range agreed with customers and safeguard mechanisms such as deposits to ensure customer procurement plans translate into actual orders and shipments, while supporting stable profitability and allowing the company to capture opportunities from improving market prices.
Responding to Goldman Sachs analyst Giuni Lee's question about the significance of LTAs, SK hynix management gave its clearest supply-demand assessment to date:
"The volumes already locked in through LTAs, plus demand currently under negotiation with customers, mean our DRAM and HBM capacity for the next three years through 2029 is effectively sold out. The volumes under negotiation are based on customers' specific procurement plans, so the probability of conversion is extremely high."
On the sustainability of HBM demand, management provided a quantified statement:
"Demand is so strong that customer purchase demand is more than double what we can supply today, which also confirms customers' continued strong preference for our products."
Supply gap seen through 2030, with capex to rise sharply in 2027
SK hynix Chief Financial Officer Woo Hyun Kim offered several specific forecasts in his outlook:
Global CSP capital expenditure is expected to grow more than 60% in 2027, with approved power capacity up more than 100% year over year;
The global AI data center infrastructure market is expected to maintain a compound annual growth rate of 40% through 2030;
The DRAM market could see demand growth of about mid-20% in 2027, with NAND at high-10%;
The company's absolute revenue and profit are both expected to reach "significantly higher levels" in 2027;
Capex in 2027 will increase significantly from 2026, funding advanced process equipment, the start of production at the first Yongin fab, and accelerated infrastructure construction at home and abroad.
Management also emphasized the structural constraints on the supply side: "Because advanced processes are becoming increasingly difficult to implement and new capacity requires long construction cycles, the supply shortage is expected to persist until 2030."
To address this long-term structural shortage, SK hynix gave clear guidance on future capacity expansion: "As we invest in advanced process equipment, ramp capacity at the first Yongin fab, and accelerate infrastructure investment at home and abroad, we expect capital expenditure in 2027 to increase sharply from 2026."
On the risk of oversupply, management was cautious:
"More investment and more fabs do not equal more supply. As the mix shifts toward high-performance products like HBM, processes become increasingly complex and require more resources, while improving bit output through finer processes is becoming increasingly constrained."
Some major customers cutting HBM from 12 layers to 8? "It won't affect our business"
Responding to market concerns over whether some major customers shifting HBM from 12 layers to 8 layers could hurt the company's performance, SK hynix offered a firm rebuttal on the call.
Reports previously said that to ease supply chain pressure, Nvidia quietly lowered HBM4 bandwidth specifications and downgraded 12-layer stacks to 8-layer configurations.
Management noted that such specification adjustments are precisely a reflection of "demand far exceeding supply": "Because HBM supply remains constrained, customers need to build more AI accelerators and systems with limited quantities. Therefore, we view reducing stack layers as a strategy to obtain more cubes at the same total capacity."
Management further explained that fewer layers do not necessarily lead to a proportional decline in HBM capacity demand:
"Even switching from 12-layer HBM4 to 8-layer HBM4E, higher capacity per die can offset a considerable portion of the impact. Therefore, even considering such adjustments, our medium- to long-term HBM demand outlook is unchanged."
More importantly, fewer layers do not weaken SK hynix's business. Management explained: "Providing the same bit capacity with fewer layers requires more cubes, which in turn increases demand for base dies and back-end processes. We have already incorporated this into our production and supply plans."
Management stressed: "Profitability is not determined by layer count alone, but by a combination of transaction terms for each product, product mix, yield, back-end production efficiency, and cost." Next year, the HBM4 mix and price will rise, mass-production yield and cost competitiveness will improve further, and HBM revenue and profit are expected to grow sharply year over year, "far above peak levels of past cycles from a profitability standpoint."
Fourth-quarter guidance: HBM4 to surpass HBM3E, DRAM shipments up about 25% quarter over quarter
SK hynix Chief Financial Officer Woo Hyun Kim disclosed that the company will further accelerate expansion in the fourth quarter to proactively meet market demand:
DRAM: Expansion of 1c nanometer processes and increased HBM4 output at the M15X line will proceed in parallel, with fourth-quarter shipments expected to grow "about 25% (mid-20%)" quarter over quarter. For conventional DRAM, the company will quickly increase the share of 1c nanometer server products and expand sales of 1c nanometer DDR5 and AI server LPDDR5 SOCAMM.
HBM4: "12-layer HBM4 is shipping smoothly on the schedule agreed with major customers; M15X will reach full capacity before year-end." In the fourth quarter, HBM4 is expected to surpass HBM3E in both share of total HBM shipments and revenue, becoming the flagship product. The company has finalized next year's HBM supply volumes and prices with major customers, "faithfully reflecting the increased product value of HBM4, and the profitability gap between HBM and general DRAM is expected to narrow significantly."
NAND: Fourth-quarter bit shipments are expected to grow "mid-single percent (about 5%)" quarter over quarter; the 321-layer process will expand to more than 50% of domestic capacity in South Korea, increasing supply of 321-layer enterprise SSDs; the company will also expand supply of high-performance, high-density enterprise SSDs suited to KV cache tiering.
The company also disclosed that on next-generation products, HBM4E completed sample shipments in the second quarter, with development and customer certification proceeding on customer timelines; it is also working closely with major customers on Custom HBM projects.
Additionally, while demonstrating extremely high earnings certainty for years ahead, SK hynix unveiled an unprecedented shareholder return plan to offset concerns about free cash flow amid heavy capital expenditure.
The company announced plans to return about 100 trillion won to shareholders for the 2026 fiscal year (including more than 30 trillion won in cash dividends and more than 60 trillion won in share buybacks and cancellations). Of the third-quarter dividend of 20,000 won per share, half (10,000 won) is a "special dividend" derived from investment gains related to Japan's Kioxia.
Management said:
"We believe our current share price is undervalued and does not fully reflect our earnings and cash flow."
The full transcript of SK hynix's third-quarter 2026 earnings call is as follows (AI-assisted translation):
Company participants
Kwak Noh Jung, President and Chief Executive Officer
Seong Hwan Park, Head of Investor Relations
Unidentified speaker
Woo Hyun Kim, Vice President, Chief Financial Officer and Head of Finance
Other participants
Dong-Hee Han, SK Securities
Giuni Lee, Goldman Sachs
Jay Kwon, JPMorgan
Peter Lee, Citigroup
Simon Woo, Bank of America
Sunwoo Kim, Meritz Securities
Young-ho Ryu, NH Investment & Securities
Youngjin Kim, Mirae Asset Securities
Presentation
Operator:
Good morning, and thank you for joining SK hynix's earnings release conference call. Today, we will first have SK hynix's earnings presentation, followed by a Q&A session. (Moderator instructions) Please note that the presentation will have simultaneous interpretation, and the Q&A session will have consecutive interpretation. Now, we are ready to begin.
Seong Hwan Park:
Good morning, good afternoon, and good evening. I am Park Seong-Hwan, Head of Investor Relations at SK hynix. Welcome to SK hynix's third-quarter 2026 earnings release conference call. Let me introduce the executives joining me today. Those attending are: President and Chief Executive Officer Kwak Noh Jung; Chief Financial Officer Kim Woo-Hyun; Head of DRAM Marketing Park Jun-Deok; Head of NAND Marketing Song Chang-Seok; and Head of HBM Sales and Marketing Kim Ki-Tae.
Please allow me to make a disclaimer: The third-quarter results included in this call are consolidated and preliminary, and the external auditor's review has not been completed, so the figures may still change. In addition, forward-looking statements, including market outlook and company plans, may be adjusted due to changes in macroeconomic and market conditions.
Now, we will formally begin SK hynix's third-quarter 2026 earnings release conference call. Before the earnings presentation begins, Chief Executive Officer Kwak Noh Jung will first speak about the Solidigm IPO and shareholder returns. Then, Chief Financial Officer Kim Woo-Hyun will present on third-quarter results, the company's future plans, and the market outlook.
Kwak Noh Jung:
Good morning, everyone. I am Chief Executive Officer Kwak Noh Jung. Amid recently heightened market volatility, I would like to express my sincerest gratitude to all shareholders for your unwavering attention and trust in SK hynix. Today, I want to explain directly to shareholders our decision regarding the Solidigm IPO, an issue that has recently drawn widespread market attention, as well as our arrangements to share the returns from strategic investments with shareholders through a special dividend.
First, Solidigm has been exploring financing options to support strategic initiatives such as expanding production scale and meeting growing demand. Among the options considered, bringing in external capital through an IPO was originally expected to reduce our investment burden and enhance our ability to return capital to shareholders. However, after comprehensively considering recent financial market volatility and the correction in semiconductor stocks, Solidigm concluded that proceeding with an IPO at this time would not fully realize the expected benefits, and therefore decided to terminate its evaluation of the Pre-IPO and IPO options.
Solidigm will continue to strengthen its competitiveness and actively seize growth opportunities in the AI data center storage market. Investments and financing arrangements to support these goals will proceed steadily in ways that enhance the corporate and shareholder value of both Solidigm and SK hynix. We remain committed to the principle of sharing most of the returns from strategic investments with shareholders. By using these returns to supplement our shareholder return capacity, which may decline temporarily due to continued future investment, we aim to protect shareholder interests while continuing to invest boldly to drive sustainable growth.
In line with this principle, we plan to return to shareholders, in the form of a special dividend, the gains realized this year from our investment in Kioxia. We will firmly honor this commitment so that shareholders can directly share in the success of our strategic investments.
Next, the Chief Financial Officer will report on this quarter's results and outlook, as well as the details of the shareholder return plan. Thank you.
Woo Hyun Kim:
Good morning, everyone. I am Chief Financial Officer Woo Hyun Kim. First, please allow me to present the company's third-quarter 2026 results. In the third quarter, as AI memory demand continued to grow, supply-demand conditions remained tight and memory prices stayed elevated. Both DRAM and NAND recorded high price increases for a second consecutive quarter, driven mainly by continued price increases for HBM, server DRAM, and enterprise SSDs.
As a result, third-quarter revenue rose 32% quarter over quarter and 328% year over year to 104.6 trillion won. This was the first time quarterly revenue exceeded 100 trillion won, setting a record high. On DRAM, we expanded sales centered on AI servers and achieved about a 10% quarter-over-quarter increase in bit shipments as planned. In particular, HBM4 officially began mass production and shipments in the third quarter. In addition, server DDR5 shipments grew more than 40% from the second quarter, driving overall DRAM bit shipment growth.
Due to continued strong conventional DRAM prices and a larger share of HBM4 sales, which carry higher prices than the previous generation, DRAM average selling price (ASP) rose by a high-teen percentage, recording a growth rate similar to the previous quarter. On NAND, although we adjusted our sales strategy to focus on high-value-added products such as 321-layer TLC enterprise SSDs, leading to a slight decline in bit shipments from the second quarter, strong overall prices and a larger share of SSD sales drove ASP up by a mid-teen percentage.
Despite unfavorable foreign exchange effects from a stronger Korean won, higher product prices and expanded sales of high-value-added products lifted third-quarter operating profit to 83.3 trillion won, up 38% quarter over quarter and 632% year over year. Operating margin rose 3 percentage points quarter over quarter to 80%, with both operating profit and operating margin setting quarterly record highs.
In the third quarter, depreciation and amortization expenses totaled 4.3 trillion won. EBITDA reached 87.7 trillion won, with an EBITDA margin of 84%. Among major non-operating items, foreign exchange-related net losses of 11.6 trillion won due to currency declines and valuation losses on investment assets of 22.9 trillion won resulted in total non-operating expenses of 32.6 trillion won. As a result, pre-tax profit was 50.7 trillion won, net profit was 39.8 trillion won, and net margin was 38%.
As of the end of the third quarter, cash and cash equivalents stood at 114.5 trillion won, up 26.5 trillion won from the end of the previous quarter. Interest-bearing debt decreased by 2.8 trillion won to 15.8 trillion won. As a result, net cash expanded to 98.7 trillion won, and the debt-to-equity ratio improved by 2 percentage points from the end of the previous quarter to 5%.
Next, I will discuss the market outlook. The AI market is continuously evolving. As AI drives productivity gains and new revenue streams across fields such as software development, customer response, and search, increased AI usage further promotes infrastructure investment, creating a virtuous cycle. This change is reflected in the investment scale and approved power capacity of major global cloud service providers (CSPs). According to the latest forecasts from market research firms, global CSP capital expenditure is expected to grow more than 60% in 2027, and approved power capacity is also expected to increase more than 100% year over year. More notably, this growth will not stop next year. Driven by commercialization of AI services and expansion of application areas, the global AI data center infrastructure market is expected to grow at a high compound annual rate of about 40% through 2030. This medium- to long-term growth trend is expected to continue.
In particular, as AI plans complex tasks and executes them over long periods using various tools, AI computing demand and the volume of data generated, transmitted, and stored are surging. As a result, demand growth across the entire memory industry has significantly exceeded initial expectations. AI infrastructure investment is moving from an initial stage of a few players racing to stake out positions to a broader stage involving more enterprises and governments. Therefore, even considering some potential investment adjustments, we expect growth in AI infrastructure investment to remain strong.
In the near term, enterprise AI and personal AI are expected to lead demand growth; over the medium to long term, physical AI such as autonomous driving and robotics is expected to further broaden the scope of demand. Although PC and mobile device shipments are expected to decline due to cost pressure from higher memory prices and supply shortages, growth in AI infrastructure demand is expected to more than offset this decline, keeping the overall memory market on a growth trajectory.
On the other hand, memory supply is struggling to keep up with such rapid demand growth, so we expect it will be difficult to fully satisfy customer volume demand in the near term. Given the complexity of implementing advanced processes and the construction cycles required for new capacity, the situation of supply failing to keep pace with demand is expected to persist for some time, and this supply bottleneck is expected to continue until 2030.
Despite strong demand, considering actual supply conditions, demand growth achievable in 2027 is expected to be similar to this year, with DRAM growth around the mid-20% level and NAND around the high-10% level. We expect this favorable market environment to provide a solid foundation supporting the company's growth next year. Based on expanded capacity and the competitiveness of high-value-added products, we plan to achieve substantial increases in both absolute revenue and profit.
Next, I will discuss the company's fourth-quarter sales plan and future operating direction. In the fourth quarter, we plan to further accelerate supply expansion and respond more actively to market demand. First, on DRAM, increased output from 1c nanometer process expansion and higher HBM4 output from the M15X ramp will proceed in parallel. As a result, fourth-quarter shipments are expected to grow sharply by about the mid-20% level quarter over quarter. On conventional DRAM, we are rapidly increasing the share of 1c nanometer products for server applications. In the fourth quarter, we plan to further strengthen our product mix centered on high-value-added server products by expanding sales of 1c nanometer DDR5 and AI server LPDDR5 SOCAMM.
Shipments of our 12-layer HBM4 are proceeding smoothly on the schedule agreed with major customers. To actively respond to the surge in customer demand, we are accelerating the M15X ramp and plan to reach full capacity before year-end. As a result, in the fourth quarter, HBM4 is expected to surpass HBM3E in both share of total HBM shipments and revenue, becoming our flagship product.
The company has completed negotiations with major customers for next year, including HBM volumes and prices. These negotiations fully reflect the increased product value of HBM4. Therefore, we expect the profitability gap between HBM and conventional DRAM to narrow significantly. On NAND, in the fourth quarter we plan to expand the share of the flagship 321-layer process to more than 50% of domestic capacity. Through this, we will expand supply of 321-layer enterprise SSDs, and fourth-quarter NAND bit shipments are expected to grow by a mid-single-digit percentage quarter over quarter.
Building on supply expansion, we plan to further expand NAND's role and business opportunities in the AI infrastructure market. We will expand supply of high-performance, high-density enterprise SSDs suited to KV cache tiering. Over the medium to long term, through HPF and system optimization technologies, we aim to evolve NAND into a core solution that improves AI service throughput and cost efficiency.
To enhance medium- to long-term demand visibility and lay the foundation for stable growth, we are continuing to expand long-term agreements (LTAs) with customers. To date, we have signed LTAs with 14 important customers, including major global CSPs. Although LTAs are typically based on five years, we are proactively discussing extensions of existing contract terms with some customers. The fact that customers are seeking longer-term cooperation based on our product roadmap and supply capability shows that LTAs are evolving into an important mechanism for strengthening customer relationships and business continuity.
Pricing structures vary by customer and product, but are generally designed to reflect market conditions within a price range agreed with customers. In addition, to ensure that customer procurement plans genuinely translate into actual orders and shipments, we have included safeguard provisions such as deposits. This structure not only helps us maintain stable profitability but also enables us to capture upside opportunities by reflecting improvements in market prices and product value.
The volumes locked in through LTAs currently account for more than 50% of our total capacity in 2027 and 2028, laying the foundation for medium- to long-term profit stability and demand visibility. At the same time, to respond to customers' medium- to long-term demand and growth opportunities, we are continuing investment to expand capacity. With equipment investment as the share of advanced processes rises, the ramp of the first Yongin fab, and accelerated infrastructure investment at home and abroad, capital expenditure in 2027 is expected to increase significantly from 2026. The scale and timing of investment will be determined after comprehensively considering medium- to long-term demand visibility obtained through customer communication, product competitiveness, profitability, and investment efficiency.
Finally, please allow me to explain the shareholder return plan. In August last year, we announced a shareholder return plan including 40 trillion won in share buybacks and cancellations, and said we would provide detailed information on the scale and method of additional shareholder returns at the third-quarter earnings briefing. Today, I will update on the execution of that plan and introduce an additional shareholder return plan reflecting our improved performance and investment results.
The company has decided to pursue about 100 trillion won in shareholder returns for the 2026 fiscal year. We plan to execute no less than 30 trillion won in cash dividends and no less than 60 trillion won in share buybacks and cancellations (including the 40 trillion won announced in August). Of that, the 40 trillion won share buyback was completed last week and will soon be cancelled.
Through dividends, we will directly share performance results with shareholders, and we will continue to buy back and cancel shares to ensure that per-share value appropriately reflects the company's intrinsic value. Through these measures, we will fulfill our commitment to return more than 50% of free cash flow to shareholders. Today, the board resolved to carry out 14 trillion won in cash dividends and 20 trillion won in share buybacks. Of that, about 10 trillion won in treasury shares is expected to be used for executive and employee compensation. Excluding this portion, third-quarter shareholder returns amount to about 24 trillion won.
I will explain item by item. First, we will expand cash dividends to reflect improved performance and investment results. We will determine dividend levels based on operating performance, and in the event of significant investment gains, we will additionally share these results with shareholders through special dividends. We will use both the operating results of our core business and the value realization of investment assets as sources of shareholder returns. Accordingly, the third-quarter dividend per share is set at 20,000 won, totaling 14 trillion won. This includes a 10,000 won dividend reflecting improved performance and a 10,000 won special dividend sharing the results of the Kioxia investment. We plan to return most of the Kioxia investment gains to shareholders through special dividends. Part of this will be paid as this quarter's dividend, and the remainder will be included in the year-end dividend after the 2026 fiscal year financial results are finalized.
Second, we will continue quarterly share buybacks. The quarterly buyback size will be resolved by the board after comprehensively considering the share price, market conditions, and cash flow. This quarter, we plan to buy back a total of 20 trillion won in shares. Of that, 10 trillion won is for shareholder return purposes and will all be cancelled; as noted earlier, the other 10 trillion won is for executive and employee compensation purposes and is not included in the total 2026 fiscal year shareholder returns.
Looking ahead, based on strong memory demand growth and the company's technological competitiveness, we will continue to steadily generate results and continuously share them with shareholders. Thank you.
That concludes my presentation, and we will now begin taking questions.
Q&A session
Operator (Q&A session):
We will now begin the Q&A session. (Moderator instructions) The first question comes from Peter Lee of Citigroup. Please go ahead.
Q - Peter Lee:
Congratulations on the strong results. My question is about AI investment. We see AI investment continuing to grow, but AI demand or the monetization from AI infrastructure buildout seems slow to materialize. As a result, there are some market concerns about whether AI demand can be sustained. Against this backdrop, my question is: on what basis does the company believe the AI investment momentum will continue?
A - Unidentified speaker:
Thank you for your question. We are indeed aware of some market discussion about the pace of AI investment, but we view this as part of a normal process, in which customers are more carefully examining the feasibility and profitability of individual projects and clarifying the timing and pace of investment. We believe AI investment momentum will continue for two main reasons.
First, we see that customers' AI investments are effectively driving broader use of AI services and producing actual business results. In their latest earnings releases, major hyperscale cloud customers reported cloud revenue growth of 30% to 80% year over year, and we believe a considerable portion of that came from expansion of their AI businesses. At the same time, we are seeing more long-term contracts for large-scale enterprise AI solutions, with cloud order backlogs tripling to quintupling year over year. Paying users are also steadily increasing, further confirming continued growth in AI-related sales and financial performance for major customers across multiple areas.
Based on this trend, major customers' AI sales and cash flow will only continue to improve, which will also help ease current market concerns about the investment burden. At the same time, customers are working from multiple angles to ease current constraints, such as multi-site deployment, small and medium-sized data centers with easier access to power, and upgrades to on-site power generation and cooling technologies. These measures will help reduce the investment burden.
The second reason is that the investment base is structurally becoming more diversified. Early AI investment was concentrated among a few large technology companies, so an investment adjustment by a single company could directly cause overall demand to fluctuate. But today, the investment base is rapidly expanding to include non-tech companies building their own AI infrastructure, sovereign AI investment, and on-premises deployment demand. This marks a structural shift in which investment adjustments by some investors will be offset by increased demand from others. For example, even if investment in certain projects slows due to interest rate pressure, policy-supported sovereign AI investment or broader enterprise AI adoption can still fill the gap. As a multi-layered structure with more diverse participants, motives, and resources takes shape, the impact of a single company's funding situation or investment adjustment on the overall investment trend will become increasingly limited.
In summary, supported by actual business results and an increasingly diversified investment base, AI investment and demand intensity will remain sufficiently strong. In addition, in the memory market, we expect the current structural supply-demand imbalance, in which demand exceeds supply, to persist for some time.
Operator:
The next question comes from Sunwoo Kim of Meritz Securities. Please go ahead.
Q - Sunwoo Kim:
Thank you. I would also like to congratulate the company on its outstanding results, and especially thank CEO Kwak Noh-Jung for his update on the company's latest situation. My question is about shareholder returns. How does the company determine the ratio between dividends and share buybacks?
A - Unidentified speaker:
Thank you very much for your question. For us, we are committed to seeking a balance between cash dividends and share buybacks and cancellations, that is, between directly sharing improved performance results with shareholders and enhancing per-share value by reducing the number of shares outstanding.
First, we set cash dividends at a sustainable level. To do so, we comprehensively considered performance improvement, historical dividend yields, global peer dividend yields, and an appropriate level of cash holdings. We also plan to share meaningful returns with shareholders when returns from the Kioxia investment are generated. Most of that will be paid as special dividends so that shareholders can truly feel the returns. Although this year's accounts have not yet been settled, we believe it is meaningful to share part of the expected returns with shareholders in advance. Therefore, part of the special dividend will be paid early in the third quarter, and the remainder will be included in the year-end dividend after the investment returns are finalized.
After cash dividends, the remainder of total shareholder returns will be used for share buybacks. We believe the company's current share price is undervalued and does not fully reflect our profitability and cash flow levels. For this reason, we believe buybacks and cancellations are an effective way to enhance per-share value. Buybacks will continue quarterly within the annual scale, and the board will determine the quarterly buyback size after comprehensively considering factors such as share price, valuation, and cash flow. Looking ahead, we will rely on strong cash generation to effectively implement our shareholder return policy and remain committed to enhancing shareholder value.
A - Seong Hwan Park:
Thank you very much. We will continue with the next question.
Operator:
The next question comes from Giuni Lee of Goldman Sachs. Please go ahead.
Q - Giuni Lee:
Congratulations on the outstanding results. My question is about long-term agreements (LTAs). It was mentioned that volumes currently locked in through LTAs account for about 50% of total capacity in 2027 and 2028. What does this mean for the company's performance and long-term growth potential?
A - Speaker:
Thank you for your question. By increasing LTA coverage, our goal is to improve long-term demand visibility and thereby achieve stable revenue and profit growth. The volumes already locked in through LTAs, plus demand currently being discussed with customers, mean our DRAM and HBM sales for the next three years through 2029 are effectively sold out. The volumes currently under discussion are based on customers' specific procurement plans, so the probability of conversion is extremely high. Overall, we believe the company's medium- to long-term sales visibility has improved significantly.
As noted earlier, volumes committed under LTAs account for more than 50% of total capacity in 2027 and 2028, and we plan to maintain that ratio over the medium to long term. This arrangement will help us secure stable profit margins. At the same time, we can use non-LTA capacity flexibly to respond to market changes and next-generation product demand, continuously strengthening future competitiveness.
A - Seong Hwan Park:
Thank you. Next question, please.
Operator:
The next question comes from Simon Woo of Bank of America. Please go ahead.
Q - Simon Woo:
My question is about the possibility of overcapacity, because we have experienced some difficult cycles in the past. Concerns about potential overcapacity stem from the fact that, with demand currently remaining strong, SK hynix is still continuing to build new fabs or expand and upgrade existing fabs, including the Yongin project. So if demand does not meet expectations, what is the company's strategy or management's response plan? I would also like to congratulate the company on its outstanding results and the very detailed shareholder return plan.
A - Speaker:
Thank you for your question. Indeed, the company is currently advancing medium- to long-term investment to meet strong demand and secure sufficient space for equipment and production needed for technology migration. But more investment or more fabs do not necessarily mean more supply. As the share of high-performance products such as HBM increases and process technologies become more complex, the resources required for production continue to rise, while improving bit output through finer processes is becoming increasingly constrained. Therefore, our investment today is not only to expand capacity but also to build the production foundation needed for technology migration and expansion of high-value-added product lines.
Our investment will be executed in stages. Infrastructure will be prepared in advance to respond to medium- to long-term demand. But key resources needed to increase supply, such as equipment installation and mass-production ramp, will be adjusted according to demand. For example, for new fabs including the Yongin cluster and M17, we plan to accelerate construction of the main structures, but cleanrooms will be opened in stages. This approach is specifically designed to ensure flexibility in future operations.
In assessing demand, we look not only at customers' capital expenditure or demand forecasts but also review their financial condition and funding capacity, as well as the readiness of key supply chains, before incorporating them into our investment plans. The company's own financial soundness is also an important parameter. While considering medium- to long-term net cash targets and an appropriate level of cash holdings to maintain sufficient financial room, we will manage the scale and pace of investment based on demand visibility and investment returns.
In summary, we will prepare medium- to long-term supply infrastructure as planned based on improved demand visibility, but actual capacity ramp will be strictly tied to demand and profitability. If actual conditions differ from expectations, we will ensure effective response to various risks by flexibly adjusting investment spending, production plans, and resource allocation.
A - Seong Hwan Park:
Thank you. Next question, please.
Operator:
The next question comes from Jay Kwon of JPMorgan. Please go ahead.
Q - Jay Kwon:
Congratulations on the strong results, and thank you very much for taking my question. You mentioned HBM4 supply earlier, and I would like to learn more about the specific details of HBM4 supply, and could you also share plans for the successor product HBM4E and Custom HBM?
A - Unidentified speaker:
Thank you very much for your question. As noted earlier, thanks to smooth mass-production progress, HBM4 is expected to surpass HBM3E in the fourth quarter in both bit shipments and revenue share. Yields are also quite stable. The company maintains industry-best yield levels in both fabs and back-end processes. In particular, what we call back-end yield covers the entire process, not just individual steps such as bumping, stacking, or KGSD testing. On this basis, HBM4 has already exceeded its established yield target, so the target has been raised. HBM3E has also exceeded planned mature yields.
Currently, we are focused on maximizing capacity to meet strong customer demand. Our performance in the HBM business will once again demonstrate our unrivaled competitiveness. Of course, we are also steadily preparing next-generation products. For example, HBM4E completed sample shipments in the second quarter, and development and certification work is proceeding in line with customer timelines.
Looking further ahead, in addition to stably supplying standard HBM, we also see the growing importance of Custom HBM, which is specifically optimized for each customer's system and workload. Therefore, we are working closely with major customers on related projects. As the preferred supplier for core customers, we will continue to drive HBM business growth based on proven product competitiveness and reliable mass-production capability.
A - Seong Hwan Park:
Thank you. Next question.
Operator:
The next question comes from Dong-Hee Han of SK Securities. Please go ahead.
Q - Dong-Hee Han:
Thank you for taking my question. My question is also about HBM, specifically about the possibility of reducing HBM stack layers. How does the company view the possibility of reducing HBM stack layers? What impact could this have on the HBM business, including shipments and profitability?
A - Unidentified speaker:
Thank you for your question. The recent moves by some customers to adjust HBM specifications actually show that demand truly far exceeds supply. Because HBM supply remains tight, customers need to build more AI accelerators and systems with limited quantities. Therefore, we view reducing stack layers as a strategy aimed at obtaining more cubes at the same capacity.
That said, fewer stack layers do not necessarily mean a proportional decline in HBM capacity. Even switching from 12-layer HBM4 to 8-layer HBM4E, higher capacity per die can offset a considerable portion of the impact. Therefore, even considering such adjustments, our medium- to long-term HBM demand outlook remains unchanged.
Demand remains very strong, with customer purchase demand more than double our current supply capacity, which fully reflects customers' continued strong preference for our products. The supply side will also not be affected. Supplying the same bit volume with fewer stack layers requires more cubes, which in turn increases demand for base dies and back-end processing. We have incorporated this factor into our production and supply plans and will execute HBM shipments and supply as planned.
As for profitability, it is not determined solely by stack layers, but by a combination of multiple factors, including transaction terms for each product, product mix, yield, back-end production efficiency, and cost. Next year, the HBM4 mix and its price will rise, while mass-production yield and cost competitiveness will improve further. HBM revenue and profit are expected to achieve significant year-over-year growth, and profitability will rise substantially. Looking ahead, profitability is expected to remain at high levels, far exceeding peak levels of past cycles.
A - Seong Hwan Park:
Thank you. Next question.
Operator:
The next question comes from Youngjin Kim of Mirae Asset Securities. Please go ahead.
Q - Youngjin Kim:
First, congratulations on the record results. My question is about 4F-square. We see other companies also researching new memory cell structures and next-generation DRAM technologies, which has raised high attention on long-term technology competition. What is the company's latest progress in preparing for 4F-square? In which areas is the company focusing to strengthen technological competitiveness?
A - Unidentified speaker:
Thank you for your question. 4F-square is a next-generation DRAM technology whose memory cell area is smaller than 6F-square, enabling higher memory density. Its applicability and strategic direction depend on each company's process scaling capability, process environment, and cost structure. For our company, our development plan is to extend the use of the 6F-square structure as long as possible through process maturity and a stable technology platform, and then transition to a 4F-square or 3D DRAM platform.
Through EUV-based process scaling and process innovation, we already have the technical capability to further extend the existing 6F-square platform. Therefore, we believe the most efficient strategy for the company at present is to fully leverage the advantages of this platform, whose mass-production reliability and cost competitiveness have already been proven, to further enhance product competitiveness.
To apply a new memory cell structure to products, it is necessary not only to ensure performance but also to guarantee yield, reliability, and cost competitiveness. In addition, product design and customer certification must all be completed. This is why the timing of technology development and the timing of commercialization are independent of each other.
We are formulating a step-by-step technology roadmap covering the full path from ultimate scaling of existing DRAM to 4F-square and then to 3D DRAM. Each technology will be applied to the right product at the right time, with "the right time" meaning when it is ready both technologically and economically. We will continue to leverage our strengths in process and design capabilities, mass-production experience, and customer collaboration to maintain technological competitiveness in next-generation DRAM by balancing overall performance, cost efficiency, and manufacturability.
A - Seong Hwan Park:
Thank you. We will take the last question.
Operator:
The last question comes from Young-ho Ryu of NH Investment & Securities. Please go ahead.
Q - Young-ho Ryu:
Thank you for the opportunity to ask a question. My question is about NAND. We have noticed that the scale of KV cache is growing rapidly, and research is currently exploring how to store and reuse it across multiple storage tiers. How does the company expect this trend to affect future NAND demand? And what is the company doing to prepare?
A - Unidentified speaker:
Thank you for your question about NAND. As agentic AI and long-context use cases become more widespread, the scale of KV cache is expanding rapidly, and due to capacity and cost constraints, it is becoming difficult to store it entirely in HBM or DRAM. Therefore, the importance of tiered storage architecture is growing, in which data is distributed across HBM, DRAM, and SSD according to access frequency. This trend is expected to create new demand for NAND to store and reuse context information needed for AI inference. SSDs are already being used in some AI systems to offload KV cache, and we believe their role will further expand into a context storage layer, supporting large-scale KV cache sharing across multiple servers and GPUs.
To seize this opportunity, the company is developing several storage tiering technologies, including Semantic-Aware Lifecycle Tiering for KV (SALT-KV) and CMM-Hybrid. SALT-KV is a software technology that optimizes data allocation among HBM, DRAM, and SSD by evaluating the reuse value and storage cost of each KV cache. CMM-Hybrid is a CXL-based hybrid storage technology that stores the full KV cache in SSD and preloads data likely to be called into DRAM in advance.
Recently, through system demonstrations based on real AI workloads, we are defining the applicability and performance requirements of eSSD for actual inference environments, in order to drive adoption by customers and platform partners' architectures. Over the medium to long term, we plan to use HPS (High Performance Storage) to fill the gap in the emerging storage tier between HBM and SSD, combining high bandwidth with NAND's high capacity and cost efficiency. By proactively positioning ourselves for new NAND demand opportunities arising from changes in AI system architecture, we will drive continued growth in the NAND business.
A - Seong Hwan Park:
Thank you all for your answers. Due to time constraints, this Q&A session is now over. Next, we will conclude SK hynix's third-quarter 2026 earnings release conference call with closing remarks from the CEO.
A - Kwak Noh Jung:
We have posted record results for six consecutive quarters, but we believe what is more profound is that the underlying business environment is undergoing fundamental change. The customer trust we have built through technology leadership is translating into long-term cooperation agreements, giving us clearer medium- to long-term demand visibility and a more solid foundation for sustained profitability. This means the potential for a new growth phase, different from the past, is gradually becoming visible through concrete changes in our business structure.
Our current task is to turn these changes into lasting competitive advantages. We will extend our existing leadership in memory, including HBM, into next-generation products and become the partner customers seek first when planning for the future. We will turn market opportunities into actual results and further strengthen our ability to continuously generate performance. As our business scale expands and investment returns materialize, the returns we share with shareholders should also increase accordingly. Today's shareholder return plan is a reflection of this principle. We will continue to share results with shareholders through sustained performance and continuously strengthen confidence in the company's future growth. Thank you very much for your continued attention and support.
A - Seong Hwan Park:
That concludes the closing remarks, and this earnings release conference call is now officially over. Thank you very much for participating.
Comments