The First ₩10 Trillion Operating-Profit Quarter, and 2026 Is Already Sold Out: A 47% Margin, HBM4 Shipping, NAND Pricing Recovered. Maintaining Outperform After a +207% Year

Results vs. Consensus

Q3 2025 Scorecard

Year-Over-Year and Quarter-Over-Quarter (K-IFRS)

Quality of Beat

Revenue: The 1.3% miss versus consensus is immaterial against a +10% sequential and +39% YoY record, and the composition is high quality: growth came from price (DRAM ASP +mid-single-digit, NAND ASP +low-10%) and high-value mix (HBM3E 12-high, 128GB-plus DDR5 doubling, premium eSSD) rather than low-margin volume. A modest revenue miss alongside a 6-point margin beat is a better outcome than the reverse.

Margins: The 47% operating margin is the centerpiece and it is clean. The six-point sequential expansion reflects the mix shift to leading-edge product and broad pricing strength across DRAM and NAND. This is operating leverage at the top of a pricing cycle, the inverse of memory's trough economics, and the figure to watch is whether it can hold as conventional volume normalizes.

EPS / net income: The +42% EPS beat is genuine on the operating result but flattered by a ₩3.3tn investment-asset valuation gain that will not recur at that scale. Pre-tax income was ₩14.8tn; the ₩3.4tn net non-operating gain (valuation plus a small FX gain on a quarter-end-stronger dollar) is the swing versus the prior quarter's non-operating loss. Strip it and the quarter is still exceptional, just on the operating line rather than the net line.

Segment Performance

Product-Line Detail — Q3 2025

DRAM — Pricing Broad, HPC Mix Compounding

DRAM bit shipments grew high-single-digit QoQ, above guidance, on HBM3E 12-high, server DDR5, and a seasonal LPDDR5 recovery. ASP rose mid-single-digit with, notably, strong ASP growth in conventional DRAM. The standout structural signal is 128GB-plus high-density DDR5 modules doubling QoQ for the second consecutive quarter, a clean read on HPC-related DRAM demand compounding. DRAM inventory is described as extremely low, with DDR5 shipping straight from production to customers.

Assessment: This is the quarter conventional DRAM pricing joined HBM as a profit driver. With HBM consuming capacity (each HBM bit needs far more wafer than conventional DRAM), the conventional pool is structurally short, and spot prices are reportedly rising daily. The whole DRAM complex is now tight, which broadens and de-risks the earnings base beyond HBM alone.

NAND — The Soft Spot Recovered

NAND bit shipments fell mid-single-digit QoQ off a high Q2 base, but the important number is ASP, which rose low-10% QoQ on a genuine price recovery plus a richer enterprise-SSD mix (eSSD shipments grew double-digit). Management framed eSSD demand as a structural shift driven by AI: RAG and vector databases require fast eSSD search, KV-cache offload from HBM to SSD is rising, and an HDD shortage is pushing hyperscalers toward high-capacity TLC eSSD.

Assessment: A meaningful positive versus our Q2 read, where NAND was the lower-quality leg on falling ASP. The pricing recovery plus the structural eSSD narrative (RAG, KV-cache, HDD substitution) elevates NAND from drag to contributor. We still weight NAND below DRAM/HBM, but the direction has reversed and the structural case is now credible.

HBM — Sold Out, and HBM4 Ships

HBM remains the franchise core, and the news is operational follow-through: management completed 2026 supply discussions with key customers, HBM4 finished development and entered mass production in September at industry-leading speed, and shipments begin in Q4 2025 with full-scale expansion in 2026. Management says HBM has been sold out since 2023, that pricing sustains current profitability, and that supply will remain tight versus demand into 2027. Custom HBM begins scaling from HBM4E, deepening multi-year customer lock-in.

Assessment: The strongest part of an already-strong quarter. A leader that has sold out HBM through 2026 at profitability-sustaining prices, shipped HBM4 first, and sees supply tight into 2027 has converted a cyclical product into something closer to a contracted, multi-year revenue stream. This is the structural spine of the Outperform.

Key Topics & Management Commentary

Overall management tone: Confident and, for the first time, explicitly making the structural-cycle argument with receipts. Management directly addressed the "is this just another boom-bust" question, contrasted the current cycle with 2017-18, and pointed to contracted 2026 demand, LTAs on conventional memory, and an OpenAI LOI as evidence that the demand is durable rather than speculative.

1. The First ₩10 Trillion Operating-Profit Quarter

"Operating profit reached KRW 11.4 trillion… This marks the first time in the company's history that quarterly operating profit has exceeded KRW 10 trillion." — Kim Woo-hyun, CFO

A symbolic and substantive milestone: operating profit crossed ₩10tn for the first time at a record 47% margin, on strong sales of leading-edge product (HBM, high-performance DRAM, enterprise SSD). The margin expanded six points sequentially even as revenue grew only ten percent, the arithmetic of a pricing-led quarter.

Assessment: The milestone matters because of how it was earned: margin, not volume. That is the higher-quality way to grow profit and the one that compounds if pricing holds.

2. 2026 Is Sold Out, Across HBM, DRAM and NAND

"Given the customer's demand and the company's capacity, for next year, not only HBM but DRAM and NAND capacity has essentially been sold out." — SK hynix management, Q3 2025 earnings call

This is the most consequential disclosure of the quarter. Management completed 2026 HBM supply discussions with key customers and has secured full customer demand for its entire DRAM and NAND production next year. Conventional-memory customers are now signing LTAs and issuing pre-purchase POs for 2026, a behavior previously confined to HBM.

Assessment: For a company in a historically volatile industry, contracting essentially all of next year's output is an extraordinary de-risking of the forward. It is the single biggest reason we maintain Outperform despite the share-price run: the earnings base for the next several quarters is now largely visible and contracted, which is precisely what a memory multiple does not normally get to assume.

3. HBM4 Complete, Shipping in Q4

"Our HBM4, which we have completed development and mass-production preparation in September… supports highest speed in the industry. We will start HBM4 shipments in Q4 this year." — Kim Woo-hyun, CFO

HBM4 reached development completion and mass-production readiness in September, with industry-leading speed, and shipments begin in Q4. Management noted HBM4's I/O count is fixed at 2048 (double HBM3E), with customers now competing on speed, and that SK hynix sampled the upgraded, above-JEDEC specifications faster than the industry.

Assessment: First-mover HBM4 into a sold-out 2026 is the cleanest possible evidence the technology lead is intact and converting to revenue on schedule. The HBM4-to-HBM4E custom transition deepens the moat further by co-developing with customers from the GPU/ASIC design stage.

4. The Super-Cycle Argument: "Different From 2017-18"

"The company sees this cycle to be a bit different from the super cycles that we witnessed in 2017 and 2018. The biggest difference is that the current demand is driven by a much broader range of applications coming from the shift to the AI paradigm." — SK hynix management, Q3 2025 earnings call

Asked directly whether this boom will bust like prior cycles, management made the structural case: AI demand is additive on top of existing applications and spawns new ones (autonomous driving, robotics); inference is driving general-purpose server demand (total server set shipments seen growing high-10% next year); and on the supply side, HBM's growing capacity share structurally constrains DRAM bit growth, which management argues turns this into a "long-drawn-out memory super cycle."

Assessment: The "this time is different" claim is the one to scrutinize, because it always precedes a top. But the supply-side half is genuinely novel: HBM cannibalizing conventional DRAM capacity is a real, structural constraint that did not exist in prior cycles, and it is corroborated by sold-out conventional memory and rising spot prices. We give the argument more credence than the usual peak-cycle optimism, while keeping the cyclical tail risk firmly in view.

5. Conventional DRAM Margins Approaching HBM

"The recent hike in DRAM price has narrowed the profitability gap between HBM and DRAM, but for the company, HBM's profitability remains high… the company does not plan to immediately adjust the capacity mix based on what can be a short-lived change in profitability." — SK hynix management, Q3 2025 earnings call

With DRAM spot prices rising daily and DDR4/DDR5 in shortage, conventional DRAM profitability is approaching HBM's. Management said it will not chase the short-term spread by shifting capacity away from HBM, because HBM requires committed long-term volume and strategic customer relationships.

Assessment: The right discipline. Protecting HBM commitments over a possibly transient conventional-DRAM margin spike preserves the multi-year franchise. That it is even a question, conventional DRAM margins rivaling HBM, is itself a measure of how tight the whole complex has become.

6. CapEx Rises Into Contracted Demand

"To respond to such surging demand, CapEx growth across the memory industry appears to be inevitable. And for the company, CapEx next year will far outpace the level of this year." — SK hynix management, Q3 2025 earnings call

2026 CapEx will rise materially, with M15X equipment installation underway (after an early cleanroom opening) to ramp HBM, accelerated tech migration for conventional DRAM and NAND, and Yongin Fab 1 plus the Indiana advanced-packaging plant lifting infrastructure spend. Management repeatedly stressed CapEx discipline and a market-aligned strategy.

Assessment: A rising CapEx cycle is the perennial memory risk, the mechanism by which booms seed the next oversupply. But this one is being poured into demand that is already contracted for 2026, which is a higher-quality basis than a speculative build. The watch item is 2027 and beyond, when this capacity arrives.

7. The Balance Sheet Turns to Net Cash

"We have achieved net cash position in Q3… the best use of cash is to reinvest it into our business while maintaining CapEx discipline." — SK hynix management, Q3 2025 earnings call

Cash rose ₩10.9tn QoQ to ₩27.9tn, and the company turned to a ₩3.8tn net cash position. On capital return, management noted this is the first year of a three-year shareholder-return policy and that, given the surging CapEx needs and high return on investment, the best current use of cash is reinvestment. No additional return is planned at this time.

Assessment: A net-cash memory maker self-funding a CapEx step-up while record cash builds is a different financial animal than in prior cycles. The decision to reinvest over returning cash is rational given the contracted demand and the ROI on capacity. We would expect the capital-return conversation to reopen once the 2026-2027 build is funded.

8. Customer Broadening and the OpenAI LOI

"I will point to our recent LOI with OpenAI for large-scale DRAM supply as an example of the very strong market demand for AI, as well as the need to secure AI memory… HBM market… will keep growing at an average of over 30% for the next five years." — SK hynix management, Q3 2025 earnings call

Management cited a 5-year HBM CAGR of over 30% (on a conservative estimate) and pointed to a recent LOI with OpenAI for large-scale DRAM supply as evidence of demand breadth, alongside its position as primary supplier for both GPU and ASIC customers.

Assessment: The OpenAI LOI and the GPU-plus-ASIC supplier breadth are evidence that demand is diversifying beyond a single accelerator vendor, which reduces customer-concentration risk and supports the durability of the franchise. A 30%-plus five-year HBM CAGR, if it holds, underwrites years of growth.

9. Inventory: Extremely Low

"DRAM inventory… remains extremely low, so much so that in the case of DDR5, the products must be shipped to customers straight out of production." — SK hynix management, Q3 2025 earnings call

Both company and customer inventories fell QoQ. Server-customer memory inventory is notably low, and SK hynix's own DRAM inventory is so lean that DDR5 ships directly from the line. This addresses the central bear worry from Q2 (that pull-forward had built excess inventory): it did not.

Assessment: Lean inventory on both sides is the strongest rebuttal to the inventory-correction bear case and a key reason the Q2 pull-forward fear did not materialize. It also means there is little buffer to absorb any sudden demand air-pocket, which cuts both ways.

10. The 2026 Demand Outlook Steps Up

"DRAM demand growth is expected to rise from high 10% this year to over 20% next year, while NAND device growth is projected to improve from mid 10% this year to high 10% in 2026." — Kim Woo-hyun, CFO

Management raised its bit-demand outlook for 2026: DRAM accelerating from high-teens to over 20%, NAND from mid-teens to high-teens, driven by AI inference broadening memory demand from HBM into high-performance DDR5 and eSSD.

Assessment: An accelerating 2026 demand outlook, paired with sold-out supply, is the combination that supports maintaining a constructive rating through a sharp price run. The structural inference-driven broadening (KV-cache, RAG, eSSD) is corroborated by hyperscaler capex commentary.

Guidance & Outlook

SK hynix guides bit shipments and frames pricing qualitatively. For Q4 2025:

The low-single-digit Q4 bit guide is a supply story, not a demand story: with inventories extremely low and DDR5 shipping straight from production, near-term shipment growth is capped by capacity, not orders. The 2026 framework is what matters, demand accelerating (DRAM >20%, NAND high-teens) into sold-out supply, with HBM4 ramping. Product roadmap: 1c-nm to exceed half of Korea conventional DRAM capacity by end-2026; 321-layer to exceed half of NAND bit production by end-2026.

Analyst Q&A Highlights

The 2026 HBM Contract

The opening question pressed for detail on the just-completed 2026 HBM supply agreements. Management confirmed volume and mix are finalized for major customers at profitability-sustaining prices, and that supply stays tight into 2027.

Q: "It was mentioned that the HBM supply negotiations for 2026 have been completed. Could you share more details about the contract?"
— Kim Sun-woo, Meritz Securities

A: "Our discussions over major issues with our clients have been completed, and the HBM supply plan for next year for major customers has been finalized… the pricing… has also been formed at a level that can sustain the current profitability… The company's HBM supply will remain tight compared to demand into 2027."
— SK hynix management

Assessment: Finalized 2026 volume and mix at profitability-sustaining prices, tight into 2027, is the clearest possible forward de-risking. This answer is the backbone of the maintained Outperform.

Is This Another Boom That Will Bust?

An analyst asked whether the current boom resembles prior memory cycles that ended in downturns. Management made its structural-cycle case.

Q: "Usually if there is a boom in the memory cycle, then it will be followed by a downturn… whether the company sees any similarities with the recent memory boom with the historic cycles in the past."
— Woo Dong-jae, Bank of America

A: "The company sees this cycle to be a bit different from the super cycles that we witnessed in 2017 and 2018. The biggest difference is that the current demand is driven by a much broader range of applications… production can only grow so much… because of the growing share of HBM. These circumstances create a structural constraint against supply increase."
— SK hynix management

Assessment: We scrutinize the demand-side "different this time" claim, but the supply-side argument, HBM structurally constraining conventional DRAM bit growth, is novel and real. It is the most credible version of the structural-cycle case we have heard.

The Structural Case for eSSD

An analyst asked for the rationale behind calling the eSSD demand surge structural. Management tied it to RAG, vector databases, and KV-cache offload.

Q: "The recently strong demand for eSSD was said to be a structural change following the advent of the AI era. So could you elaborate more on the rationale behind such assessment?"
— Ryu Young-ho, NH Investment & Securities

A: "To apply RAG on LLM, we need… vector databases. And this is where eSSD becomes a must… By offloading key value cache that was processed at the GPU all the way to the SSD… what we are seeing now is the benefits of AI infrastructure spreading from DRAM to NAND as well."
— SK hynix management

Assessment: A coherent, technically grounded structural narrative for NAND that did not exist a quarter ago. It elevates NAND from cyclical drag to AI beneficiary, though we wait to see it sustain through pricing cycles before fully re-rating the segment.

The Order-First Memory Market

An analyst asked whether AI is turning memory into a specialty, order-first market and how that changes customer dynamics. Management said conventional memory is now seeing HBM-style long-term contracting.

Q: "The memory market appears to be changing into a specialty market with order first, sell later approach thanks to AI… does the company see any differences from the past in your discussions with customers?"
— Chae Min-sook, Korea Investment & Securities

A: "We are seeing an increase in customers who want to sign long-term agreements for conventional memory products as well. Some customers are very much actively responding… by issuing pre-purchase POs for 2026… for next year, not only HBM but DRAM and NAND capacity has essentially been sold out."
— SK hynix management

Assessment: The spread of HBM-style LTAs to conventional memory is the structural change that could genuinely dampen memory's historical volatility. If it persists through a down-cycle, the appropriate multiple for the whole business rises. This is the most important multi-year watch item.

Could Conventional DRAM Pull Capacity From HBM?

With DRAM spot prices spiking, an analyst asked whether conventional DRAM margins approaching HBM's might prompt a capacity-mix shift. Management declined to chase the spread.

Q: "If the DRAM price maintains the current trend, it appears likely to reach the kind of profit margin similar to HBM… is it also foreseeable to shift the capacity mix, perhaps a bit away from the HBM to the conventional DRAM?"
— Jay Kwon, JPMorgan

A: "The recent hike in DRAM price has narrowed the profitability gap… but… HBM's profitability remains high… the company does not plan to immediately adjust the capacity mix based on what can be a short-lived change in profitability."
— SK hynix management

Assessment: Discipline over opportunism. Honoring HBM commitments rather than chasing a transient conventional-DRAM spread protects the multi-year franchise. The fact that the spread exists at all signals an extraordinarily tight conventional market.

2026 CapEx and the Fab Pull-Ins

An analyst asked the magnitude of the 2026 CapEx increase and whether M15X and Yongin schedules can be pulled forward. Management confirmed CapEx will "far outpace" 2025 and that schedules are being accelerated.

Q: "It was mentioned that the company's CapEx in 2026 will increase over this year. Then what will be the extent of the increase…? Could you pull in the schedule… for Yongin Fab 1 cleanroom readiness?"
— Ricky Seo, HSBC / Nicolas Gaudois, UBS

A: "CapEx next year will far outpace the level of this year… M15X… opened early… we are working to pull up the schedule [for Yongin Fab 1] in light of the pace of demand growth… even with growing CapEx, the company will stick to its CapEx discipline."
— SK hynix management

Assessment: Pulling fab schedules forward is a high-conviction signal, management is racing to add capacity into demand it has already contracted. The discipline caveat is appropriate; the 2027 supply picture this build creates is the key out-year risk to monitor.

Shareholder Returns vs. Reinvestment

With the move to net cash, an analyst asked whether the shareholder-return policy might change. Management said reinvestment is the priority for now.

Q: "The company has turned around to net cash position this quarter… can we expect any changes to the shareholder return policy?"
— Peter Lee, Citigroup

A: "Being in the first year of the new shareholder return policy… we are not looking into additional shareholder return at this time. But we will keep looking into how we can maximize shareholder return… the best use of cash is to reinvest it into our business while maintaining CapEx discipline."
— SK hynix management

Assessment: Rational capital allocation given contracted demand and high ROI on capacity. We would expect buybacks/dividends to re-enter the conversation once the 2026-2027 build is funded; for now, reinvestment compounds the franchise.

Market Reaction

All price levels are FMP-verified KRX closing prices; the benchmark is the KOSPI (^KS11). SK hynix discloses before the Korean market open, so the reaction is the print-day session.

• Pre-print close (2025-10-27): ₩535,000, an all-time high.
• Reaction (2025-10-28) close: ₩521,000, −2.6% on the day, while the KOSPI fell −0.8%. The stock opened down 2.6% and stayed soft.
• Volume: ~5.7M shares versus a ~4.1M 30-day average (1.4x), elevated, consistent with profit-taking.
• Setup: Into the print the shares were up ~+207% YTD (from a 2024 year-end close near ₩173,900) and +59% in the prior 30 days, a parabolic run.

A 2.6% decline on a record, milestone quarter, underperforming an already-down KOSPI, is the signature of a stock that has run too far, too fast. After tripling YTD and adding 59% in a single month, the market had priced perfection, and a quarter that was merely excellent (with a slight revenue miss and a non-operating-flattered net) gave momentum holders a reason to take profits. This is distribution into strength, not a fundamental crack.

For our rating, the reaction is a useful discipline check rather than a thesis signal. The business could hardly be executing better, but the price had gotten ahead of even this result, which is exactly why we frame the maintained Outperform with a "trim into strength" caveat: the fundamentals justify ownership, but the risk/reward at ₩521,000 is a fraction of what it was at our ₩268,500 initiation.

Street Perspective

Debate: Does "Sold Out 2026" De-Risk the Cycle, or Just Delay the Bust?

Bull view: Contracting all of 2026's DRAM, NAND and HBM output, with conventional memory now on LTAs and pre-purchase POs, gives unprecedented forward visibility and dampens the volatility that has always capped memory multiples. The supply side is structurally constrained by HBM. This is a genuinely different, longer cycle.

Bear view: "Sold out" and "this time is different" are the exact phrases heard at every memory peak. Contracts can be renegotiated or cancelled if AI capex disappoints; the 2026-2027 CapEx build the whole industry is pouring will arrive into any demand air-pocket and break pricing, as it always has.

Our take: Lean bull on the forward, cautious on the out-years. The 2026 contracting genuinely de-risks the next several quarters in a way prior cycles never did, and the HBM supply constraint is real. But we hold the cyclical tail risk for 2027+ as the CapEx wave lands, and treat the durability of conventional-memory LTAs through a future down-cycle as the unproven crux.

Debate: Is the Stock Still Investable After Tripling YTD?

Bull view: Earnings are compounding as fast as the price, operating profit nearly doubled YoY with margins still expanding, so the forward multiple has not run away. With 2026 sold out and demand accelerating, forward EPS estimates are rising faster than the stock. A −2.6% pullback is an entry.

Bear view: +207% YTD and +59% in 30 days is a parabola, and the sell-the-news reaction marks the point where good news stops working. The non-operating-flattered net makes the trailing multiple look cheaper than reality, and memory always looks cheapest at the top.

Our take: The bull has the better of the forward-multiple argument, but the bear is right that the easy money is made. We maintain Outperform because the contracted, accelerating 2026 earnings keep the forward valuation reasonable, but we explicitly flag the compressed asymmetry and would scale exposure rather than add aggressively here.

Debate: Is the HBM Moat Durable as Samsung and Micron Push HBM4?

Bull view: SK hynix shipped HBM4 first at industry-leading speed, sold out 2026, and is moving to co-developed custom HBM from HBM4E, which deepens multi-year lock-in. The lead has held every generation since HBM2E.

Bear view: The HBM profit pool guarantees Samsung and Micron spend whatever it takes; an HBM4 qualification win at a key accelerator customer would erode the scarcity premium that underpins the margin structure.

Our take: The lead is real and, on this call, widening (first HBM4, custom from HBM4E). Competitor HBM4 qualification remains the single highest-stakes catalyst to monitor, but nothing this quarter suggested the gap is closing.

Thesis Scorecard: 2025_Q2 Signposts Revisited

Our Q2 2025 initiation set out specific signposts for the second half. They graded decisively bullish, with the one near-term bear risk (pull-forward payback) failing to materialize.

Scorecard summary: Five signposts strongly bullish or bullish, one bullish-with-an-out-year-caveat (CapEx), and one mixed (net-income quality, where the operating result is clean but the headline net is non-operating-flattered). The single most important grade is the first: the pull-forward payback we feared did not arrive, and demand instead accelerated into sold-out 2026. The Q2 thesis played out about as well as it could have.

Bottom Line: Maintaining Outperform

Rating decision: We maintain Outperform on SK hynix. This is the second consecutive Outperform in our coverage, and unlike a rating held by inertia, Q3 actively confirmed and extended the thesis: the first ₩10tn+ operating-profit quarter at a record 47% margin, NAND pricing recovered, the balance sheet at net cash, HBM4 first to ship, and, decisively, 2026 demand sold out across the entire product line at profitability-sustaining prices.

The honest counterweight, and the reason we frame this as "maintain and trim" rather than "add": the shares have tripled year-to-date, ran 59% in the month into the print, sold off 2.6% on the result, and the headline net is flattered by a ₩3.3tn non-operating gain. The asymmetry that made the July initiation compelling at ₩268,500 has narrowed sharply at ₩521,000. We stay Outperform because the contracted, accelerating 2026 earnings keep the forward valuation reasonable and the franchise is executing flawlessly, but we would use strength to scale rather than chase.

What would move us to Hold: evidence that the AI-capex cycle is slowing and 2026 contracts are at risk of renegotiation; a stalling HBM4 ramp or a competitor qualification win; or the multiple running meaningfully ahead of even the now-visible 2026 earnings.

What would move us to Underperform: a genuine demand reversal that breaks the sold-out 2026 framework; signs the industry CapEx wave is creating 2027 oversupply faster than demand absorbs it; or a Samsung/Micron HBM4 breakthrough that contests SK hynix's scarcity premium.

Signposts for Q4 2025 earnings (late January 2026):