Goldman Sachs Details Korea Memory "8 Key Focus Points": Valuation, Long-Term Agreements, Inventory, CXMT Impact, Buybacks, etc.

marsbitPublicado a 2026-08-05Actualizado a 2026-08-05

Resumen

Goldman Sachs maintains a "Buy" rating on Samsung Electronics and SK Hynix, stating their recent sharp share price corrections reflect excessive market pessimism unsupported by fundamentals. The report addresses eight key market concerns: 1. **HBM Pricing:** GS forecasts HBM blended average prices for Samsung and SK Hynix to nearly double by 2027 due to persistent tight supply and a need to re-establish price premiums over standard DRAM. 2. **Long-Term Agreements (LTAs):** LTA terms are shifting in favor of suppliers, featuring longer terms, higher coverage targets (60-70% of capacity), "floor-price" protections, and significant upfront customer deposits. 3. **Inventory:** While module maker inventory is elevated, this represents a small portion of the overall market. Supplier and end-customer inventories remain healthy at low levels. 4. **NAND Supply:** Fears of a NAND oversupply are overblown. Strong enterprise SSD demand is expected to offset consumer weakness, with the supply-demand gap widening through 2027. 5. **Shareholder Returns:** GS expects both companies' actual dividends to exceed current market consensus. Share buybacks would be well-received, especially for SK Hynix to counter ADR-related dilution. 6. **SK Hynix ADR Premium:** The ADR's ~30% premium over the local share is likely persistent due to conversion constraints but aids long-term valuation re-rating by improving global investor access. 7. **SK Hynix Q2 '26 Results:** The earnings miss was due t...

Author: ZF Trading Platform, Wall Street Insights

Samsung Electronics and SK Hynix stock prices have experienced significant corrections in recent months, with valuations falling to extremely pessimistic levels. However, Goldman Sachs believes the fundamentals do not support such low valuations and reiterates its Buy ratings on both companies.

ZF Trading Platform news, August 4th - The Goldman Sachs team led by Giuni Lee published a research report systematically outlining the market's current eight core concerns regarding the Korean memory industry, covering HBM pricing outlook, long-term agreement structure, inventory status, CXMT impact, shareholder returns, and the impact of SK Hynix's US ADR, among other topics.

The analysts believe most of these concerns are over-interpreted by the market, and the actual supply-demand dynamics still support elevated memory prices.

Against this backdrop, Samsung Electronics and SK Hynix stock prices have fallen 23% and 35% respectively over the past month, causing the two companies' forward 2027 P/E ratios to drop to around 3.5-3.6x and their P/B ratios to only 1.4-1.6x.

Goldman Sachs points out that this valuation level implies extreme market distrust in the sustainability of the two companies' earnings, which significantly deviates from their actual fundamental conditions.

Focus Point One: 2027 HBM Pricing May Double, Goldman's Forecast Far Exceeds Market Consensus

Goldman Sachs expects the blended average selling price (ASP) of HBM for Samsung Electronics and SK Hynix to rise approximately 87% and 100% year-over-year respectively in 2027, both nearing $2.9 per Gb. Within this, similar product price increases account for about 60%, with the remaining increment coming from product mix improvements.

The core logic supporting this view is continuously tight supply and demand.

The report notes that AI server-driven HBM demand continues to outpace supply, while yields for the latest generation HBM have significantly declined due to more advanced process nodes and higher stack counts. Coupled with higher conversion ratios from standard DRAM to HBM, the difficulty of supply expansion is further increased.

Goldman Sachs expects the HBM supply-demand gap to be tighter in 2027 than this year.

Another key factor is the significant price differential between HBM and standard DRAM.

As of Q2 2026, because standard DRAM contracts are negotiated on a monthly or quarterly basis and can reflect market dynamics faster, their pricing has surpassed that of HBM, which is primarily based on annual fixed contracts, forming a clear inversion.

Goldman Sachs expects the standard DRAM ASP to rise from about $0.5-$0.6 per Gb at the end of 2025 to about $2 per Gb by year-end 2026, at which point HBM will inevitably re-establish its price premium and move towards the operating profit margin level of standard DRAM.

Goldman's forecast for SK Hynix's HBM ASP is approximately $2.9 per Gb, about 24% higher than the Bloomberg market consensus estimate.

Based on this calculation, the proportion of HBM revenue in Samsung and Hynix's total DRAM revenue will increase from about 8% and 14% this year to 16% and 22% respectively in 2027, and further to 18% and 25% in 2028.

Focus Point Two: Long-Term Agreement Terms Favor Suppliers More

As expectations for long-term tightness in memory supply strengthen, both supply and demand sides are actively promoting the signing of Long-Term Agreements (LTAs).

Goldman Sachs believes that, based on disclosed information and channel research, LTA terms are tilting towards suppliers across four dimensions: longer durations, broader coverage, more favorable pricing structures, and stronger binding force.

Regarding duration, most suppliers indicate contracts are primarily 5-year, with some customers at 3-year. Samsung mentioned in its earnings call that its LTAs are typically based on 5-year terms, with a rollover mechanism extending them by one year annually, theoretically allowing contract duration to exceed 5 years.

Regarding coverage, the target is leaping from 50% towards 60%-70%. Specifically:

  • Kioxia has signed contracts with 5 customers, covering about one-third of its 2027 shipments, with a long-term target of 50%;

  • Micron has signed 16 strategic customer agreements, covering about 20% of DRAM shipments and one-third of NAND shipments, with an ultimate target of LTA revenue exceeding 50%;

  • SK Hynix stated it has completed term negotiations for about 10 long-term agreements;

  • Samsung disclosed it has signed contracts with the world's top five data center customers and is in final negotiations with another 5 major customers, expecting that post-signing, multi-year contracted volume will reach about 60%-70% of planned capacity.

Regarding pricing structure, evolution is towards "price bands" and "floor price protection" mechanisms.

Micron clearly stated its largest contracts have price ceilings and floors, based on Q2 2026 market prices. Even at the floor price, gross margins remain above historical peak levels.

Samsung said it will employ different pricing models based on customer groups and product categories, and has set floor prices for commodity products to hedge against market price volatility risks.

Regarding binding force, the prepayment mechanism is the biggest highlight differentiating this cycle's LTA terms from past cycles:

  • Kioxia disclosed financial guarantees (including prepayments) exceed $11 billion;

  • Micron expects to receive $22 billion in cash deposits and related financial commitments;

  • Samsung stated contracts include substantial prepayments in the form of deposits, having received about one-quarter of total contract prepayments so far, with further increases expected as more contracts are finalized.

Focus Point Three: High Module Maker Inventory Does Not Represent Industry-Wide Risk

Recent market concerns have risen regarding high inventory levels at memory module makers.

Goldman Sachs acknowledges module maker inventory has indeed increased, especially against a backdrop of weaker consumer demand from smartphones, PCs, etc. However, the key point is that the module maker market accounts for only single-digit percentages of the overall memory market, thus its substantive impact on industry fundamentals is limited, with more impact being on sentiment.

Looking at the more critical supplier and end-customer levels, inventory conditions are healthy.

As of the end of Q2 2026, Goldman estimates Samsung and Hynix's DRAM and NAND inventory are within 2-4 weeks, below the normal level of about 4-5 weeks, and far below the 10+ weeks commonly seen before previous downturns.

Given that supply growth is expected to remain below demand growth for the next 12-18 months, this low inventory state is projected to continue.

For end-customers (especially server customers), despite active procurement over the past few quarters, inventory levels will remain within the normal range as procured products are largely used immediately for production.

Focus Point Four: NAND Supply-Demand Will Not Reverse, Server Demand Sufficient to Offset Consumer Weakness

Recent market concerns about NAND oversupply have intensified, with some bears citing falling NAND spot prices as evidence. Goldman Sachs holds a different view.

From a supply-demand perspective, Goldman expects the NAND supply-demand gap in 2027 to widen further compared to this year. The main reason is that major suppliers are focusing capital expenditure on DRAM, with NAND expansion focused more on process upgrades rather than wafer capacity expansion. Supply growth is expected to remain below demand growth over the medium term.

From a demand structure perspective, Goldman estimates enterprise SSD demand will increase from 474EB to 755EB from 2026 to 2028, with year-over-year growth rates of 66%, 31%, and 22% respectively.

Despite some weakness in consumer demand, Goldman's channel research indicates enterprise SSD demand still has upside potential, sufficient to offset consumer-side pressure.

Regarding recent spot price weakness, Goldman points out the decline is mainly concentrated in the specific TLC 512Gb product, while other specs like TLC 1Tb remain stable.

Notably, the TLC 512Gb price has risen nearly 600% over the past year, significantly outpacing the 400%+ gains for most other products. The current pullback is essentially a normal correction after its prior strong outperformance.

Focus Point Five: Goldman Expects Actual Returns Will Exceed Market Expectations

Korean memory makers' failure to provide explicit details on specific return plans during recent earnings calls has disappointed some investors.

Goldman notes that on August 3rd, after Japanese memory maker Kioxia announced its shareholder return plan, its stock rose 6% in a single day, while Samsung and Hynix shares both fell 9% on the same day. Goldman believes at least part of the divergence stems from gaps in shareholder return expectations.

Nevertheless, Goldman points out that both Samsung and Hynix clearly stated during their earnings calls that they are actively reviewing various shareholder return options.

Samsung's current three-year shareholder return policy expires this year, with a commitment to return 50% of its three-year free cash flow to shareholders.

Goldman believes the current Bloomberg consensus dividend per share estimate of 8,638 KRW has room for upward revision and updates its own forecast to 9,500 KRW. SK Hynix's three-year policy covers 2025-2027, and Goldman similarly expects actual dividends will exceed market consensus.

In addition to increased dividends, Goldman notes that buyback announcements would be strongly welcomed by the market, especially given recent sharp stock price declines. For Hynix, as ADR listing leads to share dilution, buying back and canceling shares could be an effective means to hedge dilution effects.

Focus Point Six: SK Hynix ADR Premium Unlikely to Disappear Short-Term, But Helps Repair Historical Discount

SK Hynix completed its US ADR listing on July 10th. Since then, the ADR has consistently traded at a premium to the domestic stock, averaging around 26%, with the current premium about 30%.

Simultaneously, Hynix's domestic stock's 12-month forward P/E ratio remains about 41% lower than Micron's and about 30% lower than Hynix's ADR.

Goldman attributes this premium/discount difference to two points:

  • First, procedural restrictions exist for conversion between ADRs and domestic shares, leading to investor group differentiation;

  • Second, the ADR issuance size is extremely limited, representing only about 2.4% of total shares outstanding.

Hynix has stated that ADRs can be freely converted into domestic shares, but converting domestic shares into ADRs is subject to conversion caps and involves lengthy regulatory filing procedures taking weeks or longer.

SK Hynix Chairman Chey Tae-won has expressed openness to issuing more ADRs. Nonetheless, referring to the precedent of TSMC's ADR maintaining a long-term premium, Goldman believes that as long as the two-way conversion mechanism is not substantially improved, Hynix's ADR premium relative to its domestic stock will persist.

However, in the long run, the ADR listing provides global institutional investors with direct access, helping Hynix gradually narrow its historical valuation discount versus international peers.

Focus Point Seven: SK Hynix's Q2 Earnings Miss Was a One-Time Factor, Q3 Expected to Recover Strongly

SK Hynix reported Q2 2026 revenue of 79.3 trillion KRW and operating profit of 60.5 trillion KRW. Operating profit was largely in line with Goldman's forecast of 59.1 trillion KRW, but about 7% below the Bloomberg market consensus estimate of 65 trillion KRW.

Goldman believes the primary reason for missing market consensus was DRAM ASP performance falling short of expectations, with actual sequential growth of about 29%, below Goldman's prior forecast of 39%. Within this, standard DRAM ASPs began reflecting previously locked-in contract prices with customers, while HBM ASPs were below expectations due to limited progress in the product mix shift towards HBM4.

Looking ahead to Q3, Goldman expects DRAM shipment volume to grow sequentially by about 10%, with ASP growing sequentially by about 19%, mainly benefiting from HBM4 volume ramp-up and mix improvement from 1c nm DRAM expansion. This corresponds to an operating profit forecast of about 77 trillion KRW, roughly in line with market consensus.

Goldman also notes that compared to some peers who have already locked in contracts with price ceilings, Hynix has greater exposure to standard DRAM price elasticity. If prices exceed expectations, the company's upside potential could be more significant.

Focus Point Eight: CXMT's Impact is Limited to the Chinese Domestic Market

Following CXMT's (ChangXin Memory Technologies) completion of its IPO, investor concerns about Chinese memory makers impacting the global supply-demand landscape have increased.

Goldman believes CXMT's expansion will primarily focus on meeting domestic demand, with limited substantive impact on the tight global supply-demand situation.

From a technology gap perspective, Goldman cites TrendForce data indicating CXMT's current mainstream process is equivalent to the 1z node, while Samsung and Hynix are in transition from 1a/1b to the 1c node.

From a product structure perspective, about 70% of CXMT's mobile DRAM shipments are LPDDR4(X), while Samsung and Hynix's mobile DRAM product mix already has LPDDR5(X) accounting for 75%-85%, indicating a clear product positioning misalignment.

~~~~~~~~~~~~~~~~~~~~~~~~

Preguntas relacionadas

QAccording to Goldman Sachs' report, what are the 2027 forward P/E ratios for Samsung Electronics and SK Hynix, and what does this valuation level imply about market sentiment?

AAccording to the report, the 2027 forward P/E ratios for both Samsung Electronics and SK Hynix are about 3.5 to 3.6 times, with P/B ratios at 1.4 to 1.6 times. Goldman Sachs indicates that this valuation level implies extreme market distrust in the sustainability of the two companies' earnings, which is a significant deviation from their actual fundamental conditions.

QWhat is Goldman Sachs' core logic for predicting that HBM ASP will double by 2027, and how does it compare to the Bloomberg consensus?

AGoldman Sachs' core logic is based on persistently tight supply and demand. AI server-driven HBM demand continues to outpace supply, while yields for the latest HBM generations have declined due to more advanced nodes and higher stack counts. Additionally, the significant price gap between HBM and conventional DRAM (which is currently inverted) is expected to normalize, with HBM re-establishing a premium. Goldman Sachs forecasts SK Hynix's HBM ASP to reach about $2.9 per Gb by 2027, which is approximately 24% higher than the Bloomberg market consensus.

QHow are the terms of Long-Term Agreements (LTAs) in the memory market evolving to favor suppliers, according to the report?

AAccording to the report, LTA terms are tilting in favor of suppliers across four dimensions: longer duration (typically 5-year, some 3-year), broader coverage (targets moving from 50% towards 60-70% of output), more favorable pricing structures (featuring price floors/ceilings and 'floor price protection'), and stronger enforceability, highlighted by substantial prepayment mechanisms from customers that differentiate this cycle from previous ones.

QWhat is Goldman Sachs' view on the impact of high inventory levels at memory module makers and the overall inventory health of the supply chain?

AGoldman Sachs acknowledges that module maker inventory has risen, particularly due to weak consumer demand. However, they argue that the module market only constitutes a single-digit percentage of the overall memory market, so its impact on industry fundamentals is limited and more of a sentiment shock. Crucially, inventory levels at key suppliers (Samsung, SK Hynix) and end-customers (especially server clients) remain healthy, estimated at below normal levels (2-4 weeks for suppliers) and are expected to stay low due to supply growth lagging demand.

QHow does Goldman Sachs assess the potential impact of CXMT (ChangXin Memory Technologies) on the global memory supply-demand balance?

AGoldman Sachs believes CXMT's expansion will primarily focus on meeting domestic Chinese demand, resulting in a limited substantive impact on the tight global supply-demand balance. This assessment is based on a significant technology gap (CXMT's mainstream node is equivalent to 1z, while Samsung/SK Hynix are transitioning to 1c) and a product mix misalignment (CXMT's mobile DRAM is mostly LPDDR4(X), while the Korean leaders' is 75-85% LPDDR5(X)).

Lecturas Relacionadas

On L1 Value Capture from Two Solana Proposals

The article, "Discussing L1 Value Capture Through Two Solana Proposals," by Max Resnick, explores how Layer 1 (L1) blockchain tokens derive their fundamental value, drawing parallels to traditional asset pricing theory. Resnick argues that L1 token value, like stock value, stems from claims on future income streams for holders, not merely from network activity or technological promise. This value is captured when fees are either burned (economically akin to a buyback) or distributed to stakers (akin to dividends). Inflationary staking rewards, by contrast, redistribute value among holders rather than creating it. The core challenge is the quality and defensibility of fee-based revenue. High-quality fees come from sustainable, recurring demand for the network's economic utility (e.g., long-term financial activity), not from transient speculation (e.g., meme coins, airdrops). The strength of a blockchain's network effects—liquidity, applications, users—can make its revenue more defensible and grant it greater pricing power than often assumed. The article proposes a foundational valuation framework for L1s, separating revenue (fees captured for token holders), costs, and total token supply. A key accounting principle is that inflationary rewards should not be counted as a cost unless the newly minted tokens are symmetrically counted as a value input; otherwise, it misrepresents profitability. Finally, Resnick discusses the economics of increasing protocol fees to boost revenue. Since revenue equals price times quantity, the net effect depends on demand elasticity. Research on Ethereum suggests transaction demand is somewhat elastic; a fee increase reduces volume. A uniform fee is a blunt instrument, as different transactions (e.g., small transfers vs. large settlements) have vastly different abilities to pay. The article suggests that transaction-value-based fees, potentially implemented via token programs, could be a more efficient way to capture value from high-willingness-to-pay activities. The discussion is framed around ongoing Solana proposals (SIMD-550, SIMD-553) but focuses on the universal principles of L1 value accrual.

marsbitHace 44 min(s)

On L1 Value Capture from Two Solana Proposals

marsbitHace 44 min(s)

Meme Coin with $60 Million Market Cap Plunges 65% in One Minute, FOMO Faces Renewed Scrutiny

A Solana-based meme token, $CATE, which had surged from a $20M+ to over $80M market cap in about a week, experienced a dramatic 65% crash within one minute. This flash crash has intensified scrutiny on the trading app 'fomo' and highlighted the speculative nature of the current meme coin market. The crash coincided with two events: the token's X account being suspended and the fomo app experiencing downtime, preventing users from trading. While the X suspension was straightforward, the fomo outage raised significant questions. $CATE's primary narrative driver was the open endorsement by Poorgoat, a top-ranked trader on fomo with over 200,000 followers, who had turned a ~$45,000 investment into over $2M at the peak. The token itself had no novel fundamentals, being a "cat sister" to Doge, a concept already existing on Ethereum without success. The crash, triggered by less than $1.5M in selling volume despite over 60,000 holder addresses, exposed a harsh reality: purely "organic" community-driven meme tokens (excluding past successes like $SPX) may now have a market cap ceiling around $17M, as exemplified by the long-term chart of $neet. This incident has fueled existing controversies surrounding fomo. Critics have grown skeptical of the app, alleging that rankings dominated by KOLs who receive lucrative token airdrops could be manipulated to create "pump-and-dump" schemes, luring in retail users before a rug pull. The timing of the crash during fomo's outage—preventing many of its users (who represent over 60% of $CATE holders) from reacting—was viewed as highly suspicious. Further controversy arose when another popular fomo trader publicly sold near the peak, and concerns were raised about the security of accessing private keys during the app's downtime. Fomo's official explanation of server overload due to surging user traffic was met with skepticism, given its substantial funding. The event serves as a stark reminder of the risks in meme coin speculation and the potential vulnerabilities of relying on a single trading platform during market volatility.

marsbitHace 55 min(s)

Meme Coin with $60 Million Market Cap Plunges 65% in One Minute, FOMO Faces Renewed Scrutiny

marsbitHace 55 min(s)

Trading

Spot
活动图片