Cycle on Aug 17, 2026

I will begin by listing the open threads on this topic to see what questions, investigations, or watches have already been set up in prior cycles.

I will now query the existing notes for this topic using `read_notes` to see what has been documented so far.

I will now perform some initial research on the current state of the memory market (DRAM, HBM, NAND) in August 2026. Let's see what the latest news and commentary is regarding DRAM contract pricing, HBM supply, and recent company developments (Micron, SK Hynix, Samsung, TrendForce). First, let's search for "TrendForce DRAM contract pricing Q3 2026 Q4 2026" or "HBM contract pricing 2027" or "Samsung SK Hynix Micron earnings" or any other recent updates. I will use `web_search` to find news from the last few weeks. Let's do a search on "TrendForce DRAM contract pricing 2026" or "TrendForce HBM contract pricing".

This is incredibly rich data. Let's see: 1. `https://xenospectrum.com/en/memory-spot-price-stall-ai-demand-consumer-gap-2026/` published 4 days ago: "Why Memory Prices Froze at Record Highs: The Spot Market Reveals a Fault Line Between AI Demand and Consumer Market Limits". This article talks about how commodity DRAM contract prices in 3Q26 will rise 13-18% quarter-over-quarter—a sharp deceleration from the 58-63% increase in 2Q26. It mentions that spot average price for DDR4 1Gx8 3200MT/s rose only 0.93% (from $42.11 to $42.50) with low trading volume. "Prices are rising, but goods aren't moving." 2. `https://www.trendforce.com/news/2026/08/12/news-micron-can-meet-less-than-half-of-data-center-demand-as-customers-rush-to-secure-memory-at-very-high-prices/` published 5 days ago: "Micron Meets Less Than Half of Data Center Demand as Customers Rush to Secure Memory at 'Very High' Prices". It says "SCAs Lock In Demand as Price Gains Cool" and notes that "even as these contracts lock in years of demand at prices well above prior cycle peaks, Micron also flagged that it is starting to moderate its price increases to some extent... a detail that, on its face, seems at odds with a company describing 2027 as tighter than 2026." 3. `https://www.trendforce.com/presscenter/news/20260730-13158.html` published July 30, 2026: "Diverging Memory Market Outlook in 2027 as DRAM Supply Remains Tight While NAND Flash Supply Conditions Ease, Says TrendForce". 4. `https://www.trendforce.com/presscenter/news/20260721-13148.html` published July 21, 2026: "NAND Flash Supply Growth to Outpace Demand in 2027, Easing Supply Constraints in 2H27, Says TrendForce". Let's fetch these pages using `web_fetch` to get the actual text and quotes. First, let's fetch `https://xenospectrum.com/en/memory-spot-price-stall-ai-demand-consumer-gap-2026/` as it discusses the spot price stall and the "fault line" between AI demand and consumer market limits. This is a classic late-cycle or cycle-turn indicator. Let's fetch it.

This is a highly insightful and detailed piece. It shows: - Commodity DRAM contract prices in 3Q26 are projected to rise 13-18% QoQ, which is a sharp deceleration from the +58-63% in 2Q26 and +93-98% in 1Q26. - The spot market has gone completely quiet (gridlock), with buyers refusing to chase prices higher and sellers refusing to sell lower. Low trading volume, spot prices moving by less than 1% week-over-week. - Downstream retail channels and consumer market customers have reached the limits of what they can afford to buy. - Meanwhile, the major DRAM manufacturers have shifted their capacity to HBM and server RDIMMs (which take 3x the wafers of conventional DRAM), leaving a structural shortage in mature nodes like DDR4, which has forced manufacturers to withdraw their EOL (end-of-life) plans and even extend production. - Long-term agreements (LTAs) and Strategic Customer Agreements (SCAs) are locking in supply, which might shrink the spot market's role and lead some analysts to claim a structural shift "from a cyclical commodity to a predictable, high-margin business". Let's look at the other TrendForce press releases to see what they say about 2027. Let's fetch `https://www.trendforce.com/presscenter/news/20260730-13158.html` (Diverging Memory Market Outlook in 2027) and `https://www.trendforce.com/presscenter/news/20260721-13148.html` (NAND Flash Supply Growth to Outpace Demand in 2027).

This is very interesting. The NAND Flash sufficiency ratio is expected to turn positive in 2027 (meaning a surplus), while the DRAM sufficiency ratio will remain negative (-1% to -2% in 2026, and the gap is expected to widen in 2027 due to HBM wafer consumption and delayed capacity ramps). Let's see if we have an open thread or watch on NAND Flash supply-demand balance. Yes, we do: - Thread ID `019fa2a2-e1c8-736d-9006-2adb2bb879c4`: "Monitor TrendForce and industry reports for the transition of the NAND Flash market from a supply deficit to a surplus (positive supply-demand balance), projected for 2H 2027. A surplus would mark the definitive turn of the NAND cycle." Wait, the fire date is 2027-09-01, so it hasn't fired yet, but this is a key update confirming the projection. Let's look at the other TrendForce news we found: `https://www.trendforce.com/news/2026/08/12/news-micron-can-meet-less-than-half-of-data-center-demand-as-customers-rush-to-secure-memory-at-very-high-prices/` Let's fetch this. It discusses Micron's ability to meet less than half of data center demand, customers rushing to secure memory at "very high" prices, and Micron starting to "moderate its price increases to some extent." This is a classic late-cycle dynamic where suppliers moderate price hikes to prevent demand destruction, or because long-term agreements cap the increases. Let's fetch it.

This is a brilliant primary source. Let's look at the details: - Micron's executive vice president and chief business officer, Sumit Sadana, spoke at the KeyBanc Technology Leadership Forum 2026 on August 10, 2026. - He said that customers remain eager to secure memory despite "very high" prices, with data center supply being the tightest (Micron can fulfill no more than half of customer demand). - He expects the supply-demand balance to tighten further in 2027, potentially making it even more constrained than 2026. - "Micron said its customers consistently point to DRAM — not power, real estate, data center capacity, or logic wafers — as their number one constraint they are facing today." - SCAs (Strategic Customer Agreements): Micron has signed 16 SCAs, expected to account for roughly half of revenue, carry binding take-or-pay terms, and run mostly through 2030. - "Yet even as these contracts lock in years of demand at prices well above prior cycle peaks, Micron also flagged that it is starting to moderate its price increases to some extent... a detail that, on its face, seems at odds with a company describing 2027 as tighter than 2026." - Pricing structure within SCAs: "Most SCA volume carries a price band with a ceiling, he said, and for the 16 agreements announced at earnings, that ceiling was set using CQ2 pricing. Future SCAs, however, won't be anchored to that same benchmark — Sadana said any new agreements going forward will have ceiling prices 'consistent with whatever the market price is at the time the SCAs are signed.'" - "Made in USA" premium: Micron claims it is the only company investing in front-end memory fab manufacturing in the U.S. (Samsung is logic foundry, SK Hynix is back-end packaging in Indiana). Customers value this and Micron believes it can command a pricing premium for its U.S.-made supply, which has already been factored into the SCAs. Let's do a public markets check for Micron (MU) and other key players. Let's see: `market_view(kind='ticker', id='MU')` and maybe `market_view(kind='ticker', id='Nvidia')` or `market_view(kind='ticker', id='AMD')` or `market_view(kind='ticker', id='ASML')` or `market_view(kind='ticker', id='LRCX')` or `market_view(kind='ticker', id='AMAT')`. Let's call `market_view` for MU first to see what the latest public market data looks like. This is highly relevant since the topic is the AI-driven memory cycle.

This is highly valuable data. Let's look at the numbers for Micron: - Revenue (TTM) is $90.27B (+345.7% YoY) - Gross margin is 72.6% - Operating margin is 80.4% - Profit margin is 55.9% - EPS for the latest quarter (ended May 31, 2026) was $24.89 vs $20.98 est. - Trading at $971.66 with a market cap of $1.03T. - Cash position is $25B, total debt is $6.38B. - Quarterly revenue trajectory: - 2026-05-31: rev $41.46B | op $33.33B | net $28.24B - 2026-02-28: rev $23.86B | op $16.14B | net $13.79B - 2025-11-30: rev $13.64B | op $6.14B | net $5.24B - 2025-08-31: rev $11.31B | op $3.75B | net $3.20B - Capex for the latest quarter (ended May 31, 2026) was $7.83B, and FCF was $17.56B. - News mentions: "DRAM ETF Forms Rare Bullish Pattern as Memory Stocks and Inflows Jump" on 2026-08-16. It says the Roundhill Memory ETF (DRAM) has rebounded significantly, surging nearly 30% from its July low, and is showing a bullish inverted head-and-shoulders pattern. Let's check if there are other companies or sectors we should pull. Let's search for "DRAM ETF Forms Rare Bullish Pattern as Memory Stocks and Inflows Jump" or find details about the Roundhill Memory ETF (DRAM). Or search for "Roundhill Memory ETF" or "DRAM ETF". Wait, let's also pull `market_view` for other equipment players mentioned in the brief: Lam Research (LRCX), Applied Materials (AMAT), ASML, Entegris (ENTG). Let's pull `LRCX` (Lam Research) and `ASML` to see if they are seeing similar massive results. Let's do `market_view` for LRCX and ASML.

Excellent. Let's do `market_view` for ASML too. This is great. Let's pull it.

Excellent. Let's do a quick search on the "Samsung labor situation" and "Samsung yields" or "Samsung HBM qualification" to see what has happened since July. We know Samsung averted a strike in late July with a wage deal, but there was a bonus gap crisis. Let's search for "Samsung HBM qualification 2026" or "Samsung HBM4" or "Samsung labor" in the last month.

Let's look at the incredible details we've gathered: 1. **Samsung achieves the "Golden Yield" for HBM4**: - Reports on August 9–10, 2026 (citing Seoul Economic Daily and other sources) confirm that Samsung's HBM4 production yield has risen to approximately 80%, up from below 60% when mass production began in February 2026. This is months ahead of its original year-end target. - Samsung plans to more than triple HBM4 revenue in Q3 2026, with HBM4 expected to account for over 60% of its total HBM sales in the second half of 2026. Samsung is targeting an HBM market share of around 38% by year-end. - This directly triggers one of our watches/threads! Let's check the watches: - Thread ID `019fc6af-767a-7781-9bcf-50eddacd461d`: "Monitor Samsung's Q3 2026 earnings release (expected late October/early November 2026) to verify if HBM4 sales tripled QoQ and accounted for over 60% of Samsung's total HBM revenue mix as guided." Wait! The watch is for verifying the Q3 2026 earnings release (expected late October/early November 2026). So it has not fired yet, but we have strong early indicators of the yield reaching 80% and the plans being on track. - Let's also check if there is an existing note we can update. - Note `hbm4-race-nvidia-rubin-spec-cuts-samsung-qualification` (The HBM4 Race: NVIDIA Rubin Specification Cuts, Samsung's Qualification, and the 16-Hi Stack Bottleneck). Let's update this note or create a new one. Since it's about the HBM4 race, Samsung's qualification, and yield, reusing `hbm4-race-nvidia-rubin-spec-cuts-samsung-qualification` is perfect! 2. **SK Hynix Labor Dispute**: - SK Hynix is facing a new labor dispute. On August 9, local time, SK Hynix and its union concluded their fifth round of negotiations but failed to reach an agreement on the bonus scheme (paying more than half of this year's Profit Sharing (PS) in treasury shares with a lock-up period). - A preparatory committee for a new union attracted over 3,500 participants within three days of its formation on August 5, representing about 10% of the workforce. - This labor dispute at SK Hynix comes right as Samsung is closing the yield gap, putting pressure on SK Hynix's HBM leadership. 3. **Samsung's Advanced Packaging and Backend Constraints**: - On August 15, 2026, TechTimes reported that Samsung is weighing shipping its legacy memory backend equipment to Vietnam to unlock cleanroom space and capacity for HBM packaging in Korea. - Backend throughput, not wafer starts, is Samsung's binding constraint on HBM scale. Samsung plans a 47% HBM wafer output increase (from 170k to 250k wafers per month by the end of 2026), but needs advanced packaging capacity. Shipping legacy backend lines (like conventional DRAM packaging) to Vietnam frees up space in its domestic fabs (like Cheonan and Onyang) for HBM advanced packaging. 4. **DRAM and NAND Divergence**: - TrendForce's July 30, 2026 report outlines a diverging memory market in 2027: - **DRAM remains tight**: Due to HBM wafer consumption (which takes 3x the wafers of conventional DRAM), process migrations, and delayed fab expansion ramps (Yongin, Fab 5A, etc. won't contribute meaningful output until late 2027/2028). The DRAM sufficiency ratio is estimated at -1% to -2% in 2026, and the gap is expected to widen in 2027. - **NAND Flash transitions to surplus**: NAND Flash sufficiency ratio is expected to turn positive in 2027 as process migration to higher-layer V-NAND (such as Samsung's 9th generation and others) and fab ramps increase bit supply while consumer electronics demand remains weak. - **Spot market freeze**: Commodity DRAM contract prices are decelerating rapidly (projected +13-18% in 3Q26, down from +58-63% in 2Q26 and +93-98% in 1Q26). Spot prices have frozen, moving less than 1% week-over-week on extremely low trading volumes. Buyers are resisting higher prices because downstream consumer markets (PCs, smartphones) cannot absorb them, while sellers are directing capacity to AI servers where long-term agreements (LTAs/SCAs) lock in demand. Let's check if there are other threads we should resolve or watches we should register. Wait! Let's check if we can resolve any threads. None of the threads have fully fired yet because their fire dates are in the future (e.g., FQ4 earnings for Micron is late September, Samsung Q3 earnings is late October, SK Hynix Yongin cleanroom is early 2027, etc.). But wait, let's double check if we can write a note about the spot market freeze and the contract price deceleration. Let's see if there is an existing note we can update. Yes! `dram-contract-pricing-skyrockets-to-record-highs-q1-2026` is titled: "DRAM Contract Pricing Moderates in Q3 2026 as Long-Term Agreements Cap Price Hikes". Let's update this note to include the latest August 2026 spot market freeze data, the 3Q26 deceleration to +13-18% QoQ, and the spot-contract divergence. Let's write a note for `dram-contract-pricing-skyrockets-to-record-highs-q1-2026`. Wait! Is there an existing note on the HBM4 race? Yes: `hbm4-race-nvidia-rubin-spec-cuts-samsung-qualification`. Let's update `hbm4-race-nvidia-rubin-spec-cuts-samsung-qualification` to cover Samsung's 80% yield milestone, SK Hynix's labor dispute, and Samsung's plan to ship legacy backend to Vietnam to free up HBM packaging space. Let's also look at Micron's recent KeyBanc forum comments. Is there an existing note on Micron's SCAs? Yes: `micron-scas-business-model-transformation-cyclicality` (Micron Redefines Memory Cycle with 16 Take-or-Pay Strategic Customer Agreements in Record Q3 2026). We can update this note with Micron's August 10, 2026 KeyBanc forum disclosures (customers scrambling for memory at "very high" prices, meeting less than half of data center demand, SCA price caps set to CQ2 prices, and the "Made in USA" premium). Let's read these three existing notes to make sure we integrate our new findings seamlessly and maintain their structure. Wait, let's fetch the contents of: - `dram-contract-pricing-skyrockets-to-record-highs-q1-2026` - `hbm4-race-nvidia-rubin-spec-cuts-samsung-qualification` - `micron-scas-business-model-transformation-cyclicality` Wait, how do we read them? We can fetch their full bodies by using `read_notes` with a limit, but `read_notes` only returned previews. Let's see if we can query them or if there's a tool to read notes. Ah, `read_notes` lists them, but wait, does it return the full body or just the preview? The description of `read_notes` says "Returns slug, title, preview, updated_at, and source_count for each note so the cycle can reference prior findings without re-fetching them." Wait, can we retrieve the full note? No, there is no tool to fetch a specific note's full body. But the previews and our memory of them (or semantic search) are usually sufficient, or we can write the updated body directly. Let's check if we can write a note with the same slug. Yes, "slug is the slug parameter you pass. Reusing a slug updates the existing note and snapshots the prior content into note_revisions." Let's verify what the previous notes were about from the previews and write comprehensive, updated versions. Let's look at the previews: 1. `coreweave-hardware-hedging-derivatives-late-cycle-signal`: "As the AI-driven memory and hardware upcycle reaches unprecedented heights, specialized AI cloud giant **CoreWeave** is actively evaluating the use of financial derivatives—specifically futures..." 2. `micron-scas-business-model-transformation-cyclicality`: "Micron Technology has reported a historic performance for its third quarter of fiscal 2026 (ended May 28, 2026), demonstrating an unprecedented structural transformation of its business model. The..." 3. `hbm4-race-nvidia-rubin-spec-cuts-samsung-qualification`: "As the high-bandwidth memory (HBM) upcycle progresses into 2026, the competitive landscape has pivoted to the sixth-generation high-bandwidth memory technology: **HBM4**. Unlike previous generations,..." 4. `dram-contract-pricing-skyrockets-to-record-highs-q1-2026`: "While the global memory market remains exceptionally tight, the explosive quarter-over-quarter run-up in contract prices is showing clear signs of moderation as it enters the third quarter of 2026...." Let's do some more web searches to see if there are other developments. Let's search for "SK Hynix Yongin Phase 1" or "Samsung Taylor Fab 2" or "Nanya Fab 5A" to see if there are any updates on these capacity additions. We saw: `[News] Nanya Tech Unveils Up to US$10.7B Fab 5A Investment Plan Through 2029; Targets 10nm-Class DRAM with EUV` on August 6, 2026. Let's search for "Nanya Tech Fab 5A" or "Nanya Technology investment 2026" to get the details of this massive capex. This is a classic late-cycle tell: "everyone adds capacity at once." Let's search for it.

This is incredibly fascinating. Let's look at the new data: 1. **Nanya Technology's massive capex expansion**: - On August 5, 2026, Nanya's board approved an NT$346.6 billion (approx. **$10.7 billion**) capital expenditure budget for Fab 5A from 2026 to 2029, with the goal of bringing 10nm-class EUV DRAM online. - Nanya also raised its 2026 capital expenditure ceiling by 34%—from NT$52 billion to NT$69.7 billion—to accelerate Fab 5A build-out and make advance payments on equipment. - On August 14, 2026, UDN reported that Nanya is eyeing new fabs in Yunlin and Pingtung for 1d DRAM and custom memory, with investments expected to top NT$300 billion. - This represents a classic late-cycle tell where second-tier DRAM makers (like Nanya) are making massive, multi-billion-dollar bets on advanced EUV lithography and capacity expansions (such as Fab 5A and additional fabs in Yunlin/Pingtung) to capture AI opportunities. - Let's check our notes: `semiconductor-equipment-re-rating-hbm-complexity-gains` (Semiconductor Equipment Leaders Reap Record Revenues from the AI Memory Capex Boom) and `ai-memory-capacity-expansion-race-computex-2026` (The AI Memory Capacity Expansion Race and the Late-Cycle Capex Boom). Reusing `ai-memory-capacity-expansion-race-computex-2026` is perfect because it tracks the global memory capacity expansion race and the late-cycle capex boom! Let's update `ai-memory-capacity-expansion-race-computex-2026` to cover Nanya's $10.7 billion Fab 5A investment and its Yunlin/Pingtung plans. 2. **Winbond and Adata's disclosures**: - Winbond says customers are seeking LTAs through 2030, and Adata expects DRAM supply to be even tighter in 2027. This aligns with Micron's SCA playbook. - SanDisk's gross margin guidance flatlines at 83-85% under its new contract structures. Let's write our notes. We have three main findings to write/update: 1. **DRAM Contract Pricing Moderating and Spot Market Freeze**: Update `dram-contract-pricing-skyrockets-to-record-highs-q1-2026`. We'll detail the sharp deceleration of DRAM contract price increases (+13-18% QoQ in 3Q26 vs +58-63% in 2Q26), the complete stall in the spot market (DDR4 spot prices rising <1% on extremely thin volume), downstream consumer resistance, and the diverging 2027 outlook (DRAM structurally tight, NAND transitioning to surplus). 2. **The HBM4 Race (Samsung's 80% Golden Yield, SK Hynix Labor Dispute, and Samsung's Vietnam Relocation)**: Update `hbm4-race-nvidia-rubin-spec-cuts-samsung-qualification`. We'll detail Samsung's breakthrough 80% yield on HBM4 (ahead of schedule), its plans to triple HBM4 revenue in Q3, SK Hynix's labor dispute over bonuses (opposing management's proposal to pay bonuses in treasury shares), and Samsung's plan to ship legacy backend equipment to Vietnam to free up domestic fab space for HBM advanced packaging. 3. **Micron's KeyBanc Disclosures (SCAs, Data Center Shortages, and USA Premium)**: Update `micron-scas-business-model-transformation-cyclicality`. We'll cover Micron's August 10, 2026 KeyBanc Forum disclosures, including meeting less than half of data center demand, customers scrambling despite "very high" prices, the structure of SCA price caps (tied to CQ2 pricing), and the "Made in USA" front-end premium. 4. **Nanya's Massive $10.7B Fab 5A Expansion and Yunlin/Pingtung Fabs**: Update `ai-memory-capacity-expansion-race-computex-2026`. We'll detail Nanya's massive board-approved NT$346.6B ($10.7B) investment through 2029 for Fab 5A to bring 10nm-class EUV DRAM online, its 34% increase in 2026 capex to NT$69.7B, and its plans for new fabs in Yunlin and Pingtung. This perfectly illustrates the classic late-cycle phenomenon of second-tier players making massive capital investments simultaneously. Let's double check if there are other notes we should write. We have four distinct slugs, and we are updating four existing notes. This is perfect! It keeps our findings as a living, updated set under stable slugs. Let's write the notes now. First, let's update `dram-contract-pricing-skyrockets-to-record-highs-q1-2026`. Wait, let's draft the note body for `dram-contract-pricing-skyrockets-to-record-highs-q1-2026`. We need to make sure we include: - Verbatim quotes from the articles we fetched. - Links to the articles. - Analysis of what it means. Let's see: from `https://xenospectrum.com/en/memory-spot-price-stall-ai-demand-consumer-gap-2026/`: Quote: "According to a TrendForce weekly report published on August 12, the spot average price for DDR4 1Gx8 3200MT/s rose only 0.93%, from $42.11 the previous week to $42.50, with trading volume remaining low. A state in which buyers and sellers cannot reach agreement on price has persisted for more than three weeks since late July." And: "TrendForce forecasts that commodity DRAM contract prices in 3Q26 will rise 13-18% quarter-over-quarter—a sharp deceleration from the 58-63% increase in 2Q26." From `https://www.trendforce.com/presscenter/news/20260730-13158.html`: Quote: "On the DRAM side, continued HBM capacity allocation, robust AI server demand, and growing procurement of CPU memory and HBM are expected to keep supply constrained and prices on an upward trajectory. By contrast, NAND Flash is projected to enter a looser supply environment in 2H27 as new production capacity comes online while consumer electronics demand remains weak, increasing downward pressure on prices." Let's write the note using `write_note`. Let's structure it beautifully. No YAML frontmatter, first heading as title. Let's do it.

Fantastic. Now let's write the second note, updating `hbm4-race-nvidia-rubin-spec-cuts-samsung-qualification`. Wait, let's look at the details we have: - Samsung's HBM4 yield reached 80% (the "golden yield") as of early August 2026, up from below 60% in February. This is months ahead of its year-end target. - Samsung plans to triple HBM4 revenue in Q3 2026, and expects HBM4 to represent over 60% of total HBM revenue in 2H26. It targets a 38% HBM market share by year-end. - SK Hynix's HBM4 yield is also estimated to be in the 80% range, but the company is facing a major labor dispute. On August 9, negotiations failed over a profit-sharing bonus scheme (paying bonuses in treasury shares with a lock-up). A new union has quickly gathered over 3,500 members (10% of the workforce) within days of its August 5 formation. - Samsung is considering shipping legacy memory backend equipment to Vietnam. This frees up domestic cleanroom and packaging space (at fabs like Cheonan and Onyang) to expand its HBM advanced packaging capacity. This highlights that backend packaging throughput, rather than raw wafer starts, is the binding constraint on HBM scale. Let's write this note under the slug `hbm4-race-nvidia-rubin-spec-cuts-samsung-qualification`. Let's draft it.

Outstanding! Now let's update the third note, `micron-scas-business-model-transformation-cyclicality`. Let's see: we have Micron's KeyBanc Technology Leadership Forum disclosures from August 10, 2026. We want to cover: - Customers scrambling for memory despite "very high" prices, with data center supply being the tightest (can fulfill no more than half of demand). - CBO Sumit Sadana expects the supply-demand balance to tighten further in 2027, potentially making it even more constrained than 2026. - DRAM is cited by customers as their number one constraint today (not power, real estate, data center capacity, or logic wafers). - SCAs (Strategic Customer Agreements): 16 SCAs signed, covering ~half of revenue, binding take-or-pay terms through 2030. - SCA pricing structure: Most carry a price band with a ceiling. For the first 16 agreements, the ceiling was set using CQ2 pricing. Future SCAs will have ceilings consistent with market price at the time of signing. This explains why Micron is starting to "moderate its price increases to some extent" (as price ceilings kick in, price gains cool). - "Made in USA" premium: Micron is the only company investing in front-end memory fab manufacturing in the U.S. (Samsung's U.S. investments are logic foundry, SK Hynix is back-end packaging in Indiana). Customers are willing to pay a premium for U.S.-fabricated memory, and this premium is already baked into the SCAs. Let's write this note under the slug `micron-scas-business-model-transformation-cyclicality`. Let's draft it.

Excellent. Now let's write the fourth note, `ai-memory-capacity-expansion-race-computex-2026`. Let's see: we want to cover: - Nanya Technology's massive board-approved NT$346.6 billion (approx. **$10.7 billion**) capital expenditure budget for Fab 5A from 2026 to 2029, with total potential investment reaching up to $16 billion depending on market conditions. - Nanya also raised its 2026 capital expenditure ceiling by 34%—from NT$52 billion to NT$69.7 billion—to accelerate Fab 5A build-out and make advance payments on equipment. - On August 14, 2026, UDN reported that Nanya is eyeing new fabs in Yunlin and Pingtung for 1d DRAM and custom memory, with investments expected to top NT$300 billion. - This is a classic late-cycle tell: "everyone adds capacity at once." The second-tier players (like Nanya) are making massive, multi-billion-dollar bets on advanced EUV lithography and capacity expansions (such as Fab 5A and additional fabs in Yunlin/Pingtung) to capture AI opportunities. - Let's also mention that Winbond has customers seeking LTAs through 2030, and Adata expects DRAM supply to be even tighter in 2027. Let's write this note under the slug `ai-memory-capacity-expansion-race-computex-2026`. Let's draft it.

Excellent. All 4 notes have been successfully updated with high-quality, cited, and verbatim-quoted evidence. Let's review our threads and watches. Is there any other thread or watch we should record? Let's see: we have a watch on "SK Hynix Yongin Phase 1 Cleanroom Opening" in early 2027. And "Samsung Taylor Fab 2 Construction" before end of 2026. And "Samsung Q3 2026 HBM4 Sales Growth". And "NAND Flash Supply-Demand Balance Turns Positive" in 2H 2027. And "Execution of Compute/RAM Price Futures". And "Samsung UFS 5.0". And "Micron Gross Margin Turn Tripwire". And "Micron FQ4 2026 Earnings". And "Samsung 1d DRAM Equipment Delivery". And "2027 HBM4 Contract Pricing". Wait! Let's register a new watch for Nanya's Fab 5A equipment delivery or EUV installation. This is a great forward-looking catalyst. Let's register a watch for Nanya's Fab 5A EUV installation or equipment move-in, which is scheduled to support its 2027-2028 production. Let's call `record_watch`. Let's see: - `watch_type`: `dated_catalyst` or `live_process` or `tripwire` - `cardinality`: `one_shot` - `title`: `Nanya Fab 5A EUV Equipment Delivery and Installation` - `entity`: `Nanya` - `fire_date`: `2027-06-30T00:00:00Z` - `description`: `Monitor Nanya's progress in installing EUV lithography equipment in its new Fab 5A, scheduled for 2027, to prepare for 10nm-class DRAM production.` - `note`: `Fires when Nanya officially confirms the delivery or installation of its first EUV lithography system at Fab 5A.` Let's record this watch.

This is great. Let's also record another watch for SK Hynix's labor negotiations. Since it's an ongoing dispute that could disrupt HBM4 supply, let's create a watch: - `watch_type`: `live_process` - `cardinality`: `one_shot` - `title`: `SK Hynix and Union Wage/Bonus Agreement` - `entity`: `SK Hynix` - `description`: `Monitor SK Hynix's labor negotiations to see if they reach a final wage and bonus agreement, resolving the dispute over paying bonuses in treasury shares.` - `note`: `Fires when SK Hynix and its union officially sign a final collective bargaining agreement or if a strike is formally called.` Let's record this watch.

Excellent. Let's see if there are any other open threads or actions. No, we listed the threads earlier and none of them have been fully resolved this cycle (since they are tied to future earnings releases, cleanroom openings, or 2027 contract outcomes). However, we've provided massive updates that advance the story significantly. Let's call `finish_cycle` now. I will not pass `no_findings_reason` because I wrote/updated four findings this cycle.

I have completed the research cycle for topic `019e8ec8-ffb3-71d7-981c-e48354ab25e7` on the AI-driven memory/storage cycle. Here is a summary of the findings and updates made during this cycle: ### 1. DRAM Contract Pricing Deceleration & Spot Market Freeze * **Slug:** `dram-contract-pricing-skyrockets-to-record-highs-q1-2026` (Updated) * **Summary:** Commodity DRAM contract price increases are decelerating sharply, projected to rise by **13% to 18% QoQ in Q3 2026**, down from +58% to 63% in Q2 and +93% to 98% in Q1. Meanwhile, the spot market has entered a complete "gridlock phase" with extremely thin trading volumes and spot prices moving by less than 1% week-over-week. This divergence reveals a clear fault line between robust AI server demand and the affordability limits of downstream consumer markets (PCs, smartphones). * **2027 Outlook:** TrendForce outlines a stark divergence for 2027. DRAM will remain in a structural deficit (-1% to -2% sufficiency ratio) due to HBM wafer cannibalization (HBM requires 3x the wafers of conventional DRAM). Conversely, NAND Flash is projected to transition to a supply surplus in 2H27 as process migrations and new fab ramps boost bit supply while consumer demand remains weak. ### 2. The HBM4 Race: Samsung’s Yield Breakthrough, SK Hynix Labor Crisis, & Vietnam Relocation * **Slug:** `hbm4-race-nvidia-rubin-spec-cuts-samsung-qualification` (Updated) * **Summary:** * **Samsung Yields:** Samsung has achieved a major manufacturing milestone, with its HBM4 production yield reaching the **80% "golden yield"** mark as of early August 2026, up from below 60% in February. This is months ahead of its original year-end target. Samsung plans to more than triple HBM4 revenue QoQ in Q3 2026, targeting a 38% HBM market share by year-end. * **SK Hynix Labor Dispute:** SK Hynix's HBM leadership is facing operational headwinds as labor negotiations deadlock. The union has rejected management's proposal to pay more than half of bonuses in treasury shares subject to a lock-up. A newly formed union has quickly mobilized over 3,500 participants (10% of the workforce) within days of its August 5 formation. * **Vietnam Relocation:** Samsung is considering shipping its legacy memory backend packaging equipment to Vietnam. This strategy frees up valuable cleanroom space at its domestic fabs (Cheonan and Onyang) for advanced packaging, which is currently the binding constraint on its HBM scale. ### 3. Micron’s KeyBanc Forum Disclosures: SCA Price Caps & Domestic Premium * **Slug:** `micron-scas-business-model-transformation-cyclicality` (Updated) * **Summary:** Micron executives at the KeyBanc Technology Leadership Forum on August 10, 2026, disclosed that they can fulfill **less than half of data center demand**, with customers pointing to DRAM as their number one constraint. Crucially, they revealed that most of their 16 Strategic Customer Agreements (SCAs) carry a price band with a ceiling set using **CQ2 2026 pricing**. This explains why Micron's price increases are starting to moderate even as the company guides 2027 to be tighter than 2026. Additionally, Micron is commanding a pricing premium for its "Made in USA" front-end memory fabrication. ### 4. Nanya’s $10.7 Billion Fab 5A Investment & Southern Taiwan Expansion * **Slug:** `ai-memory-capacity-expansion-race-computex-2026` (Updated) * **Summary:** Second-tier players are aggressively entering the capacity race, illustrating a classic late-cycle tell where everyone adds capacity simultaneously. Nanya Technology's board approved a massive **NT$346.6 billion (approx. US$10.7 billion) capital expenditure budget for Fab 5A** through 2029 to bring 10nm-class EUV DRAM online. Nanya also raised its 2026 capex by 34% to NT$69.7 billion to accelerate the build-out, and is reportedly planning new fabs in Yunlin and Pingtung representing over NT$300 billion in additional investment. ### 5. New Watches Registered * **Nanya Fab 5A EUV Equipment Delivery and Installation:** Tracking Nanya's progress in installing EUV systems in Fab 5A by mid-2027. * **SK Hynix and Union Wage/Bonus Agreement Resolution:** Monitoring the union dispute to see if it reaches a resolution or escalates into a strike.