Cycle on Jun 3, 2026

I will begin by listing the open threads for this topic to understand any existing research questions or priorities established in previous cycles.

I will now read the existing notes for this topic to understand what has already been researched and captured, so I can build upon prior findings or identify key gaps.

There are no existing notes or open threads for this topic. This is the first cycle or a clean slate. The topic brief is: "Track the AI-driven memory/storage cycle — HBM, DRAM, NAND — and whether this upcycle is structurally different or the usual boom that busts... Core entities: Micron/MU, SK Hynix, Samsung and the HBM/DRAM/NAND mix; the demand pull (Nvidia/AMD attach rates, hyperscaler orders); supply signals (capex, wafer starts, the Samsung labor situation, yields); the equipment layer (Lam, Applied, ASML, Entegris). Track contract pricing (TrendForce/DRAMeXchange commentary), HBM allocation and "sold-out" claims, bit-supply guidance, inventory, and earnings commentary on pricing power. Flag where pricing/allocation diverges from the "permanently sold out" story, and the classic late-cycle tell (everyone adds capacity at once)." Let's begin by pulling some public market data for key tickers: - Micron (MU) - SK Hynix (000660.KS - though let's search if ticker kind supports it, or if we look at MU and others first) - Samsung Electronics (005930.KS) - Equipment providers: ASML, Applied Materials (AMAT), Lam Research (LRCX) Let's first use `market_view` for MU, and then search the web for recent news, earnings, TrendForce pricing updates, and HBM capacity expansions. Let's check `market_view` for MU.

The market view for MU is incredibly interesting: - MU is trading at $1064.10, which is its 52-week high (100% position). - Look at the revenue trajectory: - Q Ending 2026-02-28: $23.86B revenue, $16.14B operating income, $13.79B net income, $18.52B EBITDA. - Q Ending 2025-11-30: $13.64B revenue, $6.14B operating income. - Q Ending 2025-08-31: $11.31B revenue, $3.75B operating income. - Q Ending 2025-05-31: $9.30B revenue, $2.17B operating income. - Revenue has exploded from $9.30B in May 2025 to $23.86B in Feb 2026. Yes, that is a massive upcycle! Operating margin is 67.6%, and profit margin is 41.5%. - Morgan Stanley recently (June 3, 2026) raised targets for MU and SNDK, citing 2-3 years of tight memory supply. - There's a neutral article from Morgan Stanley warning that AI 'chipflation' is spreading from data centers to the wider economy, causing margins for other hardware to compress. - MU's market cap has reached $1.20T. Let's do some search on recent reports, specifically: - TrendForce memory contract pricing updates for Q2/Q3 2026. - HBM supply, "sold-out" announcements by SK Hynix, Micron, and Samsung. - Samsung labor strikes or yield issues. - ASML, Applied Materials, Lam Research views. Let's search for TrendForce DRAM NAND contract pricing updates in 2026.

This is exceptionally fresh data (June 1-3, 2026). Let's fetch the actual press releases or articles to get verbatim quotes: 1. `https://www.trendforce.com/presscenter/news/20260601-13070.html` (Rapid Contract Price Surge Drives 1Q26 DRAM Industry Up 81% QoQ, Says TrendForce) 2. `https://www.trendforce.com/presscenter/news/20260602-13074.html` (Tight DRAM Supply Gives Suppliers Greater Pricing Power in HBM, with HBM Contract Prices Expected to Surge Multiples Higher in 2027, Says TrendForce) 3. `https://www.theregister.com/storage/2026/06/02/expect-more-of-those-dram-price-hikes-as-memory-shortage-continues-to-bite/5250049` (Expect more of those DRAM price hikes as memory shortage continues to bite) 4. `https://wccftech.com/memory-giants-gain-pricing-power-over-ai-firms-as-ddr5-rdimms-beat-hbm-profitability-setting-stage-for-2027-hikes/` (Memory Giants Gain Pricing Power Over AI Firms as DDR5 RDIMMs Beat HBM Profitability, Setting Stage for 2027 Hikes) Let's fetch these four URLs to extract exact details. I'll start with the TrendForce press releases.

This is an incredibly rich update. Let's look at the key details: - **1Q26 DRAM Industry Revenue** surged **81% QoQ** to **$97 billion**, driven by conventional DRAM contract prices increasing by **93% to 98% QoQ**! - **2Q26 Outlook**: DRAM suppliers' inventory levels are "extremely low," and conventional DRAM contract prices are expected to rise another **58-63% QoQ** in 2Q26. - **Market Share**: Samsung is #1 with $37.32B revenue (+93.4% QoQ, 38.5% share); SK Hynix is #2 with $27.98B revenue (+62.5% QoQ, 28.8% share); Micron is #3 with $21.75B revenue (+81.6% QoQ, 22.4% share). - **HBM vs RDIMM shift**: The Shift to General-Purpose Servers is expanding memory procurement needs beyond just HBM3e, LPDDR5X, and high-capacity RDIMMs to include a broader range of RDIMM capacities. - **Wafer output**: Suppliers are relying primarily on process migrations to expand bit output in 2026, so wafer input growth remains limited. Let's fetch the second TrendForce report: `https://www.trendforce.com/presscenter/news/20260602-13074.html` (Tight DRAM Supply Gives Suppliers Greater Pricing Power in HBM, with HBM Contract Prices Expected to Surge Multiples Higher in 2027, Says TrendForce). This is critical because it talks about HBM pricing power, HBM4 negotiations, and the 2027 outlook! This will directly address the "permanently sold out" story and pricing power.

This is a fascinating dynamic: - **HBM vs DDR5 RDIMM Profitability**: HBM wafer revenue was actually **overtaken** by DDR5 64GB RDIMM in 1Q26, and HBM profitability fell below it. Why? Because HBM contract prices are set annually, so they haven't caught up with the massive quarterly spikes in conventional DRAM (which rose ~95% in 1Q26 and is expected to rise 58-63% in 2Q26). - **Supplier leverage in 2027**: Because of this, suppliers are pointing to DDR5 RDIMM profitability to demand massive price increases for HBM4 in 2027 negotiations, which are happening *now* (in Q2 2026). - **HBM capacity allocation**: HBM wafer input as a % of total DRAM wafer input is expected to reach 18% in 2025, 22% in 2026, and 30% in 2027. This represents a massive crowding-out of conventional DRAM capacity (since HBM takes 3x more wafer capacity for the same bits due to lower yields and larger die sizes). - **Demand drivers**: - 2026: AI ASICs capacity upgrades (96GB/192GB to 216GB/288GB) and higher NVIDIA Rubin platform shipment volumes. - 2027: NVIDIA Rubin Ultra (384GB per GPU) and AI ASICs (Google TPU, etc.) deployment. Let's fetch the other two articles to see more details about this dynamic. `https://wccftech.com/memory-giants-gain-pricing-power-over-ai-firms-as-ddr5-rdimms-beat-hbm-profitability-setting-stage-for-2027-hikes/` And `https://www.theregister.com/storage/2026/06/02/expect-more-of-those-dram-price-hikes-as-memory-shortage-continues-to-bite/5250049` Let's fetch both.

Let's also fetch `https://www.theregister.com/storage/2026/06/02/expect-more-of-those-dram-price-hikes-as-memory-shortage-continues-to-bite/5250049`.

This provides incredible, specific details: - **SK Hynix wafer capacity expansion**: Chairman Chey Tae-won said in Taipei (June 2, 2026) that it plans to double its silicon wafer output capacity *gradually over the next five years*. He warned the shortage could persist until 2030, though other analysts expect it to last until the end of 2027. - **Micron capacity expansion**: Micron started DRAM manufacturing at its Manassas, Virginia, plant and expects initial wafer output at its Boise, Idaho, fab in "mid-calendar year 2027." Meaningful new capacity is projected to come fully online in 2027 and 2028. - **Samsung labor strike update**: A threatened strike by workers at Samsung Electronics was *called off at the end of May 2026* after the company agreed to create a fund that will share profits with workers. This averted a major disruption to memory production. - **"Chipflation" and PC prices**: PC prices are up by double-digit percentages in Europe due to the memory crunch. Let's search for the SK Hynix Reuters article or similar articles to confirm their capacity plans: "SK Hynix plans double wafer capacity next five years group chairman says 2026-06-02". Let's also do a search on Samsung's recent HBM status (specifically HBM3e qualifications with Nvidia, which has been a major point of discussion in the memory space). Let's search for "Samsung Nvidia HBM3e qualification 2026" or similar.

This is absolutely incredible. Look at the date: June 1-3, 2026. Computex 2026 is happening *right now*. Let's see what is happening: - **Roundhill Memory ETF (DRAM)**: Launched on April 2, 2026, and has run 128% in 40 trading days! - **SK Hynix**: Chairman Chey Tae-won confirmed at Computex 2026 that SK Hynix will double its wafer capacity within five years, citing a memory shortage that could stay locked through 2030. - **Samsung**: On Friday, May 29, 2026, Samsung Electronics started shipping samples of its latest HBM4E chip (high-bandwidth memory) to customers, pulling ahead of rivals in the HBM4 race. Also, Samsung unveiled an HBM5 prototype with a 2nm base die at Computex 2026. And Samsung's fifth-generation 12-layer HBM3E product has passed Nvidia's qualification. - **Micron**: Micron's valuation has crossed $1 trillion (as seen in the MU market view, it's at $1.20T). Its entire 2026 HBM capacity is sold out. Let's fetch the following URLs to get full details and exact quotes: 1. `https://startupfortune.com/sk-hynix-pledges-to-double-wafer-capacity-within-five-years-as-ai-memory-shortage-deepens-toward-2030/` (SK Hynix pledges to double wafer capacity within five years as AI memory shortage deepens toward 2030) 2. `https://www.reuters.com/world/asia-pacific/samsung-electronics-ships-hbm4e-chip-samples-global-customers-2026-05-28/` (Samsung Electronics ships faster HBM4E chip samples to customers; shares jump) 3. `https://en.sedaily.com/finance/2026/06/02/samsung-applies-2nm-to-hbm5-first-sk-to-accelerate-output` (Samsung Applies 2nm to HBM5 First; SK to Accelerate Output via AI Factory) 4. `https://www.techtimes.com/articles/317695/20260603/ai-memory-shortage-locked-through-2030-computex-2026-brings-agent-economy-hbm-crisis.htm` (AI Memory Shortage Locked Through 2030: Computex 2026 Brings Agent Economy, HBM Crisis) Let's fetch these. I'll start with the Reuters article on Samsung shipping HBM4E samples and the Startup Fortune article on SK Hynix doubling capacity.

This is incredibly detailed! Let's pull some core figures: - **SK Hynix Capex & Capacity**: - Capex in 2026 is projected to surpass **KRW 30 trillion (~$21 billion)**, up from the mid-KRW 20 trillion range in 2025. - New Cheongju fab, **M15X**, coming online in H1 2026, starting at 10k wafers/month and ramping to 50k wafers/month by Q4 2026. - Total DRAM capacity target is **~620,000 wafers per month by H2 2026**, nearly double the mid-300k level in 2023. - SK Hynix currently holds **58% of the global HBM market**. - **Samsung Capex & Capacity**: - 2026 capex expected to reach **~$20 billion** (up 11% YoY), focused on advanced process nodes and its **P4L facility**. - **Micron Capex & Capacity**: - 2026 capex projected at **$13.5 billion** (up 23% YoY). - Micron expects initial wafer output at its Boise, Idaho, fab in "mid-calendar year 2027." - **The HBM4/HBM4E Pivot**: - SK Hynix completed its HBM4 mass-production setup in Sept 2025 and has been supplying Nvidia with paid samples. Expected to supply ~2/3 of Nvidia's HBM4 requirements. - Samsung is pushing hard into HBM4/HBM4E. Let's fetch the Reuters article on Samsung shipping HBM4E samples: `https://www.reuters.com/world/asia-pacific/samsung-electronics-ships-hbm4e-chip-samples-global-customers-2026-05-28/` And then `https://en.sedaily.com/finance/2026/06/02/samsung-applies-2nm-to-hbm5-first-sk-to-accelerate-output` And `https://www.techtimes.com/articles/317695/20260603/ai-memory-shortage-locked-through-2030-computex-2026-brings-agent-economy-hbm-crisis.htm` Let's fetch these three.

This is an incredibly important piece of news: - **Samsung ships 12-layer HBM4E samples** (on May 28/29, 2026). This is sixth-generation, 10-nanometer-class (1c DRAM) process technology, with a 4-nanometer foundry logic base die. - **Timing**: It comes just 3 months after Samsung began shipping HBM4 samples in February 2026. - **HBM Market Share (Q4 2025 - Counterpoint Research)**: - SK Hynix: 57% - Samsung: 22% - Micron: 21% - **Anthropic Funding & Strategic Alliance**: Anthropic raised funding at a **$965 billion valuation** (!) and named Samsung, Micron, and SK Hynix as strategic partners. Samsung is specifically referenced for logic chip capabilities. Let's fetch the other two articles: - `https://en.sedaily.com/finance/2026/06/02/samsung-applies-2nm-to-hbm5-first-sk-to-accelerate-output` - `https://www.techtimes.com/articles/317695/20260603/ai-memory-shortage-locked-through-2030-computex-2026-brings-agent-economy-hbm-crisis.htm` Let's fetch them now.

This is incredibly high-quality primary source material. Let's look at the details: - **SK Hynix Yongin Cluster Fab (Y1)**: - Investing **31 trillion won** (~$22.5 billion) in its first fab (Y1) at the Yongin semiconductor cluster. - Expected to begin full operation after completion in **Q1 2027** (or "next year", which is 2027 since we are in June 2026). - Target DRAM production capacity of **300,000 wafers per month** by the end of 2027. - **Samsung's HBM5 Prototype**: - Unveiled at Computex 2026. - Applies a base/logic die built on its own **2nm foundry process**. - Considering structures of **up to 20 layers**. - **Supply-Demand Mismatch**: - Industry sources say that the volume the three major memory makers can supply is **only around 50%** of the medium- to long-term demand required by global Big Tech firms. - **Strategy Swap**: - A year ago, SK Hynix focused on technology innovation while Samsung focused on capacity expansion. - Now, they have swapped strategies: SK Hynix is aggressively focusing on capacity expansion (doubling capacity in five years, building Yongin Y1, etc.) while Samsung is focusing on technology leadership (HBM4E, 2nm logic die HBM5, vertical integration). Let's fetch the last URL: `https://www.techtimes.com/articles/317695/20260603/ai-memory-shortage-locked-through-2030-computex-2026-brings-agent-economy-hbm-crisis.htm` to see if there are any other specific angles (e.g. equipment suppliers, Nvidia/AMD attach rates, contract pricing divergences, etc.).

This is incredibly rich and comprehensive. Let's analyze the findings we have gathered: 1. **DRAM Memory Pricing Explosion and Record Revenue (1Q26 and 2Q26)**: - 1Q26 DRAM industry revenue surged **81% QoQ** to **$97 billion**. - Driven by a massive **93% to 98% QoQ** rise in conventional DRAM contract prices. - 2Q26 is expected to see another **58% to 63% QoQ** increase in conventional DRAM contract prices. - This price spike is driven by extremely low supplier inventories and the prioritization of wafer capacity for high-capacity RDIMMs and HBM for AI servers, crowding out PC and smartphone clients. - PC prices in Europe are climbing by double-digit percentages as a result. 2. **The "HBM vs. DDR5 RDIMM" Profitability Arbitrage & 2027 Price Negotiations**: - Because HBM contract prices are negotiated annually, they have lagged behind the explosive, near-doubling quarterly price spikes of conventional DRAM. - In 1Q26, HBM per-wafer revenue and profitability was actually **overtaken** by DDR5 64GB RDIMMs! - This has given the three major memory makers (SK Hynix, Samsung, Micron) massive leverage in current Q2 2026 negotiations for 2027 HBM4 contracts. - If buyers do not accept massive price increases for 2027 HBM4, suppliers can simply threaten to allocate tight wafer capacity to lucrative DDR5 RDIMMs instead. 3. **Massive Capacity Expansion Plans (The Late-Cycle Tell? Or Structural Shift?)**: - SK Hynix Chairman Chey Tae-won announced at Computex 2026 (June 2026) that SK Hynix will **double its wafer capacity over the next five years** to address an AI memory shortage expected to persist through 2030. - SK Hynix's capex is projected to surpass **KRW 30 trillion (~$21 billion)** in 2026 (up from mid-20s in 2025). - SK Hynix is investing **31 trillion won** in its first Yongin cluster fab (Y1), expected to operate in Q1 2027 and target **300,000 wafers/month** by the end of 2027. - Its Cheongju fab, **M15X**, is ramping from 10k/month in H1 2026 to 50k/month by Q4 2026, aiming for total DRAM capacity of **620,000 wafers per month by H2 2026** (nearly double 2023 levels). - Samsung's 2026 capex is expected to reach **~$20 billion** (up 11% YoY), focusing on P4L and advanced process nodes. - Micron's 2026 capex is projected at **$13.5 billion** (up 23% YoY), with its Boise fab starting initial wafer output in mid-2027. - The thesis states: "everyone adds capacity at once." This is the classic late-cycle tell, even if the executives are framing it as a structural shortage until 2030. We are seeing tens of billions of dollars in capex and massive wafer-capacity targets (like SK Hynix aiming to double wafer capacity and hitting 620k wafers/month by H2 2026). 4. **Samsung's HBM Technology Leap (HBM4E and HBM5)**: - Samsung is fighting back aggressively after falling behind in HBM3E. - On May 28/29, 2026, Samsung announced shipping **12-layer HBM4E samples** using its sixth-generation 10nm-class (1c DRAM) process and its own 4nm foundry logic base die. - At Computex 2026, Samsung unveiled a mockup of **HBM5** featuring a base/logic die built on its own **2nm foundry process** and up to 20 stacked layers. - This vertical integration (combining memory and foundry logic in-house) is Samsung's key differentiator against SK Hynix and Micron, which rely on TSMC for the logic base die in HBM4/HBM5. Let's check if there are any other aspects. Is there any equipment layer news? Let's search for "Lam Research" "Applied Materials" "ASML" recent commentary or news regarding DRAM / HBM capex in 2026. Let's do a search.

Let's look up the Applied Materials presentation from the Bank of America 2026 Global Technology Conference (which happened recently, around late May / early June 2026): `https://seekingalpha.com/article/4911070-applied-materials-inc-amat-presents-at-bank-of-america-2026-global-technology-conference` Let's see if we can fetch this or search for it to get some details. Let's search for "Applied Materials" "BofA" 2026 "HBM" "trade ratio" or similar to see what they said about HBM wafer trade ratios. This is highly relevant to the "how much wafer capacity is HBM consuming" question. Let's do a search.

Let's fetch `https://www.forbes.com/sites/greatspeculations/2026/06/03/does-applied-materials-stock-deserve-its-epic-rally/` to read the full context on Applied Materials (AMAT) and its HBM/DRAM wafer-fab-equipment (WFE) drivers. This is perfect for the equipment layer of the supply cycle.

This is a phenomenal piece of analysis. Let's summarize the key points for Applied Materials (AMAT) and the equipment layer: - **Epic Rally**: AMAT's stock has nearly tripled over the last year, trading at ~37x forward earnings and 12x trailing sales (up from less than 4x sales just 12 months ago). - **Consensus Growth**: Revenue is expected to grow 17% this year (2026) and 25% next year (2027). - **HBM Impact on Equipment**: - HBM units occupy roughly **three times the wafer area** of standard DRAM. - HBM involves **19 manufacturing steps, 15 of which pertain to semiconductor equipment**. - Applied Materials captures **over half of the equipment value** linked to those 15 steps. - This means as HBM expands, equipment expenditures are outpacing memory volumes. - **Advanced Packaging Revenue**: AMAT expects advanced packaging revenue to surge by **over 50% in calendar year 2026**. - **Visibility Shift**: AMAT now has up to **eight quarters (2 years) of forward demand visibility**, compared to 3-6 months historically. This shifts the perception of AMAT from a highly cyclical toolmaker to a stable infrastructure provider. Let's double check if we need to pull the `market_view` for AMAT to get its actual numbers /markets URL. Yes, let's pull `market_view(kind="ticker", id="AMAT")`. This is a critical step for public-market analysis. Let's do that right now.

This is exceptional! AMAT is also at its 52-week high, with a market cap of $389.08B. Let's look at the quarterly trajectory: - Q Ending 2026-04-30: $7.91B revenue, $2.52B operating, $2.81B net. - Q Ending 2026-01-31: $7.01B revenue. - Q Ending 2025-10-31: $6.80B revenue. - Q Ending 2025-07-31: $7.30B revenue. Revenue is growing at a slower pace compared to MU (~11.4% YoY), but the *forward valuation re-rating* is what is driving the stock (nearly tripling over the last year, now at 46.1 P/E trailing, or ~37x forward). Let's also look at Lam Research (LRCX) and ASML just to be thorough and see if we need their views. Let's pull `market_view` for LRCX.

This is excellent. We now have a very clear picture of what's happening. Let's organize our findings. We'll write four distinct notes, each focusing on a single theme or story with concrete evidence, quotes, and links: 1. **DRAM Memory Pricing Explosion**: The 93-98% QoQ contract price explosion in Q1 2026, expected 58-63% rise in Q2 2026, and record-breaking DRAM industry revenues reaching $97 billion. 2. **The HBM vs. DDR5 RDIMM Profitability Arbitrage**: How high-capacity DDR5 RDIMMs became more profitable than HBM in 1Q26 due to lagged annual HBM contracts, giving memory giants massive pricing power for 2027 HBM4 negotiations. 3. **The AI Capacity Expansion Race (The Capex Boom)**: High-stakes capacity expansion announcements at Computex 2026, with SK Hynix pledging to double wafer capacity within 5 years and investing KRW 31T in Yongin Fab Y1, Samsung's $20B capex, and Micron's $13.5B capex. 4. **Samsung's Vertical Integration and HBM Technology Leap**: Samsung's shipping of 12-layer HBM4E samples and unveiling of HBM5 featuring a 2nm base die, leveraging its in-house foundry logic to challenge SK Hynix's dominance. Wait, let's write these findings clearly. Each finding must carry 1-2 representative verbatim quotes, copied exactly, with links (e.g. TrendForce press releases, Reuters, or Forbes articles). Let's carefully format each note. ### Note 1: DRAM Contract Pricing Skyrockets to Record Highs - Slug: `dram-contract-pricing-skyrockets-to-record-highs-q1-2026` - Source: - TrendForce press release: `https://www.trendforce.com/presscenter/news/20260601-13070.html` - The Register: `https://www.theregister.com/storage/2026/06/02/expect-more-of-those-dram-price-hikes-as-memory-shortage-continues-to-bite/5250049` - Verbatim quotes: - "TrendForce’s latest survey reveals that the memory industry experienced a significant boost in 1Q26 due to rapidly rising contract prices for conventional DRAM, which increased by approximately 93% to 98% QoQ. This surge contributed to an overall industry revenue increase of 81% QoQ, reaching $97 billion." - "The continuing AI memory crunch saw DRAM prices effectively double in calendar Q1, and the bad news is they are likely to rise again by more than 50 percent in the current quarter, if TrendForce forecasters are on the money." ### Note 2: The HBM vs DDR5 Profitability Arbitrage - Slug: `hbm-vs-ddr5-profitability-arbitrage-leverage-for-2027` - Source: - TrendForce press release: `https://www.trendforce.com/presscenter/news/20260602-13074.html` - Wccftech: `https://wccftech.com/memory-giants-gain-pricing-power-over-ai-firms-as-ddr5-rdimms-beat-hbm-profitability-setting-stage-for-2027-hikes/` - Verbatim quotes: - "Based on TrendForce’s analysis of per-wafer revenue for HBM and conventional DRAM—estimated using die size, yield rates, and per-Gb pricing—HBM wafer revenue was overtaken by DDR5 64GB RDIMM in 1Q26. This has led HBM profitability to also fall below that of DDR5 64GB RDIMM since 1Q26." - "The higher DRAM prices also provide the memory manufacturers with greater incentive to dictate HBM pricing terms in their negotiations, says TrendForce. In case their HBM prices are not accepted by the buyers, the memory manufacturers can simply allocate the tight production capacity to DRAM instead" ### Note 3: The AI Memory Capacity Expansion Race - Slug: `ai-memory-capacity-expansion-race-computex-2026` - Source: - Startup Fortune: `https://startupfortune.com/sk-hynix-pledges-to-double-wafer-capacity-within-five-years-as-ai-memory-shortage-deepens-toward-2030/` - Seoul Economic Daily: `https://en.sedaily.com/finance/2026/06/02/samsung-applies-2nm-to-hbm5-first-sk-to-accelerate-output` - Tech Times: `https://www.techtimes.com/articles/317695/20260603/ai-memory-shortage-locked-through-2030-computex-2026-brings-agent-economy-hbm-crisis.htm` - Verbatim quotes: - "SK Group Chairman Chey Tae-won announced at Computex 2026 that SK Hynix will double its wafer capacity within five years, citing a memory shortage he expects to persist through 2030. The company holds 58 percent of the global HBM market and is projected to spend over KRW 30 trillion in capex this year alone." - "Behind SK hynix's emphasis on capacity lies a structural memory supply shortage driven by the AI boom. According to industry sources, the volume that the three memory makers, including SK hynix, can currently supply stands at only around 50% of the medium- to long-term demand required by global Big Tech firms. SK hynix is accordingly investing 31 trillion won in its first fab (Y1) at the Yongin semiconductor cluster" ### Note 4: Samsung's Vertical Integration and HBM Technology Leap - Slug: `samsung-vertical-integration-hbm4e-hbm5-leap` - Source: - Reuters: `https://www.reuters.com/world/asia-pacific/samsung-electronics-ships-hbm4e-chip-samples-global-customers-2026-05-28/` - Seoul Economic Daily: `https://en.sedaily.com/finance/2026/06/02/samsung-applies-2nm-to-hbm5-first-sk-to-accelerate-output` - Verbatim quotes: - "The South Korean tech company said the new chip, the 12-layer HBM4E, is more than 20% faster than its previous-generation HBM4 products. Samsung said the chip uses its latest 1c DRAM process technology - sixth-generation, 10-nanometer-class DRAM - alongside Samsung's 4-nanometer foundry logic base die." - "Samsung's HBM5 applies a base (logic) die built on its own 2-nanometer (one-billionth of a meter) foundry process, demonstrating a technology gap that vaults well beyond the current 4nm." ### Note 5: Semiconductor Equipment Layer Re-rating (AMAT and LRCX) Wait, let's also write a note on the equipment layer itself, since the brief explicitly mentions "the equipment layer (Lam, Applied, ASML, Entegris)". - Slug: `semiconductor-equipment-re-rating-hbm-complexity-gains` - Source: - Forbes: `https://www.forbes.com/sites/greatspeculations/2026/06/03/does-applied-materials-stock-deserve-its-epic-rally/` - Verbatim quotes: - "Management indicated that each HBM unit occupies roughly three times the wafer area of standard DRAM and involves 19 manufacturing steps, 15 of which pertain to semiconductor equipment. Applied captures over half of the equipment value linked to those additional procedures." - "Applied is now capable of obtaining up to eight quarters of forward demand visibility, or around two years, contrasted with the three to six months that historically preceded equipment orders. This transformation is significant as it alters how investors assess the business’s value." Let's draft and write these five notes. Remember: - No YAML frontmatter. The slug is passed as the `slug` parameter in `write_note`. - Carry verbatim quotes with their links. - Use `[[slug]]` to wikilink other notes. - Ground everything in the market views and actual fetched pages. Let's write Note 1.

Now let's write Note 2 on the HBM vs. DDR5 RDIMM Profitability Arbitrage. Let's make sure we refer to Note 1 via its wikilink `[[dram-contract-pricing-skyrockets-to-record-highs-q1-2026]]`.

Now let's write Note 3 on the AI Memory Capacity Expansion Race (the capex boom and the classic late-cycle tell). We will link the previous notes. Let's write it.

Now let's write Note 4 on Samsung's Vertical Integration and HBM Technology Leap. We will link the previous notes. Let's write it.

Now let's write Note 5 on the Semiconductor Equipment Layer Re-rating (AMAT, LRCX). This is highly relevant to the "equipment layer" mentioned in the brief. We will link the previous notes. Let's write it.

We have written five fantastic, highly detailed, and thoroughly cited notes covering all aspects of the memory/storage supply-cycle mechanics. Let's review if there are any open questions that we should record as a thread for future cycles. One interesting open question is whether Samsung will successfully complete its HBM3E and HBM4E qualifications with Nvidia, as this will determine whether the HBM market remains an SK Hynix/Micron duopoly or shifts to a highly competitive landscape. Another open question is tracking whether the massive capex additions (e.g. SK Hynix Yongin Y1 starting in Q1 2027, Micron Boise starting in mid-2027) will result in a supply glut in late 2027 or 2028, or if the HBM wafer penalty and agentic token demand indeed absorb all capacity through 2030. Let's record a thread for these. I'll open a thread using `record_thread`. Let's check the parameters. Parameters: - `topic_id`: 019e8ec8-ffb3-71d7-981c-e48354ab25e7 - `cycle_id`: empty or a value? Let's check the kickoff message or metadata. The Chat ID is `topic-019e8ec8-ffb3-71d7-981c-e48354ab25e7-cycle-098efba7-aa58-46a8-8d8b-e40f1c195840`. The cycle ID is `098efba7-aa58-46a8-8d8b-e40f1c195840`. Let's use this cycle ID. - `kind`: question | investigation | contradiction - `title`: Samsung HBM Nvidia Qualification and 2027-2028 Supply Glut Risk - `description`: Track Samsung's ongoing qualification progress with Nvidia for HBM3E/HBM4E, and monitor whether the massive multi-billion dollar capex plans (Yongin Y1, Cheongju M15X, Samsung P4L, Micron Boise) risk triggering a severe supply glut in late 2027 or 2028. Let's call `record_thread`.

Ah, the error says: `invalid thread kind "investigation"`. Wait, let me check the description: `kind is one of 'question', 'investigation', 'contradiction'`. Let me look at the error: `invalid thread kind "investigation` (missing closing quote or maybe the enum is strict). Let me try `question` instead of `investigation` to be safe. Let's call `record_thread` with `kind="question"`.

Excellent, the thread has been created successfully. Now I will call `finish_cycle` to complete this research cycle. I have written five highly detailed notes and recorded an open thread for future cycles. I do not have any open threads to resolve since this was the first cycle and `list_threads` returned an empty list. So I will go straight to `finish_cycle`.

I have completed the research cycle on the AI-driven memory/storage cycle. Here is a summary of the findings that have been successfully registered: 1. **DRAM Contract Pricing Skyrockets to Record Highs in Q1 2026** (`dram-contract-pricing-skyrockets-to-record-highs-q1-2026`): - Conventional DRAM contract prices surged by **93% to 98% QoQ** in Q1 2026, driving global DRAM industry revenue up **81% QoQ to $97 billion**. - TrendForce expects another **58% to 63% QoQ** price increase in Q2 2026. This acute shortage is driven by the structural diversion of wafer capacity to HBM and high-capacity server RDIMMs, which has severely squeezed consumer PC and smartphone OEMs. - Micron's (MU) financials reflect this: revenue grew **196.3% YoY** to $23.86 billion last quarter, with net income surging **756% YoY** to $13.79 billion. 2. **HBM vs DDR5 Profitability Arbitrage Grants Suppliers Leverage for 2027 HBM4 Negotiations** (`hbm-vs-ddr5-profitability-arbitrage-leverage-for-2027`): - Because HBM contract prices are set annually, they lagged behind the near-doubling quarterly spikes of conventional DRAM. - Consequently, in Q1 2026, HBM per-wafer profitability actually fell **below** that of high-capacity DDR5 64GB RDIMMs. - Memory makers (Samsung, SK Hynix, Micron) are now using this profitability inversion as massive leverage in negotiations for 2027 HBM4 contracts, threatening to allocate wafer capacity back to DDR5 RDIMMs if AI customers do not accept substantial HBM4 price increases. 3. **The AI Memory Capacity Expansion Race and the Late-Cycle Capex Boom** (`ai-memory-capacity-expansion-race-computex-2026`): - Memory giants are embarking on a massive capex race. SK Hynix pledged at Computex 2026 to **double total wafer capacity over the next five years**, projecting the shortage to last through 2030. - SK Hynix's 2026 capex will surpass **KRW 30 trillion (~$21 billion)**, with KRW 31 trillion targeted for its Yongin cluster fab (Y1) to hit **300,000 wafers/month** by late 2027. - Samsung is targeting **~$20 billion** in 2026 capex (focusing on P4L), and Micron is projecting **$13.5 billion** (ramping Boise for mid-2027 output). - This simultaneous multi-billion dollar buildout represents the classic late-cycle tell, setting up a potential supply glut risk for 2027–2028. 4. **Samsung Leverages Vertical Integration to Leapfrog Competitors in HBM4E and HBM5** (`samsung-vertical-integration-hbm4e-hbm5-leap`): - Samsung is fighting back against SK Hynix's ~57% HBM market share by leveraging its unique in-house vertical integration (memory + foundry logic + packaging). - Samsung began shipping **12-layer HBM4E samples** in late May 2026, using its sixth-generation 10nm-class (**1c DRAM**) process and its own **4nm foundry logic base die**. - At Computex 2026, Samsung unveiled its **HBM5** mockup, which will use its own **2nm foundry process** for the base die to solve crucial heat dissipation and power efficiency bottlenecks in stacks up to 20 layers tall. 5. **AI Memory Bottlenecks Drive Epic Valuation Re-rating for Semiconductor Equipment Leaders** (`semiconductor-equipment-re-rating-hbm-complexity-gains`): - The complexity of HBM dramatically increases equipment intensity. Producing HBM consumes **3x the wafer area** of standard DRAM and involves **19 manufacturing steps (15 of which require advanced equipment)**. Applied Materials (AMAT) captures over half of this value. - AMAT expects its advanced packaging revenue to surge by **over 50% in 2026**. - Equipment makers are seeing their visibility shift from 3–6 months to up to **eight quarters (2 years)**, re-rating them as stable infrastructure providers. AMAT ($389.08B market cap, 46.1 P/E) and Lam Research ($418.20B market cap, 63.34 P/E) are both trading at all-time 52-week highs. ### Future Research Thread Opened: - **Samsung HBM Nvidia Qualification and 2027-2028 Supply Glut Risk**: To track Samsung's HBM3E/HBM4E qualification progress with Nvidia and monitor whether the massive global capex additions risk oversupplying the market in late 2027 or 2028.