Cycle on Jul 15, 2026

I will begin by listing the open threads of interest for this topic to understand what specific questions or watches need to be addressed during this cycle.

I will call `market_view` for Nvidia (NVDA) first to get the latest public-market numbers and financial trajectory. This will serve as our core baseline.

I see that TSMC is reporting its Q2 2026 earnings on Thursday, July 16, 2026 (tomorrow, given the current date of July 15, 2026). Also, Taiwan's semiconductor ecosystem logged full-sweep gains in June 2026. This is highly relevant to our TSMC Q2 2026 Earnings watch (`019f1cf3-530f-7f29-a5c5-dbe8ad17023d`). Let's run a web search to check the current status of Nvidia's AI capex demand, hyperscaler capex updates for Q2/July 2026 earnings, any news on TSMC's earnings expectations or pre-releases, and what the overall narrative is regarding Blackwell/Rubin demand and whether the capex story remains intact in mid-July 2026.

I want to search for TSMC's Q2 2026 earnings expectations or any pre-announced figures, as well as the latest reports on hyperscaler capex projections for July 2026 earnings. Let's do a search specifically targeting TSMC Q2 2026 earnings, capex guidance, and monthly revenue data.

Let's fetch the page about TSMC's expected guidance and capex increase at its Q2 meeting to see what the analysts are saying, particularly around TSMC's CoWoS packaging capacity expansion, N5/N3/N2 demand, and full-year outlook. I'll fetch `https://seekingalpha.com/news/4612235-tsmc-expected-to-increase-guidance-capex-at-q2-analyst-meeting-gf-securities` or the MSN/TechTimes articles. Let's do a fetch on the Seeking Alpha article. Wait, let's fetch the MSN article or look for more detailed articles. Let's try `https://www.techtimes.com/articles/320142/20260711/tsmc-q2-earnings-july-16-three-cowos-signals-that-test-ais-spending-ceiling.htm` first to see if it's fetchable, or the Seeking Alpha one. Let's try Seeking Alpha first. Since Seeking Alpha might have paywalls, let's search and fetch. Let's search for "TSMC Q2 earnings July 16 three CoWoS signals" or similar. Let's fetch `https://www.techtimes.com/articles/320142/20260711/tsmc-q2-earnings-july-16-three-cowos-signals-that-test-ais-spending-ceiling.htm` first. Let's see if we can read it. Let's check `https://investor.tsmc.com/english/quarterly-results/2026/q2` or search for the June monthly revenue. Let's run a search.

This is extremely important. Let's look at the TSMC monthly revenue YoY growth rate watch (`019ecfb3-79b3-76db-aeec-0bc8d0635c41`). It tracks whether TSMC's monthly revenue year-over-year growth drops below 20%. In June 2026, TSMC's consolidated net revenue reached NT$442.68 billion, up 67.9% year-on-year. This is well above the 20% trigger. Let's fetch the CNBC article `https://www.cnbc.com/2026/07/13/tsmc-june-revenue-rises-percent-ahead-second-quarter.html` or the Yahoo Finance / The Street article to get more details on TSMC's Q2 revenue beat relative to its own guidance. Let's do `web_fetch` on `https://www.cnbc.com/2026/07/13/tsmc-june-revenue-rises-percent-ahead-second-quarter.html`.

That's very clear. Let's do another search to see if we have more details about TSMC's Q2 revenue calculation. If June was NT$442.68B, what were April and May's revenues? Let's search for "TSMC April 2026 revenue" and "TSMC May 2026 revenue" or "TSMC Q2 2026 revenue total in USD". Let's run a search.

Let's fetch the Reuters article `https://www.reuters.com/world/asia-pacific/tsmc-q2-revenue-jumps-36-year-earlier-beating-market-expectations-2026-07-13/` to see the actual Q2 2026 revenue numbers and how they beat expectations. Let's do `web_fetch`.

This is fantastic. Let's look at the details: - TSMC's second-quarter revenue rose 36% from a year earlier to NT$1.27 trillion ($39.62 billion), which is slightly above LSEG SmartEstimate of NT$1.264 trillion and near the top-end of TSMC's own guidance of $39.0 billion to $40.2 billion. - For June alone, revenue rose 67.9% year-on-year to NT$442.68 billion, which is a 6.2% increase from May. - This confirms that TSMC's June revenue YoY growth was 67.9%, which is well above the 20% trigger. - The TSMC Q2 2026 Earnings conference is scheduled for tomorrow, July 16, 2026. This is the date of our watch `019f1cf3-530f-7f29-a5c5-dbe8ad17023d`. Wait, let's explore if there are any other updates or news on the AI capex story. Let's see: Is Nvidia's AI capex story still intact? - Let's check hyperscaler capex updates. Are they still planning massive physical AI infrastructure investments? - Let's search for "hyperscaler capex projections Q2 2026" or "AI capex July 2026". Let's run a search.

Let's fetch the Yahoo Finance article `https://finance.yahoo.com/sectors/technology/article/meta-microsoft-amazon-and-alphabet-are-about-to-spend-a-shocking-amount-of-money-to-dominate-the-ai-era-115359575.html` to read the details on Goldman Sachs' updated capex numbers and expectations. Let's do `web_fetch`.

This is an excellent source. Let's look at the numbers. - Combined capex for Google, Amazon, Microsoft, and Meta is expected to reach $725 billion in 2026. - This is up 77% from 2025's $410 billion. - Amazon is projecting $200 billion in capex. - Alphabet is targeting $175 billion to $185 billion. - Meta is guiding $115 billion to $135 billion. - Microsoft is tracking toward $190 billion. - Goldman Sachs (strategist Amanda Lynam) expects a combined $5.3 trillion of capex spending for the four largest hyperscalers from FY2025 to FY2030 (up from $4.5 trillion prior to Q1 earnings). - The baseline aggregate capex estimate stands at $7.6 trillion between 2026 and 2031 across compute, data centers, and power. This confirms that the hyperscaler capital expenditure story is not just intact; it has *risen* in scale and projections as of June/July 2026. Let's check if there are any other specific updates regarding Blackwell or Rubin delays. In the previous find, there was a mention: "Nvidia's Blackwell successor, the Rubin GPU platform, is seeing downward revisions to wafer starts due to next-generation HBM4 memory supply..." Let's search for "Nvidia Rubin GPU wafer starts HBM4" or "Rubin delay HBM4 June July 2026". Let's run a search.

Let's fetch the SDXCentral article `https://www.sdxcentral.com/news/nvidias-rubin-gpus-hit-the-brakes-as-hbm4-memory-drought-threatens-jensens-supply-chain-magic-report/` or the Network World article `https://www.networkworld.com/article/4156508/nvidia-rubin-gpus-may-be-delayed-slowing-the-next-phase-of-ai-infrastructure.html`. Let's fetch the SDXCentral one first. Let's see if we can read it. Let's do `web_fetch`.

This is a very thorough article. Let's analyze the timeline and dates. The article was published on **April 9, 2026**. It references TrendForce reports about Rubin delays and HBM4 supply. Let's see if there are more recent updates from June or July 2026 on this Rubin delay or if there are any other supply chain developments. Let's do a search for "Nvidia Rubin delay July 2026" or "Nvidia Rubin HBM4 July 2026". Let's run a search.

Let's fetch the TechTimes article `https://www.techtimes.com/articles/319410/20260701/nvidia-rubin-ultra-four-die-gpu-cancelled-packaging-limits-cut-2027-performance-half.htm` which discusses "NVIDIA Rubin Ultra Four-Die GPU Cancelled" due to manufacturing constraints with TSMC's CoWoS-L advanced packaging. This was published around **July 1, 2026**. This is extremely timely and highly relevant! Let's do `web_fetch`.

This is an absolutely critical, highly detailed, and deeply sourced piece of evidence. Let's process what this means for Nvidia's AI capex story. - **The Core Finding:** The original four-die single-package design of Nvidia's flagship **Rubin Ultra** GPU (unveiled at GTC 2026) has been **cancelled** due to manufacturing constraints that TSMC cannot currently solve, specifically **CoWoS-L substrate warpage** at the four-die scale. - **The Pivot:** Instead of a single-package four-die design, Nvidia is pivoting to a dual-die design (same chiplet count as standard Rubin) paired with HBM4E memory (delivering 384 GB of capacity per GPU, up from 288 GB in standard Rubin). To achieve the quad-die count at the rack level, Nvidia is expected to use a "2+2" board-level configuration in the Kyber rack system. - **The Implications:** 1. This cuts the performance of the single-package Rubin Ultra in half compared to what was announced at GTC 2026. 2. It exposes a physical ceiling in TSMC's CoWoS-L packaging technology (warpage at the 4-die, 7.5x to 8x reticle limit). Its successor, CoPoS (Chip-on-Panel-on-Substrate), which uses glass or sapphire materials to reduce thermal warpage, won't reach mass production until late 2028 or early 2029. 3. It reduces aggregate HBM4E demand, which affects memory suppliers like SK Hynix and Samsung, who accelerated HBM4E qualification specifically for Rubin Ultra. 4. It narrows the performance gap between Nvidia and AMD's Instinct MI500 (also targeting 2027 with HBM4E). 5. The standard Rubin GPU remains on track for mass shipments in the summer of 2026 to eight confirmed cloud partners (AWS, Google, Azure, Oracle, CoreWeave, Lambda, Nebius, Nscale). Wait, let's look at the prior findings that we can refine. We have: - `nvidia-ai-capex-cycle-sustainability`: "Nvidia's AI Capex Story Remains Robustly Intact Amid Soaring Projections and Kyber Rack Delay Debates" - `enterprise-ai-roi-gap-revenue-reckoning`: "The Hyperscaler Capital Crossover: AI Capex Outstrips Cash Flow, Triggering Historic Debt and Equity Waves" Let's refine both of these or write a highly specific note on the Rubin Ultra cancellation under a stable slug like `nvidia-rubin-ultra-cancellation-packaging-limits`. But wait, the instructions say: "One finding = ONE coherent subject tracked under one stable slug across cycles — a developing story, a running debate, or an entity's trajectory — NOT a fresh single-story-of-the-day note each cycle. Name a slug for the DURABLE SUBJECT it tracks: the throughline, entity, or running debate (`ai-coding-backlash`, `nvidia-datacenter-margins`) — never the transient story-of-the-day (`sp-h-c99-library`, `q3-earnings-beat`), which can never recur. When this cycle advances a subject you already have a finding for, call `write_note` with that SAME slug to revise it in place — an upsert that snapshots a new revision. Mint a fresh slug ONLY for a genuinely new subject. Before you write, scan your prior findings (the injected block + `read_notes`) for the slug that already owns this subject and reuse it;" Wait, let's read the prior findings to see if there is an existing note that matches this. Let's call `read_notes` to see all our notes. Let's do that. This is the best way to check if there is an existing note tracking Nvidia's product roadmap, delays, or advanced packaging issues.

Let's analyze the existing notes. We have: - `nvidia-ai-capex-cycle-sustainability`: "Nvidia's AI Capex Story Remains Robustly Intact Amid Soaring Projections and Kyber Rack Delay Debates" - `nvidia-q1-2027-record-financials-agentic-ai`: "Nvidia's Record Financial Trajectory: S&P Upgrade to 'AA', $25B Jumbo Bond Sale, and Relentless Blackwell Demand" - `custom-silicon-surge-openai-jalapeno-qualcomm-dragonfly`: "The Custom Silicon Surge: OpenAI's Jalapeño and Qualcomm's Dragonfly C1000 Challenge Nvidia's Moat" Let's see: The user wants to know "Is Nvidia's AI capex story still intact?" This is directly addressed by `nvidia-ai-capex-cycle-sustainability`! We can update this note to integrate the latest data points: 1. **TSMC's blowout June 2026 revenue and Q2 2026 revenue beat:** TSMC reported a 67.9% YoY revenue surge in June 2026 to NT$442.68 billion, bringing Q2 revenue to NT$1.27 trillion ($39.62 billion), beating LSEG SmartEstimates and landing near the top end of its own guidance. This confirms that the physical semiconductor supply chain and AI chip demand remain extremely robust. 2. **Hyperscaler Capex Projections:** Goldman Sachs (strategist Amanda Lynam) updated hyperscaler capex projections in June 2026, expecting a massive combined $5.3 trillion from FY2025 to FY2030 (up from $4.5 trillion prior to Q1 earnings). Combined capex for Google, Amazon, Microsoft, and Meta is expected to reach $725 billion in 2026 alone (up 77% YoY). 3. **The Rubin Ultra Cancellation/Redesign:** On June 30, 2026, SemiAnalysis confirmed that the original four-die single-package design of Nvidia's flagship 2027 Rubin Ultra GPU has been cancelled due to TSMC's CoWoS-L substrate warpage limits. Nvidia is pivoting to a dual-die design with HBM4E memory, using a "2+2" board-level configuration in the Kyber rack system to achieve the quad-die count. This highlights a physical bottleneck in advanced packaging, narrowing the performance gap with AMD's MI500 and reducing HBM4E demand, but standard Rubin is still shipping in summer 2026. This is a perfect way to update `nvidia-ai-capex-cycle-sustainability`! Let's see if there are other notes we should update or if there are other things to watch. Wait, let's verify if TSMC's Q2 earnings conference watch or any other watch has been triggered. Let's look at the watches we retrieved earlier: 1. `019f1cf3-530f-7f29-a5c5-dbe8ad17023d`: "TSMC Q2 2026 Earnings Conference" on July 16, 2026. Wait, today's date in the context is **2026-07-15**. The earnings conference itself takes place tomorrow, July 16, 2026. However, TSMC has already reported its June monthly sales and Q2 revenue figures (on July 13, 2026, as per the Reuters article). Wait, does the watch resolve now or should we wait for the actual July 16 conference? The watch title is: "TSMC Q2 2026 Earnings Conference" and the description is "Track TSMC's Q2 2026 earnings results and updated full-year revenue/capex guidance on July 16, 2026". Since today is July 15, we are still one day before the conference itself. We can't record the event as completed yet because the actual conference and capex guidance update happen tomorrow. But we can note that the pre-release Q2 revenue beat has already been announced and is highly bullish. Wait, let's look at another watch: - `019ecfb3-79b3-76db-aeec-0bc8d0635c41`: "TSMC Monthly Revenue YoY Growth Rate Track". Description: "Track TSMC's monthly revenue year-over-year growth rate. A drop below 20% YoY would signal that the physical semiconductor supply chain is finally catching up with demand or that hyperscaler chip orders are starting to soften." Wait, TSMC's June revenue growth rate was 67.9% YoY. This does not drop below 20%. So the watch does not fire (it fires if it drops below 20%). Since it's an ongoing watch, we don't resolve it, but we can keep tracking it. Let's check if there are other watches: - `019eb0cd-20a0-71c6-b91e-92d528f6792e`: "Hyperscaler Capex Guidance Track (July 2026 Earnings)". Description: "Track the quarterly capex guidance and actual capex spending reported by the major AI hyperscalers (Microsoft, Meta, Alphabet, and Amazon) during the July 2026 earnings calls. This is the single most critical forward-looking signal for the AI capex story." This has a fire date of July 31, 2026. It hasn't fired yet since today is July 15, 2026. - `019ee44e-b944-74a7-bc73-b87fb1bf1ac5`: "Aggregate Hyperscaler Free Cash Flow Crossover (Q3 2026)" - fire date October 31, 2026. - `019edf29-c3f7-7561-bccb-ef09f3a7fb09`: "Nvidia Q2 FY2027 Revenue Guidance Cross" - fire date August 25, 2026. - `019f4b49-2c52-7a54-9f80-a4fb82cea464`: "Nvidia Q2 FY2027 Earnings - Rubin Ultra / Kyber NVL144 Timeline Update" - fire date August 25, 2026. Wait, has there been any regulatory action or other events? Let's check if there are other news. Let's write the updated note for `nvidia-ai-capex-cycle-sustainability`. Let's first read the current body of `nvidia-ai-capex-cycle-sustainability` using a web or note reading tool if possible, but wait, `read_notes` only gives us a preview. We don't have a direct tool to read the full body of an existing note unless we search for it or read it. Wait, `read_notes` does not return the full body, it returns slug, title, preview, updated_at, and source_count. Ah! Is there a way to read the note? No, there isn't a specific tool to read the full body of a note, but we can overwrite/update it by writing the full body with the same slug. Let's make sure we write a comprehensive, high-quality note that preserves the core story and adds the new mid-July 2026 findings (TSMC June sales, Goldman's updated $5.3T hyperscaler capex, and the SemiAnalysis Rubin Ultra four-die cancellation/redesign). Let's check if we can search for the existing note's content or if we should write a beautifully structured note that covers both the robust demand side (TSMC, hyperscaler capex) and the supply-chain/packaging bottleneck side (Rubin Ultra cancellation). Let's do a search on TSMC Q2 2026 and Nvidia's capex story to see if there is any other major news we should include. Let's search for "Nvidia AI capex July 2026" to get any additional angles.

Let's fetch the Futurum Group article `https://futurumgroup.com/insights/ai-capex-2026-the-690b-infrastructure-sprint/` to read about their insights on the sustainability of the AI capex cycle in 2026. This will give us excellent context. Let's do `web_fetch`.

Ah, the fetch failed with a timeout. Let's try to fetch another article, or we can use our existing knowledge from other sources. Let's try searching for "AI Capex 2026: The $690B Infrastructure Sprint" or search for "sustainability of AI capex 2026" and see if there are other articles we can fetch. Wait, let's look at the Yahoo Finance article we fetched earlier: - Goldman Sachs expects a combined $5.3 trillion of capex spending for the four largest hyperscalers (Meta, Microsoft, Amazon, and Alphabet) from FY2025 to FY2030, up from $4.5 trillion. - The baseline aggregate capex estimate stands at $7.6 trillion between 2026 and 2031 across compute, data centers, and power. - Google, Amazon, Microsoft, and Meta collectively plan to allocate $725 billion to capital expenditures in 2026 alone, up 77% from last year's $410 billion. - Amazon is projecting $200 billion, Alphabet $175 billion to $185 billion, Meta $115 billion to $135 billion, and Microsoft $190 billion. And from TSMC's pre-release: - TSMC's Q2 2026 revenue rose 36% YoY to T$1.27 trillion ($39.62 billion), beating LSEG SmartEstimates. - TSMC's June 2026 revenue rose 67.9% YoY to T$442.68 billion, up 6.2% MoM. - This represents the fastest monthly growth in 2026 and breaks a four-year seasonal pattern, confirming relentless demand for AI accelerators (Nvidia, AMD, etc.). - TSMC is adding advanced packaging facilities (Chiayi Science Park) to expand CoWoS capacity. And on the Rubin Ultra redesign: - SemiAnalysis confirmed on June 30, 2026, that the four-die single-package design of Nvidia's flagship Rubin Ultra (originally planned for 2027) has been cancelled due to CoWoS-L substrate warpage limits at the 4-die scale. - Nvidia is pivoting to a dual-die design with HBM4E memory, using a "2+2" board-level configuration (Kyber rack) to achieve the quad-die count at the rack level rather than the package level. - This represents a physical bottleneck in advanced packaging (CoWoS-L hitting a physical wall at 4-die scale) and narrows the performance gap with AMD's Instinct MI500. - Standard Rubin remains on track for summer 2026 shipments. These three pillars together provide a complete, nuanced, and exceptionally detailed answer to the question: "Is Nvidia's AI capex story still intact?" The answer is: **Yes, the demand-side capex story remains robust and has actually expanded in scale, but the supply-side technology has hit a physical wall in advanced packaging, forcing a major redesign of Nvidia's flagship 2027 Rubin Ultra GPU.** Let's write a comprehensive update to the note `nvidia-ai-capex-cycle-sustainability`. Wait, let's review the required body shape for `write_note`: - First level heading `# <Title>` - Verbatim quotes kept with their links - Sources block at the bottom - No YAML frontmatter - Slug: `nvidia-ai-capex-cycle-sustainability` Let's carefully construct the markdown body for the note. I will write a highly professional, detailed update that integrates these new July 2026 developments. Let's draft the note. Wait, let's double check if there are any other notes we should write. Is there any other theme? We have: 1. **The physical packaging wall:** Since this is a massive technical shift (the cancellation of the four-die Rubin Ultra package and TSMC's CoWoS-L warpage issues), does it deserve its own separate finding, or should it be integrated into `nvidia-ai-capex-cycle-sustainability`? The instructions say: "For each distinct theme or story worth persisting, call `write_note` — ONE note per theme, never a combined write-up... Name a slug for the DURABLE SUBJECT it tracks... Before you write, scan your prior findings for the slug that already owns this subject and reuse it." The subject of Nvidia's AI capex sustainability, chip roadmaps, and physical supply chain/rack delays is already owned by `nvidia-ai-capex-cycle-sustainability`. The title of that note is: "Nvidia's AI Capex Story Remains Robustly Intact Amid Soaring Projections and Kyber Rack Delay Debates". The Rubin Ultra redesign and packaging limits directly address the supply-chain and physical roadmap debates of Nvidia's capex story, so they fit perfectly within this note. We can update this note to give a balanced, comprehensive view of the capex story: robust demand (TSMC sales, Goldman's $5.3T projections) vs. physical limits (Rubin Ultra packaging cancellation). Wait, let's also check if there are any other public players we should pull. Let's look at TSMC. Can we pull TSM? Yes, let's do a `market_view` for TSM to get its latest market numbers as well. This will ground our TSMC figures. Let's do that.

This is extremely helpful. We have TSM's market view as of July 11, 2026. Note that the numbers are in NT$ for local reporting, but we can cite both USD and NT$ figures. Let's see: TSM has a market capitalization of $2.27 trillion (USD) as a company, and its TTM revenue is $4.10T NTD. Its latest quarter reported (ending March 31, 2026) showed revenue of NT$1.13 trillion and net income of NT$572.48 billion. Let's make sure we cite these views: - `/markets/NVDA/2026/07/15` - `/markets/TSM/2026/07/11` Let's write a highly detailed update to `nvidia-ai-capex-cycle-sustainability`. Wait, let's verify if there are other threads or watches we should register. Let's review our watches: - `019f1cf3-530f-7f29-a5c5-dbe8ad17023d`: "TSMC Q2 2026 Earnings Conference" on July 16, 2026. Wait, since today is July 15, we can't resolve it yet. But we can register a new watch if we find something forward-looking. Wait, is there any new forward-looking thing? Yes! The SemiAnalysis article mentions that **TSMC's CoPoS (Chip-on-Panel-on-Substrate)** pilot lines are expected by the end of 2026, and mass production is targeted for late 2028 or early 2029. We could register a watch for the first commercial pilot production of TSMC's CoPoS packaging technology, which is a key milestone for resuming quad-die single-package scaling in future GPU architectures (such as the successor to Rubin, or a revised Rubin Ultra). Let's see: - Card: `one_shot` - Entity: `TSM` - Title: `TSMC CoPoS Packaging Pilot Line Launch` - Description: `Track TSMC's launch of its first CoPoS (Chip-on-Panel-on-Substrate) pilot line, utilizing glass or sapphire substrates to overcome CoWoS-L substrate warpage limits.` - Fire date: `2026-12-31T00:00:00Z` - Watch type: `dated_catalyst` Let's look at another watch we could register: - **AMD Instinct MI500 launching with HBM4E in 2027**: Since AMD's MI500 is directly competing with Nvidia's revised dual-die Rubin Ultra in 2027, we can register a watch for the first official benchmark comparison or commercial launch of the AMD Instinct MI500 to see if it narrows the performance gap. - Card: `one_shot` - Entity: `AMD` - Title: `AMD Instinct MI500 Launch and Rubin Ultra Comp` - Description: `Track the official commercial launch of AMD's Instinct MI500 GPU with HBM4E in 2027 and its first head-to-head performance benchmarks against Nvidia's revised dual-die Rubin Ultra.` - Fire date: `2027-06-30T00:00:00Z` - Watch type: `dated_catalyst` Wait, let's look at the existing watches. Are there any watches that we can record an event for? Let's check if any of them have fired. - `019f5599-dcfb-788b-a861-4a9c86598587`: "Low-Cost Chinese Model Breakthrough (Thesis Breaker)" - Chinese open-source models. No event yet. - `019f4b49-2c52-7a54-9f80-a4fb82cea464`: "Nvidia Q2 FY2027 Earnings - Rubin Ultra / Kyber NVL144 Timeline Update" - fire date August 25, 2026. Not fired. - `019f2c65-b0bd-7794-add6-7b6b15ef3fff`: "First Commercial Shipment of Nvidia Rubin GPUs" - fire date December 31, 2026. Not fired. - `019f1cf3-530f-7f29-a5c5-dbe8ad17023d`: "TSMC Q2 2026 Earnings Conference" - fire date July 16, 2026. Today is July 15, 2026, so it hasn't happened yet. - `019f17cc-fae4-7cda-927b-fb7b20231712`: "Official Launch of ICE GPU Compute Futures Contracts". Not fired. - `019efe0e-ea69-7ac4-9af7-50d0d58f3a57`: "Micron Quarterly Capex FY2027 Track". Not fired. - `019ef3bf-47e0-7795-ba54-6be76a9a84e7`: "First Commercial Deployment of Nvidia-LG Co-Developed Smart Factory Standard". Not fired. - `019ee44e-b944-74a7-bc73-b87fb1bf1ac5`: "Aggregate Hyperscaler Free Cash Flow Crossover (Q3 2026)" - fire date October 31, 2026. Not fired. - `019edf29-c3f7-7561-bccb-ef09f3a7fb09`: "Nvidia Q2 FY2027 Revenue Guidance Cross" - fire date August 25, 2026. Not fired. - `019ed4d9-3d3f-7390-b1e8-77f8753e9989`: "Kazakhstan Data Center Valley Phase 1 Groundbreaking" - 2027. Not fired. - `019ed4d9-3256-78b5-94ac-9e710b346c1d`: "Helix Digital Infrastructure Project Commitments Track". Not fired. - `019ecfb3-79b3-76db-aeec-0bc8d0635c41`: "TSMC Monthly Revenue YoY Growth Rate Track". The trigger is "drops below 20%". June 2026 was 67.9% YoY, so it did not drop below 20%. Ongoing. - `019ec568-ed82-7d7b-a8d0-36e0e5717e8a`: "Ornn Compute Price Index Blackwell GPU Rental Rate Track". Not fired. - `019ec040-d231-7dc2-9387-902ca1d82c1d`: "Nvidia Gross Margin Track". NVDA gross margin is 74.1% (per `market_view`). Not fired. - `019ebb1c-c3cb-7db0-b267-77ac380cc943`: "Regulatory Scrutiny of Off-Balance-Sheet AI SPVs". Not fired. - `019eb0cd-20a0-71c6-b91e-92d528f6792e`: "Hyperscaler Capex Guidance Track (July 2026 Earnings)" - fire date July 31, 2026. Not fired. - `019eaba5-b004-719f-8669-34a8aae85368`: "Nvidia Standalone Vera CPU Revenue Track". Not fired. - `019e9c33-1f68-7b87-b126-0aa020f67c7f`: "Apple Revamped Siri Launch with Blackwell Workloads" - fire date September 15, 2026. Not fired. Wait, since none of the watches have officially fired yet, we don't call `record_event` or `resolve_thread`. But we are actively tracking them. Let's write our updated finding under slug `nvidia-ai-capex-cycle-sustainability`. Let's draft the note body first. ### Draft of Note Body: `nvidia-ai-capex-cycle-sustainability` Title: Nvidia's AI Capex Story Remains Robustly Intact Amid Soaring Projections and Kyber Rack Delay Debates Content outline: 1. **Executive Summary:** In mid-July 2026, the structural demand driving Nvidia's AI dominance remains robustly intact, supported by staggering new hyperscaler capex projections and record-breaking supply chain performance from TSMC. However, the physical limits of semiconductor packaging have emerged as a critical supply-side bottleneck, forcing Nvidia to cancel its highly anticipated four-die single-package Rubin Ultra GPU design. 2. **The Demand Side: Staggering Hyperscaler Capex Projections:** - Cite Goldman Sachs' June 2026 update: Amanda Lynam projects a combined $5.3 trillion in capex from the four largest hyperscalers (Meta, Microsoft, Amazon, Alphabet) between FY2025 and FY2030 (up from $4.5 trillion previously). - In 2026 alone, combined capex for these four giants is projected to reach $725 billion, a 77% YoY surge from $410 billion in 2025. - Breakdown: Amazon ($200B), Microsoft ($190B), Alphabet ($175B–$185B), Meta ($115B–$135B). - Link: [[enterprise-ai-roi-gap-revenue-reckoning]] which tracks the ROI gap and crossover. 3. **The Supply Chain: TSMC's Blowout June & Q2 2026 Revenue:** - Cite TSMC's June 2026 monthly sales report (announced July 13, 2026): Consolidated net revenue surged 67.9% YoY to NT$442.68 billion ($13.2 billion), representing a 6.2% sequential increase from May. This is the fastest monthly growth in 2026 and breaks a four-year seasonal pattern. - For Q2 2026 (April–June), TSMC's total revenue rose 36% YoY to a record NT$1.27 trillion ($39.62 billion), beating LSEG SmartEstimates and landing near the top end of TSMC's guidance of $39.0B to $40.2B. - TSMC is aggressively expanding CoWoS capacity, adding two advanced chip packaging plants in Chiayi Science Park, with the first already in mass production. - Link: `/markets/TSM/2026/07/11` and `/markets/NVDA/2026/07/15` 4. **The Supply Side Bottleneck: Rubin Ultra Four-Die Package Cancelled:** - Introduce the SemiAnalysis confirmation (published July 1, 2026) that Nvidia has cancelled the original four-die single-package design of the flagship Rubin Ultra GPU (originally announced at GTC 2026 for a 2027 release). - Explain the technical bottleneck: TSMC's CoWoS-L packaging has hit a physical wall at the four-die scale (requiring an interposer 7.5x to 8x the reticle limit, causing severe substrate warpage and yield-killing electrical failures under thermal cycling). - The successor packaging technology, CoPoS (Chip-on-Panel-on-Substrate), which uses glass or sapphire to reduce thermal expansion mismatch, won't reach mass production until late 2028 or early 2029. - Nvidia's pivot: A revised dual-die Rubin Ultra paired with HBM4E memory (providing 384 GB capacity per GPU, up from 288 GB in standard Rubin). To approximate the performance of the quad-die package at the rack level, Nvidia will use a "2+2" board-level configuration in its Kyber rack system. - Competitive impact: Narrows the performance advantage over AMD's Instinct MI500 (also launching in 2027 with HBM4E) and decreases aggregate HBM4E demand for suppliers like SK Hynix and Samsung. - Standard Rubin remains on track for summer 2026 shipments. 5. **Sources Block:** Include links and quotes. Let's verify the exact quotes we can use: From the CNBC article: - "Taiwan Semiconductor Manufacturing Co. (TSMC) reported a 67.9% year-on-year rise in its June sales on Monday, ahead of its second-quarter earnings release later this week." - "For the first half of 2026, TSMC’s total revenue reached 2.4 trillion new Taiwan dollars ($74.99 billion), representing a 35.6% increase compared to the same period in 2025." From the Reuters article: - "TSMC, the world's largest contract chipmaker, reported on Monday second-quarter revenue that rose 36% from a year earlier to a record high on surging interest in artificial intelligence applications." - "Revenue in the April-June period of this year came in at T$1.27 trillion ($39.62 billion), according to Reuters calculations, slightly above a T$1.264 trillion LSEG SmartEstimate drawn from 20 analysts." From the Yahoo Finance (Brian Sozzi) article: - "Goldman now expects a combined $5.3 trillion of capex spending for the four largest hyperscalers — Meta (META), Microsoft (MSFT), Amazon (AMZN), and Alphabet (GOOGL) — from fiscal year 2025 to fiscal year 2030." - "Google, Amazon, Microsoft, and Meta alone collectively plan to allocate $725 billion to capital expenditures in 2026 — up a staggering 77% from last year's already record-breaking $410 billion." From the TechTimes (Richard L. Wells) article: - "Semiconductor research firm SemiAnalysis confirmed on June 30 that the original four-die version of the Rubin Ultra GPU — unveiled with considerable fanfare at GTC 2026 just three months ago — has been cancelled due to manufacturing constraints TSMC cannot yet solve. The GPU that will actually ship in 2027 under the Rubin Ultra name will deliver roughly half the compute and half the memory bandwidth of what was announced." - "The cancellation also signals something that reaches beyond NVIDIA's product roadmap: TSMC's CoWoS-L advanced packaging technology has hit a physical ceiling at the four-die scale, and its successor — CoPoS — is not expected to enter mass production until late 2028 at the earliest." These are perfect, highly specific, and authoritative quotes. Let's write the note now! This is a substantial update to `nvidia-ai-capex-cycle-sustainability`. Let's perform the write_note call.

I will write a comprehensive update to the existing note tracking Nvidia's AI capex sustainability (`nvidia-ai-capex-cycle-sustainability`). This update integrates the latest mid-July 2026 data points: TSMC's record-breaking Q2 revenue beat, Goldman Sachs' updated $5.3 trillion hyperscaler capex projections, and the significant technical setback of Nvidia's flagship 2027 Rubin Ultra GPU redesign due to physical packaging limits. Following this, I will register a couple of crucial forward-looking watches and finish the cycle.

I will register a couple of important forward-looking watches that directly emerge from this cycle's findings: 1. **TSMC CoPoS Packaging Pilot Line Launch:** This tracks whether TSMC can successfully build and scale its next-generation glass/sapphire packaging technology, which is the key to resuming quad-die single-package scaling in the future. 2. **AMD Instinct MI500 Launch and Rubin Ultra Comp:** This tracks the competitive dynamics in 2027 between AMD's Instinct MI500 and Nvidia's revised dual-die Rubin Ultra, which have a narrowed performance gap due to the cancellation. Let's register the first watch.

Now, let's register the second watch.

Now that all findings have been written and updated, and new watches have been recorded to track future catalysts, I will call `finish_cycle` to complete this research cycle.

I have successfully completed this research cycle. ### Summary of Findings: Is Nvidia's AI Capex Story Still Intact? Yes, the **demand-side capex story remains robustly intact and has actually expanded in scale**, but the **supply-side technology has hit a physical wall in advanced packaging**, forcing a major redesign of Nvidia's flagship 2027 Rubin Ultra GPU. Here is the breakdown of the most recent mid-July 2026 developments: 1. **Unprecedented Hyperscaler Capex Projections:** Goldman Sachs (strategist Amanda Lynam) updated its hyperscaler capital expenditure projections in June 2026, forecasting a combined **$5.3 trillion** in capex for the four largest hyperscalers (Meta, Microsoft, Amazon, and Alphabet) from FY2025 to FY2030 (up from $4.5 trillion prior to Q1 earnings). In 2026 alone, combined capex for these four giants is projected to reach **$725 billion**, representing a staggering 77% YoY surge from 2025's $410 billion. 2. **TSMC's Blowout June and Q2 2026 Revenue:** TSMC's June 2026 monthly sales report (announced July 13, 2026) showed a **67.9% YoY surge** to NT$442.68 billion ($13.2 billion), representing a 6.2% sequential increase from May and breaking a four-year seasonal pattern. For Q2 2026, TSMC's total revenue rose 36% YoY to a record NT$1.27 trillion ($39.62 billion), beating LSEG SmartEstimates and landing near the top end of TSMC's guidance of $39.0B to $40.2B. 3. **The Supply-Side Bottleneck — Rubin Ultra Single-Package Cancelled:** On June 30, 2026, semiconductor research firm SemiAnalysis confirmed that the original four-die, single-package version of Nvidia's flagship **Rubin Ultra GPU** (originally announced at GTC 2026 for a 2027 release) has been **cancelled** due to manufacturing constraints that TSMC cannot currently solve. - **The Advanced Packaging Limit:** TSMC's CoWoS-L packaging has hit a physical ceiling at the four-die scale. Assembling more than 3,400 square millimeters of active silicon onto a single interposer introduces severe substrate warpage under thermal cycling, causing electrical failures and yield-killing defects. TSMC's successor packaging technology, CoPoS (Chip-on-Panel-on-Substrate), which uses glass or sapphire to reduce thermal expansion mismatch, is not expected to enter mass production until late 2028 or early 2029—too late for the 2027 Rubin Ultra timeline. - **Nvidia's Dual-Die Pivot:** Nvidia is pivoting to a dual-die design—the same chiplet count as the standard Rubin—paired with HBM4E memory (providing 384 GB of capacity per GPU, up from 288 GB in standard Rubin). To achieve the quad-die count at the rack level, Nvidia will use a "2+2" board-level configuration in its Kyber rack system. This achieves the four-die count at the rack level, which is much easier to manufacture but introduces a latency tradeoff. - **Competitive Impact:** This redesign narrows the performance advantage over AMD's Instinct MI500 (also launching in 2027 with HBM4E) and decreases aggregate HBM4E demand for memory suppliers like SK Hynix and Samsung. However, the standard Rubin GPU remains on track for mass shipments in the summer of 2026 to eight confirmed cloud partners, confirming that near-term capex deployment is secure. I have updated the living note `nvidia-ai-capex-cycle-sustainability` with these findings and registered new watches to track the future launch of TSMC's CoPoS glass-substrate packaging pilot lines as well as the head-to-head performance benchmarks of the AMD Instinct MI500 vs. the revised dual-die Rubin Ultra in 2027.