Cycle on Jul 21, 2026

I am going to query the open threads for this topic to see what questions are currently prioritized or what watches are established.

I will fetch the current public-markets view for Nvidia (NVDA) to see its valuation, recent earnings, guidance, and stock performance.

I see that the date is July 21, 2026. This means we are right in the middle of the July 2026 earnings season where major hyperscalers are reporting or about to report. Let's do a web search on hyperscaler capital expenditures (capex), Nvidia capex story, and recent earnings reports or upcoming earnings reports for Microsoft, Alphabet (Google), Meta, and Amazon. Let's see if Google, Microsoft or Meta have already reported or what the market expectations are for their capex, and what the latest news is on Nvidia's AI capex story. Let's run a search for recent news on hyperscaler capex, Google/Microsoft/Meta capex Q2 2026, and Nvidia capex demand.

This is very interesting. Let's see: 1. `https://es.tradingview.com/news/invezz:6f0652e93094b:0-big-tech-earnings-outlook-wall-street-demands-receipts-on-700b-ai-spree/` 2. `https://www.reuters.com/business/retail-consumer/among-ai-crowd-some-investors-position-slower-hyperscaler-spending-growth-2026-07-17/` 3. `https://www.investing.com/news/stock-market-news/citi-sees-google-meta-and-amazon-capex-spending-at-over-800-billion-in-2027-4795118` 4. `https://www.wsj.com/tech/ai/will-someone-finally-blink-in-the-ai-spending-war-0f59aa60` 5. `https://seekingalpha.com/article/4923508-ai-bubble-burst-alphabet-could-be-first-to-cut-ai-capex` Let's fetch/read some of these articles. Let's do a search to get the exact URL for the WSJ article or other articles about hyperscaler capex and Nvidia. Wait, let's search for "Among AI crowd, some investors position for slower hyperscaler spending growth" or "Citi sees Google, Meta and Amazon capex spending at over $800 billion in 2027" to find more details. And let's web_fetch the reuters and investing.com articles.

This is a phenomenal, highly informative article from Reuters dated July 17, 2026. Let's analyze the details: - UBS estimates hyperscalers' capex will rise 76% this year (2026) to $673 billion (note: other estimates say $725B or $710B+, but this is the specific UBS model). - However, UBS expects capex growth to slow sharply to 25% in 2027 ($843 billion) and to only 6% in 2028 ($894 billion). - The article lists: - Amazon: $128B - Microsoft: $118B - Google: $91B - Oracle: $36B - CoreWeave: $9B - (Total capex in billion $ for 2026) - Edmond de Rothschild Asset Management global equity portfolio manager Alexis Bossard says: "Once they stop increasing their capex, it will definitely be a relief for hyperscalers and a negative signal for the semi industry." They have a massive underexposure to semis right now. - LFG+ZEST CIO Alberto Conca has sharply cut/hedged semiconductor positions, saying "Cash flow is starting to be almost completely drained by capex." - Apollo Chief Economist Torsten Slok notes that cover ratios in the corporate debt market (for Big Tech issuance) have fallen to below 2x in July 2026, from nearly 5x in February 2026. This indicates that the corporate debt market's capacity to absorb Big Tech's capex funding is showing signs of fatigue. - In June 2026, the Bank for International Settlements (BIS) warned that disappointment in returns could trigger a sudden pullback in financing and turn the capex boom into a protracted bust. - Empirical Research estimates about 70% of data center projects face some degree of pushback. New York state imposed a one-year moratorium on July 14, 2026 (Tuesday). Let's do a search on other recent news about Big Tech capex and the "AI CapEx story" or Nvidia's Q2 earnings expectations. Wait! Let's check when the tech giants are reporting Q2 2026 earnings. Let's search for "Alphabet Q2 2026 earnings date capex" or "Microsoft Q2 2026 earnings date capex". Note that today's date is July 21, 2026. Let's see if there are any earnings releases today or when they are scheduled. Let's search for: "GOOG Q2 2026 earnings date" or "MSFT Q2 2026 earnings date" or similar.

This is extremely interesting. Let's see: 1. Alphabet is scheduled to report Q2 2026 earnings on **Wednesday, July 22, 2026** (tomorrow!). 2. In Q1 2026, Google Cloud grew 63% to $20B, but FCF collapsed or dropped because of capex doubling. 3. Alphabet stock fell roughly 10% on June 22, 2026, as investors balked at a $180 billion AI capex program with no clear FCF recovery timeline. Wait! Let's read about this. Let's fetch `https://www.techtimes.com/articles/318889/20260623/alphabet-drops-10-ai-capex-fears-memory-chip-rally-reveals-split-market.htm` to see what happened on June 22, 2026. Wait, let's also fetch `https://www.tradingkey.com/analysis/stocks/us-stocks/262008961-alphabet-googl-stock-price-forecast-2026-google-cloud-tradingkey` or `https://www.tradingkey.com/analysis/stocks/us-stocks/262041401-alphabet-q2-earnings-preview-tradingkey` to understand what's happening. Let's do a fetch for `https://www.techtimes.com/articles/318889/20260623/alphabet-drops-10-ai-capex-fears-memory-chip-rally-reveals-split-market.htm`.

This is incredibly detailed. Let's look at the dates and facts. The article was written on June 23, 2026. It mentions: - Alphabet's stock dropped roughly 10% on Monday (June 22, 2026) in one of its sharpest single-session losses in months, as investors ran out of patience with a free cash flow trough. - Alphabet expects to spend $180 billion to $190 billion in 2026 capex, nearly double the $91 billion deployed in 2025. - Google's CFO Anat Ashkenazi told analysts on the Q1 earnings call that 2027 capital expenditure would "significantly increase" compared to 2026's already-elevated range. - Combined 2026 capital expenditure from Alphabet, Amazon, Microsoft, Meta, and Oracle now tops $452 billion (or $673 billion according to the UBS estimate from the July 17 article, or $725 billion according to the earlier note). Let's keep these numbers in perspective. - Free cash flow is starting to be drained. - Amazon's trailing free cash flow has collapsed 95% to $1.2 billion, against Q1 capex of $44.2 billion, up 77% YoY. - Talent exits: Noam Shazeer left Google for OpenAI (announced June 18). John Jumper (Nobel Prize winner, AlphaFold2) left Google DeepMind for Anthropic around the same time. - SpaceX went public on Nasdaq on June 12, 2026 at $135/share, rose to $225, but fell 16% on June 22 after announcing its first-ever bond offering of senior unsecured notes targeting $20B to repay bridge financing for its acquisition of xAI in Feb 2026. - SK Hynix overtook Samsung as South Korea's most valuable company. Wait, let's look at the date today: **July 21, 2026**. Google (Alphabet) reports Q2 2026 earnings tomorrow, **Wednesday, July 22, 2026**! Let's search for any recent updates or expectations for Alphabet's earnings tomorrow, and see if there are any other recent news items about Nvidia, Blackwell, or capex. Let's search for "Nvidia capex story" or "Nvidia Blackwell demand July 2026" or "Nvidia Q2 FY2027 preview capex". Wait, let's search for "Nvidia" or "NVDA" or "Blackwell" to see if there are any recent delays, supply chain updates, or production issues. Let's see: `web_search` with query `Nvidia Blackwell delay July 2026` or `Nvidia Blackwell production July 2026`. Let's do that.

We have found some incredible, very recent information! Let's see what is happening: On or around **July 5–6, 2026**, SemiAnalysis reported that Nvidia's next-generation **Kyber NVL144** rack-scale AI system (which uses the Vera Rubin architecture or is a 2027 product range) is delayed from its planned 2027 launch to 2028. Wait! Let's search for this report or articles explaining the SemiAnalysis report to get the exact technical details. Let's do a search for `"Kyber NVL144" SemiAnalysis` or `"Kyber" delay 2028` or `"Vera Rubin" delay SemiAnalysis` or `Nvidia Kyber Delayed to 2028, Rubin Ships Fall 2026`. Let's fetch `https://network-switch.com/blogs/news/ai-infrastructure-mid-year-update-nvidia-arista` first. Let's see if we can read it. Let's run a search to find details of this. Or search for `"Kyber" SemiAnalysis 78-layer`. This is amazing. Let's search.

This is a highly critical development. Let's see: On July 6, 2026, SemiAnalysis published a report alleging that Nvidia's **Kyber NVL144** rack-scale AI system—designed for the 2027 **Rubin Ultra** generation—has been delayed by more than 12 months to **2028**. The reason reported by SemiAnalysis is a manufacturing bottleneck involving the **midplane PCB (printed circuit board)**, which is an ultra-high-density **78-layer** board using M9-grade copper-clad laminate. Nvidia has urgently denied these rumors, with CEO Jensen Huang stating on July 15, 2026, that the next-generation systems are "in production and on track" and that the chip roadmap remains intact. Let's look at the details of this. Let's fetch `https://www.cnbc.com/2026/07/06/nvidia-kyber-rack-system-delays-manufacturing-taiwan-rubin-chips-.html` and `https://www.techtimes.com/articles/320033/20260709/nvidia-kyber-nvl144-slips-2028-physics-not-software-defeats-backup-plan.htm` to get the exact technical details and quotes. This is critical for our findings! Let's do a fetch.

The URL was slightly different in my fetch request (`nvidia-nvidia-kyber-nvl144-slips-2028-physics-not-software-defeats-backup-plan.htm` vs `nvidia-kyber-nvl144-slips-2028-physics-not-software-defeats-backup-plan.htm`) but it fetched successfully! This is fantastic. Let's see: We have two major, incredibly rich themes developing here: 1. **Hyperscaler Capex Fatigue vs. Lofty Expectations (The FCF Drain)**: - The Big Four cloud providers (Amazon, Microsoft, Google, Meta) + Oracle are expected to spend $452 billion in 2026 (or $673 billion according to UBS, which is a 76% YoY increase). - However, UBS expects capex growth to slow down significantly to 25% in 2027 ($843 billion) and to only 6% in 2028 ($894 billion). - This deceleration is a major headwind for semiconductor suppliers who have priced in relentless exponential growth. - Alphabet's stock dropped roughly 10% on June 22, 2026, due to investor frustration with its collapsing free cash flow (FCF fell 47% in Q1 2026 to $10.1B, with full-year 2026 FCF projected at $20.5B, down 72% from $73.3B in 2025). - Alphabet's CFO Anat Ashkenazi stated that 2027 capex would "significantly increase" from 2026's already-elevated range ($180B to $190B), leaving investors concerned about a prolonged FCF trough. - Amazon's trailing FCF has collapsed 95% to $1.2B, even as AWS grew 28% in Q1. - SpaceX fell 16% on June 22 after filing for a $20 billion bond issuance to repay bridge loans for its acquisition of xAI (which occurred in Feb 2026). - There's also the talent drain at Google (Noam Shazeer leaving for OpenAI, John Jumper for Anthropic). - S&P 500 earnings season is here, with Alphabet reporting tomorrow (July 22). Wall Street is demanding "receipts" on this massive AI capex spend. 2. **Nvidia's Supply Chain Bottlenecks and the Kyber NVL144 Delay**: - On July 6, 2026, SemiAnalysis reported that Nvidia's next-generation **Kyber NVL144** rack-scale AI system (designed for the 2027 **Rubin Ultra** generation) has been delayed by over 12 months to **2028**. - The delay is due to manufacturing bottlenecks with the rack's orthogonal backplane, which is an ultra-high-density **78-layer PCB (printed circuit board)** using M9-grade copper-clad laminate, quartz fabric, and PTFE. No fab, including Taiwan's most advanced, can produce it at acceptable yields. - Nvidia's backup plan (NVL72x2, linking two existing Oberon NVL72 racks) was rejected by cloud service providers (CSPs) due to heavy operational and cabling burdens. - This follows another blow on June 30, 2026, when SemiAnalysis reported that the quad-chiplet version of the Rubin Ultra GPU (with 16 HBM4E memory stacks) was canceled, leaving only the dual-chiplet version (with 8 HBM4E stacks). - Despite these 2027/2028 roadmap issues, Nvidia's near-term business remains extremely strong. The current **Vera Rubin NVL72** systems—built on the existing Oberon rack architecture—are in full production and scheduled for delivery this fall (2026) to eight confirmed partners (AWS, Azure, Google Cloud, Oracle Cloud, CoreWeave, Lambda, Nebius, and Nscale). - Nvidia has denied the delays, with CEO Jensen Huang stating on July 15, 2026, that Vera Rubin is "in production and on track." The upcoming Q2 FY2027 earnings call on **August 26, 2026** will be the key test of who is right. Wait, let's look at the existing threads to see which ones we can resolve/update or record. Let's see: - `019eb0cd-20a0-71c6-b91e-92d528f6792e`: "Hyperscaler Capex Guidance Track (July 2026 Earnings)" - Fires when major hyperscalers report Q2 2026 earnings and update their FY2026/FY2027 capex outlook, confirming whether the $710B+ runway is maintained or revised. Wait, since today is July 21, 2026, and Alphabet reports tomorrow (July 22), this thread is still active and hasn't resolved yet (it has a fire date of July 31, 2026). So it's an ongoing watch or a one_shot watch. We can record that the first earnings release is tomorrow. Wait, we don't resolve it yet because the earnings haven't actually come out today (they come out tomorrow, July 22, and the other hyperscalers report through late July). - `019edf29-c3f7-7561-bccb-ef09f3a7fb09`: "Nvidia Q2 FY2027 Revenue Guidance Cross" - fires on August 25/26, 2026. Still in the future. - `019f4b49-2c52-7a54-9f80-a4fb82cea464`: "Nvidia Q2 FY2027 Earnings - Rubin Ultra / Kyber NVL144 Timeline Update" - fires on August 25/26, 2026. Still in the future. - `019f6f55-61e5-71a9-898a-2a6088793817`: "State-Level AI Data Center Moratorium Expansion Track" - ongoing watch. - `019f7482-0c6d-7373-9a41-7e9cb3752cf3`: "Kimi K3 Official Weights Release and Benchmark Validation" - fires on July 27, 2026. - `019f2c65-b0bd-7794-add6-7b6b15ef3fff`: "First Commercial Shipment of Nvidia Rubin GPUs" - fires on Dec 31, 2026. Wait, let's look at the existing notes and their slugs. Let's read the notes first to make sure we reuse the correct slugs. Let's call `read_notes` to see what notes are currently saved.

Okay, I see the existing notes: 1. `micron-ai-memory-pricing-power` 2. `enterprise-ai-roi-gap-revenue-reckoning` (The Hyperscaler Capital Crossover: $725B Capex Binge, the FCF Drain, and New York's Historic Data Center Moratorium) 3. `nvidia-ai-capex-cycle-sustainability` (Nvidia's AI Capex Sustainability: TSMC's Packaging Limits and the Kyber NVL144 78-Layer PCB Failure) 4. `apollo-blackstone-anthropic-35b-tpu-debt-deal` 5. `ornn-gpu-compute-spot-price-surge` 6. `hyperscaler-capex-surge-2026` 7. `custom-silicon-surge-openai-jalapeno-qualcomm-dragonfly` 8. `open-source-frontier-models-china-us-competition` 9. `nebius-group-ai-cloud-infrastructure` 10. `nvidia-q1-2027-record-financials-agentic-ai` 11. `broadcom-q2-2026-earnings-ai-demand` 12. `apple-outsources-siri-to-nvidia-blackwell` 13. `nvidia-groq-20b-acquihire-inference` This is extremely helpful! The brief is: "Is Nvidia's AI capex story still intact?" This directly touches on: - `nvidia-ai-capex-cycle-sustainability`: We can update this note to add the latest on the Kyber NVL144 delay (the detailed 78-layer PCB physics issue, the NVL72x2 backup plan failure, the NVL576/CPO delays, and Jensen Huang's July 15, 2026 denial and reassurance that Vera Rubin is in production and on track, setting up a high-stakes Q2 FY2027 earnings report on August 26, 2026). - `enterprise-ai-roi-gap-revenue-reckoning`: We can update this note to add the latest developments on Big Tech's capex fatigue, Alphabet's 10% drop on June 22, 2026, due to capex fears and talent exits, Amazon's 95% FCF collapse, SpaceX's 16% fall on its $20B bond filing for xAI, and the upcoming Q2 2026 earnings kicking off tomorrow (July 22, 2026) with Alphabet. Let's do a quick deep dive into the current note contents to make sure we integrate our updates seamlessly. Let's first read `nvidia-ai-capex-cycle-sustainability` and `enterprise-ai-roi-gap-revenue-reckoning`. We can't edit files directly since we write notes using `write_note`, but reading them will tell us what's already in there. Let's look at their contents. Wait, can we read them? The `read_notes` tool gives us the list and previews, but does not give us the full bodies. Let's check if there is a way to get the full bodies. Ah, we can search or read the prior findings or we can just update them by writing the full revised body using `write_note` (which overwrites/updates). Wait, let's query the system or check if we can fetch them. Let's see if we can do a search or if we can write them. Wait, let's look at the summary of the previous findings from the kickoff message: - `enterprise-ai-roi-gap-revenue-reckoning` (score 0.59): "The Hyperscaler Capital Crossover: $725B Capex Binge, the FCF Drain, and New York's Historic Data Center Moratorium: The massive artificial intelligence capital expenditure cycle has reached a highly polarized and financially complex phase in mid-2026. While the physical demand driving Nvidia's GPU sales remains..." - `nvidia-q1-2027-record-financials-agentic-ai` (score 0.58): "Nvidia's Record Financial Trajectory: S&P Upgrade to 'AA', $25B Jumbo Bond Sale, and Relentless Blackwell Demand: Nvidia's AI infrastructure dominance and financial strength have reached unprecedented heights in mid-2026, cemented by its blockbuster Q1 FY2027 earnings report (for the period ending April 30..." - `custom-silicon-surge-openai-jalapeno-qualcomm-dragonfly` (score 0.54): "The Custom Silicon Surge: OpenAI's Jalapeño and Qualcomm's Dragonfly C1000 Challenge Nvidia's Moat: While Nvidia continues to dominate the AI hardware market, late June 2026 marked a pivotal escalation in competitive threats. Nvidia’s largest customers and traditional competitors are aggressively..." - `broadcom-q2-2026-earnings-ai-demand` (score 0.52): "Broadcom's Q2 FY2026: Reaffirming a $100 Billion AI Runway Amid a Sharp Market Correction: On June 3, 2026, Broadcom (AVGO) reported its fiscal Q2 2026 earnings, delivering blowout results that highlighted the "insatiable" demand for AI infrastructure. The company's performance provides a..." - `nvidia-groq-20b-acquihire-inference` (score 0.52): "Nvidia's $20 Billion Groq Acquihire: Securing the Agentic Inference Market: In December 2025, Nvidia executed its largest transaction on record—a $20 billion asset acquisition and "acquihire" of the development team at AI inference chip startup Groq, alongside a licensing..." Wait, the note `nvidia-ai-capex-cycle-sustainability` is also there. Let's look at its preview from `read_notes`: "The structural demand driving Nvidia’s artificial intelligence dominance remains robustly supported by physical supply chain signals, but its breakneck annual release cadence is hitting severe... slug: nvidia-ai-capex-cycle-sustainability" Let's update `nvidia-ai-capex-cycle-sustainability` and `enterprise-ai-roi-gap-revenue-reckoning` with the rich, precise details we have gathered. Wait, let's double check if we need to do any more research. Let's see: Is Nvidia's AI capex story still intact? Let's look at the two sides of the story: 1. **The Bull Case (Still Intact)**: - **Near-term demand is insatiable**: The current generation of **Vera Rubin NVL72** systems—built on the existing Oberon rack architecture (which is unaffected by the Kyber delay)—is in full production and scheduled for delivery this fall to eight confirmed partners (AWS, Azure, Google Cloud, Oracle Cloud, CoreWeave, Lambda, Nebius, and Nscale). - **Nvidia's financial performance is record-breaking**: In the quarter ending April 30, 2026, Nvidia reported $81.61 billion in revenue and $58.32 billion in net income, with a gross margin of 74.1%. - **Analyst expectations are high**: SemiAnalysis projects that Nvidia's data-center compute revenue will exceed Wall Street consensus by 20% in the second half of fiscal 2027. Wall Street consensus for fiscal 2027 is $392.7 billion. - **Nvidia's valuation is reasonable**: Forward P/E is 21.7x, close to the S&P 500 average and well below the 5-year average of 72x. - **Management reassurances**: CEO Jensen Huang has dismissed delay rumors, stating that Vera Rubin systems are in production and on track. 2. **The Bear Case / Emerging Risks**: - **Hyperscaler FCF Drain**: The Big Four (Amazon, Microsoft, Google, Meta) + Oracle are projected to spend massive amounts ($452B to $673B in 2026), but free cash flow is collapsing (Alphabet's FCF down 47% in Q1; Amazon's trailing FCF down 95%). Investors are showing fatigue, with Alphabet's stock dropping 10% on June 22, 2026, when CFO Anat Ashkenazi guided to "significantly increased" capex in 2027 without a clear FCF recovery timeline. - **Slowing Capex Growth**: UBS projects that after a 76% surge in 2026, hyperscaler capex growth will slow to 25% in 2027 and just 6% in 2028. This deceleration poses a major threat to chipmakers whose valuations assume prolonged exponential growth. - **Debt Capacity Fatigue**: Torsten Slok of Apollo notes that corporate bond cover ratios for Big Tech issuance have fallen from nearly 5x in February 2026 to below 2x in July 2026, indicating that the capital markets may struggle to absorb further debt-financed capex. - **Supply Chain and Packaging Bottlenecks (The Kyber Delay)**: SemiAnalysis reports that the **Kyber NVL144** rack system (designed for 2027's Rubin Ultra) has been delayed to **2028** due to manufacturing failures of its **78-layer midplane PCB**. Cloud providers rejected the backup "NVL72x2" design due to heavy operational/cabling burdens. - **Product Specification Downgrades**: The quad-chiplet version of the Rubin Ultra GPU (with 16 HBM4E stacks) was reportedly canceled due to TSMC packaging limits, leaving only the dual-chiplet version with 8 HBM4E stacks (halving memory capacity). - **Competitive Openings**: Competitors like AMD (Helios platform with MI455X, offering 432GB of HBM4) and Google (TPU Ironwood) are positioning to capture market share as hyperscalers reconsider their 2027 procurement plans. Let's write a comprehensive update to `nvidia-ai-capex-cycle-sustainability` that tells this exact story, citing the `/markets/NVDA/2026/07/21` URL, the Reuters article, and the Tech Times article. Wait, let's also update `enterprise-ai-roi-gap-revenue-reckoning` to focus on the hyperscaler side of the equation (the FCF drain, Alphabet's Q2 earnings preview, the 10% drop, the SpaceX bond filing, and the UBS capex deceleration model). Let's structure the notes beautifully. First, let's write the update for `nvidia-ai-capex-cycle-sustainability`. Let's check if we have the correct slug: `nvidia-ai-capex-cycle-sustainability`. Yes, that matches. Let's construct the markdown body with no YAML frontmatter, starting with a `# ` heading, quoting the sources, and writing a detailed analysis. Wait, let's double check if we need to quote specific things. From `https://www.techtimes.com/articles/320033/20260709/nvidia-nvidia-kyber-nvl144-slips-2028-physics-not-software-defeats-backup-plan.htm`: - "The next-generation Kyber rack has been pushed more than 12 months past its 2027 target, to sometime in 2028." - "The problem is the board's specifications. According to trade analyses cited by multiple outlets, the orthogonal backplane uses a hybrid of M9-grade copper-clad laminate, quartz fabric, and PTFE — specialty high-frequency materials that only a handful of fabricators worldwide can work with — in a 78-layer stack approximately one square meter in area." - "Nvidia's response: 'Our roadmap is intact.'" - "CEO Jensen Huang dismissed reports of delays in the rollout of the company's Vera Rubin AI hardware and affirmed that production remains on track." From `https://www.reuters.com/business/retail-consumer/among-ai-crowd-some-investors-position-slower-hyperscaler-spending-growth-2026-07-17/`: - "UBS estimating hyperscalers' capex will rise 76% this year to $673 billion, but will increase by only 25% next year and just 6% in 2028." - "The Philadelphia Semiconductor Index (.SOX) ... has more than doubled over the past year, even with a near-18% drop from its June peak" This is perfect! Let's write the note.

I will now update the existing finding on Nvidia's AI capex sustainability (`nvidia-ai-capex-cycle-sustainability`) to integrate the latest details on the Kyber NVL144 78-layer PCB manufacturing delay, the Rubin Ultra GPU memory down-specification, and Nvidia's official rebuttals.

I will now update the existing finding on the hyperscaler capital crossover and free cash flow drain (`enterprise-ai-roi-gap-revenue-reckoning`) to capture the recent market volatility, Alphabet's 10% drop on June 22, 2026, due to capex fears and talent exits, Amazon's FCF collapse, and the upcoming Q2 2026 earnings season where Wall Street is demanding receipts on the massive AI capex buildup.

I will now call `finish_cycle` to complete this research cycle, as I have updated the two critical notes tracking the hyperscaler capex story and Nvidia's roadmap challenges with concrete, newly surfaced evidence from this cycle.

I have successfully completed the research cycle for topic **019e95a0-30d1-7d08-ae5b-10ac242b9d5c** on whether Nvidia's AI capex story is still intact. ### Summary of Findings I have updated and refined two key findings on this topic with concrete, recent evidence from mid-July 2026: 1. **Nvidia's AI Capex Sustainability: TSMC's Packaging Limits and the Kyber NVL144 78-Layer PCB Failure** (`[[nvidia-ai-capex-cycle-sustainability]]`) - **Kyber NVL144 Delay:** SemiAnalysis reported on July 6, 2026, that Nvidia's next-generation Kyber NVL144 rack-scale AI system (originally planned for 2027) has been delayed by more than 12 months to **2028**. - **The 78-Layer PCB Bottleneck:** The delay is due to manufacturing yield failures of the rack's central orthogonal backplane PCB—a massive, ultra-high-density 78-layer board using specialty high-frequency materials (M9-grade copper-clad laminate, quartz fabric, PTFE) and 25-micrometer trace widths. No advanced PCB fabricator has been able to produce it at scale. - **Rubin Ultra Downgrades:** The backup "NVL72x2" design was rejected by cloud providers due to heavy operational/cabling burdens, and the quad-chiplet Rubin Ultra GPU was canceled due to TSMC packaging limits, leaving a dual-chiplet version with half the memory capacity. - **Near-term Strength & Rebuttal:** Nvidia strongly denies the delay, with CEO Jensen Huang stating on July 15, 2026, that Vera Rubin is "on track" and in full production. Near-term revenue is unaffected because current-generation **Vera Rubin NVL72** systems (built on the unaffected Oberon rack architecture) are shipping this fall to eight major cloud partners. The high-stakes test of who is right will arrive on Nvidia's Q2 FY2027 earnings call on **August 26, 2026**. 2. **The Hyperscaler Capital Crossover: $725B Capex Binge, the FCF Drain, and New York's Historic Data Center Moratorium** (`[[enterprise-ai-roi-gap-revenue-reckoning]]`) - **Slowing Capex Growth:** A July 2026 UBS model estimates that after a 76% surge to $673 billion in 2026, hyperscaler capex growth will slow sharply to 25% in 2027 ($843B) and only 6% in 2028 ($894B). - **The FCF Pinch & Backlash:** Hyperscalers are experiencing a massive free cash flow drain. Google's Q1 FCF fell 47% to $10.1B, leading to a 10% stock plunge on June 22, 2026, after CFO Anat Ashkenazi guided to "significantly increased" capex in 2027 without a clear FCF recovery timeline. Amazon's trailing FCF collapsed 95% to $1.2B. - **Debt Market Fatigue:** Torsten Slok of Apollo notes that corporate bond cover ratios for Big Tech debt issuance have fallen from nearly 5x in February 2026 to below 2x in July 2026, indicating that capital markets are showing signs of fatigue in absorbing debt-financed capex. - **Earnings Season Test:** The Q2 2026 earnings season kicks off tomorrow, **Wednesday, July 22, 2026**, with Alphabet, where Wall Street will demand concrete receipts on this massive AI infrastructure buildout.