Cycle on Jul 11, 2026

I will begin by listing the open threads and watches for this topic to see what specific questions or triggers have been set up by previous cycles.

I will call `market_view` for Nvidia (NVDA) to get the latest public market figures and updates on its financials and stock performance, which will ground our quantitative analysis.

I will perform a web search to check for recent news regarding Nvidia's AI capex story, specifically focused on July 2026 updates, hyperscaler capex plans, and any developments around the Kyber rack delays, TSMC's upcoming Q2 2026 earnings (which is in a few days on July 16, 2026), and overall market sentiment.

I will perform a search on the CNBC report from July 5, 2026, about the Kyber rack system delay and Nvidia's denial or reaction, to get the precise details and quotes regarding this development.

I will fetch the Seeking Alpha article or the CNBC article to read the exact details of the Kyber NVL144 rack delay, the PCB snag, and any response from Nvidia. Let's fetch `seekingalpha.com/news/4611538-nvidia-next-gen-kyber-ai-rack-delayed-to-2028-on-manufacturing-snags-report` first as it's a reliable financial reporting source, or the CNBC article if available. Let's search for the exact Seeking Alpha article or CNBC article URL first to see if we can fetch it. I'll search for seekingalpha.com URL.

I will fetch the CNBC article or another detailed source on the Kyber delay to get more technical specifics, particularly the PCB midplane issue, and what it implies for Rubin Ultra and Nvidia's revenue projections. Let's do a search to find a detailed report that we can fetch. I'll search for `Nvidia Kyber rack SemiAnalysis PCB midplane 2028`.

I will fetch the Datacenter Dynamics or Digitimes article to understand the details of the PCB midplane snags and Nvidia's denial. Let's fetch `https://www.datacenterdynamics.com/en/news/nvidia-pushes-kyber-release-to-2028-following-manufacturing-concerns-report/` first.

I will also fetch the Digitimes article `https://www.digitimes.com/news/a20260708VL212/nvidia-pcb-rubin-materials-ccl.html` to find out more about the PCB midplane snags, materials, and how the supply chain is reacting.

The Digitimes article is short but provides confirmation of Nvidia's denial and the context of the PCB midplane. Let's search for "Nvidia Kyber rack SemiAnalysis" to see what further details there are about the PCB midplane. Specifically: 1. What are the technical details of the PCB midplane? (e.g., number of layers, thickness, manufacturability challenges) 2. What are the implications for Rubin Ultra and Rubin chips? Are they delayed? 3. How does this affect Nvidia's short-term and medium-term revenue? (The DCD article mentions SemiAnalysis projected Nvidia's data-center compute revenue will exceed Wall Street consensus by 20% in the second half of fiscal 2027 even with product delays). Let's do a search to find more in-depth reports or discussions.

I will fetch the Techzine article `https://www.techzine.eu/news/devices/142682/ambitious-nvidia-roadmap-hits-snag-as-kyber-pushed-to-2028/` to read the details on the 78-layer PCB midplane and the exact defect risks.

Let's search for "Vik's Newsletter" `Kyber Midplane Design, Making NVL72x2 Work, and All Roads Led to Kyber` to get a deeper understanding of the technical challenges, and see if there are any other details about the PCB midplane.

I will fetch `https://www.viksnewsletter.com/p/kyber-midplane-design-making-nvl72x2` to read the engineering details of the PCB midplane. This will give us high-quality technical evidence.

This is an excellent engineering deep dive. Let's summarize the key facts we have: 1. **Kyber NVL144 Rack Delay**: SemiAnalysis reported on July 6, 2026, that Nvidia's next-gen Kyber NVL144 rack-scale architecture (designed to house 144 of the 2027 Rubin Ultra GPUs) has been delayed by over 12 months to 2028. 2. **Technical Snag**: The delay is due to manufacturing challenges involving a complex 78-layer PCB midplane. The midplane connects the vertical compute trays ("like books on a shelf") to the network switches on the other side. This 78-layer PCB is extremely complex, requiring M9-grade copper clad laminate (CCL) and Quartz glass (Q-glass) reinforcement to minimize the dissipation factor to ~0.0007. Lamination defects, via drilling through abrasive Q-glass, and yield compound risks across the 1-square-meter board mean defects scrap the entire board. 3. **Backup Plan Abandoned**: Nvidia's proposed stopgap backup plan—a back-to-back NVL72x2 configuration using two existing Oberon-style racks connected back-to-back—was rejected by cloud service providers and hyperscalers due to its "odd design and heavy operational burden" on data hall layout (which are laid out in straight rows with specific cable raceways, power, and cooling orientations). This back-to-back design has reportedly been cancelled. 4. **Nvidia's Response**: Nvidia has officially denied the delay, stating "Our roadmap is intact." 5. **No Impact on Short-term Revenue**: Despite these roadmap debates, SemiAnalysis notes that Nvidia's data-center compute revenue is still projected to exceed Wall Street consensus by 20% in the second half of FY2027, as Nvidia will simply sell more of the Blackwell-era "Oberon" racks to cover any Kyber shortfall. 6. **Rubin Production**: Current Rubin systems (the base Rubin, not Rubin Ultra) are in full production and will be shipped this fall (2026) to eight cloud partners, including the big three hyperscalers (AWS, Microsoft Azure, Google Cloud). Wait! Let's check if there are any other recent news items about Nvidia's AI capex story remaining intact. What about the hyperscalers' actual capex spending? The thread list shows: - `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, today's date is July 11, 2026. The Q2 2026 earnings for the major hyperscalers (Alphabet, Microsoft, Meta, Amazon) usually start in late July (around July 23-30). So this watch has not fired yet, but we should check if there are any pre-earnings previews or recent capex comments. Let's search for "hyperscaler capex predictions Q2 2026" or "AI capex projections July 2026". Let's do 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 see the latest figures on hyperscaler spending as of July 2026. This will give us excellent context.

This is very useful data. Goldman Sachs now expects a combined **$5.3 trillion of capex spending** for the four largest hyperscalers (Meta, Microsoft, Amazon, Alphabet) from FY2025 to FY2030, which is up from a prior estimate of $4.5 trillion. For calendar year 2026, the four hyperscalers collectively plan to allocate **$725 billion** to capital expenditures, up 77% from last year's $410 billion. The breakdown: - Amazon: projecting $200 billion - Alphabet: targeting $175 billion to $185 billion - Meta: guiding $115 billion to $135 billion - Microsoft: tracking toward $190 billion Wait, let's look at the TSMC Q2 2026 earnings conference watch (`019f1cf3-530f-7f29-a5c5-dbe8ad17023d`). It is scheduled for July 16, 2026, which is in 5 days (as of today, July 11, 2026). Wait, what about TSMC's recent June revenue? TSMC reports monthly revenue around the 10th of every month. Since today is July 11, 2026, TSMC's June 2026 revenue should have just been reported! Let's search for "TSMC June 2026 revenue" or "TSMC monthly revenue June 2026" to see if there's any fresh data that would trigger the TSMC monthly revenue YoY growth watch (`019ecfb3-79b3-76db-aeec-0bc8d0635c41`), which triggers if the YoY growth rate drops below 20%. Let's search for this.

Ah, TSMC's Monthly Sales for June 2026 is scheduled to be released on **July 13, 2026 (Mon)**, because it was postponed due to a typhoon day-off on July 10th, 2026! This is a very interesting piece of real-time schedule information. So the monthly sales will release on July 13, and the Q2 2026 earnings conference is on July 16. Both of these are coming up in the next few days. Let's search for "TSMC Q2 2026 earnings preview" or "TSMC July 16 2026 earnings" to see if there are any updates or expectations. Let's do a search.

Let's fetch the Seeking Alpha article `https://seekingalpha.com/news/4612235-tsmc-expected-to-increase-guidance-capex-at-q2-analyst-meeting-gf-securities` to get the expectations for TSMC's July 16 earnings. This is a very critical forward-looking indicator for the AI capex story.

The article mentions that GF Securities expects TSMC to raise its full-year guidance along with higher capital expenditures for 2027 at the July 16 analyst meeting. Wait, let's look for more details in other articles to see if there are quotes or numbers. Let's do a search for "TSMC expected to increase guidance, capex at Q2 analyst meeting: GF Securities". I'll search on TipRanks or other platforms. Wait, the tradingview article said: "TSMC also raised its 2026 capital expenditure outlook to the high end of its $52-$56 billion range to expand advanced-node capacity". Let's search for "TSMC capital expenditure outlook" or "TSMC capex 2026 52 billion" to verify if TSMC recently raised its capex or if that's expected.

Let's fetch `https://www.techtimes.com/articles/320142/20260711/tsmc-q2-earnings-july-16-three-cowos-signals-that-test-ais-spending-ceiling.htm` which was published today (July 11, 2026)! This is incredibly fresh and highly relevant to the TSMC Q2 earnings watch. Let's fetch it.

This is a fantastic article! It's from today (July 11, 2026) and contains extremely rich information. Let's synthesize what we have and see if we can resolve any watches or threads, and write/update findings. First, let's look at the threads and watches again: 1. `019f1cf3-530f-7f29-a5c5-dbe8ad17023d`: "TSMC Q2 2026 Earnings Conference" - scheduled for July 16, 2026. This is a dated catalyst. It is in the future, so we shouldn't resolve it yet, but we have excellent context to write about it. 2. `019ecfb3-79b3-76db-aeec-0bc8d0635c41`: "TSMC Monthly Revenue YoY Growth Rate Track" - Triggers if TSMC's monthly revenue YoY growth drops below 20%. Let's look at the TSMC monthly revenue. The article says: "A typhoon that shut down Taiwan's financial markets on Friday [July 10, 2026] forced TSMC to postpone its June 2026 monthly sales data release... That data now arrives July 13 [2026]". This is a great piece of information. Since it hasn't been released yet (it was postponed from July 10 to July 13), this watch has not fired yet, but we will note that it's coming on July 13. 3. `019eb0cd-20a0-71c6-b91e-92d528f6792e`: "Hyperscaler Capex Guidance Track (July 2026 Earnings)" - We have fresh data on this! The Yahoo Finance article from June 3, 2026, says that Goldman Sachs strategist Amanda Lynam raised combined capex spending for the four largest hyperscalers (Meta, Microsoft, Amazon, Alphabet) from FY2025 to FY2030 to **$5.3 trillion** (up from $4.5 trillion). For calendar year 2026, they collectively plan to allocate **$725 billion** to capital expenditures (up 77% from last year's $410 billion). Wait, has the actual July 2026 earnings happened yet? No, today is July 11, 2026, and the earnings calls start in late July. So this watch is still ongoing, but we can update our findings with the raised Goldman Sachs estimates. 4. `019f4b49-2c52-7a54-9f80-a4fb82cea464`: "Nvidia Q2 FY2027 Earnings - Rubin Ultra / Kyber NVL144 Timeline Update" - This is for Nvidia's earnings on August 25, 2026. This is in the future. 5. Let's write a major finding about the Kyber rack-scale delay and Nvidia's response, and how it impacts the AI capex story. Wait! Is there an existing finding for this? Yes: `Nvidia's AI Capex Story Remains Robustly Intact Amid Soaring Projections and Kyber Rack Delay Debates` (slug: `nvidia-ai-capex-cycle-sustainability`). Let's see the full content of `nvidia-ai-capex-cycle-sustainability` by listing or reading the note. Wait, we should call `read_notes` to see what is already in that note so we can revise/update it properly. Let's do that!

I will read the note `nvidia-ai-capex-cycle-sustainability` to understand its current content and how we should update it with the new details from today (the 78-layer PCB midplane, M9-grade CCL, Q-glass, the back-to-back NVL72x2 cancellation, the upcoming TSMC Q2 earnings and June monthly sales release, and the updated Goldman Sachs hyperscaler capex projections). Wait, does the tool `read_notes` let me read a specific note's body? No, `read_notes` lists the notes. Ah, wait, can I get the full body from `read_notes`? The tool parameters for `read_notes` are `topic_id` and `limit`. It returns the slug, title, preview, updated_at, and source_count. Let me check if there's any file in the workspace or if I can read it. Let's list files in the workspace directory. Wait, let's look at the Guidelines: "Before modifying a file, read it first. Do not assume files or directories exist." Since this tool environment has a workspace directory, let's see what is in `/var/lib/hey-lefty/notes/019e89b4-58b3-7e3b-a0c2-6a68bd3242db/019e95a0-30d1-7d08-ae5b-10ac242b9d5c`. Wait, let's check if the notes are stored as local files or if there is a tool to read the note body. Wait, let's search if there is a way to get the note body or if we can find it in the workspace. Or is the note body actually in the notes directory? Let's check with standard terminal commands or check if we can run shell commands? No, we don't have a bash tool declared! The tools available are: - `default_api:list_threads` - `default_api:market_view` - `default_api:read_notes` - `default_api:record_event` - `default_api:record_thread` - `default_api:record_watch` - `default_api:resolve_thread` - `default_api:web_fetch` - `default_api:web_search` - `default_api:write_note` - `default_api:finish_cycle` Ah! There is no tool to read the full body of a note, but `read_notes` gives us the preview. Let's look at the preview for `nvidia-ai-capex-cycle-sustainability`: `The structural demand driving Nvidia’s artificial intelligence dominance remains robustly intact in July 2026, even as the company navigates a fresh wave of product roadmap debates. On July 5,…` Wait, let's search the web for what happened on July 5, 2026, or if there's any other way to get the exact text. Ah, the preview of `nvidia-ai-capex-cycle-sustainability` says: "The structural demand driving Nvidia’s artificial intelligence dominance remains robustly intact in July 2026, even as the company navigates a fresh wave of product roadmap debates. On July 5,..." This shows that the previous cycle (which was on July 10, 2026) already wrote a finding under the slug `nvidia-ai-capex-cycle-sustainability` covering the July 5th/6th SemiAnalysis report about the Kyber delay! Let's see: we have gained much more detailed engineering information today (July 11, 2026) from Vik's Newsletter (published recently), explaining the exact mechanism of the 78-layer PCB midplane, the M9-grade CCL, Q-glass, the cancellation of the NVL72x2 backup plan due to hyperscaler pushback over data hall layout, and the upcoming TSMC Q2 earnings on July 16, 2026 (including the June monthly sales postponement to July 13 due to a typhoon). We also have the updated Goldman Sachs hyperscaler capex forecast of $5.3 trillion through 2030 (up from $4.5 trillion) and $725 billion for 2026 (up 77% YoY). Let's write a highly detailed update to `nvidia-ai-capex-cycle-sustainability` that incorporates these fresh engineering and market insights. Wait, let's first check if there are other notes we should update or if we should write a new note. Wait, the guidelines say: "Slugs are stable identities, not per-cycle labels — this is what makes findings a living document. 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; if you find yourself coining a near-synonym of an existing slug, stop and reuse the existing one." So we should definitely reuse `nvidia-ai-capex-cycle-sustainability` to update the Nvidia AI capex story! Let's also check if there is another note we can update. What about `hyperscaler-capex-surge-2026` or `enterprise-ai-roi-gap-revenue-reckoning`? Let's look at `hyperscaler-capex-surge-2026` and see its preview: `As the artificial intelligence capex boom faces severe physical constraints from power availability and grid capacity, a major private-capital solution emerged in June 2026. KKR, in partnership with…` Wait, does `hyperscaler-capex-surge-2026` track the hyperscaler spending? Yes, it does. Let's see if we should update `nvidia-ai-capex-cycle-sustainability` first. Let's check if there are any other specific pieces of information we should fetch. What about the TSMC monthly revenue growth watch? The watch `019ecfb3-79b3-76db-aeec-0bc8d0635c41` is "TSMC Monthly Revenue YoY Growth Rate Track". It fires if the monthly revenue YoY growth drops below 20%. Since the June revenue is postponed to July 13, 2026, we cannot resolve it yet. But wait! What about TSMC's May 2026 revenue? The May 2026 revenue report says: "Revenue for January through May 2026 totaled NT$1,961.80 billion, an increase of 30.0 percent compared to the same period in 2025." And the TSMC May monthly revenue YoY growth was 30% YoY. So it remained well above 20%. Wait, let's check if there are any other watches we can resolve or record events for. Let's review the active watches in `list_threads`: - `019f1cf3-530f-7f29-a5c5-dbe8ad17023d`: TSMC Q2 2026 Earnings Conference (July 16, 2026) - Future, cannot resolve. - `019f2c65-b0bd-7794-add6-7b6b15ef3fff`: First Commercial Shipment of Nvidia Rubin GPUs (Dec 31, 2026) - Future, cannot resolve. - `019f4b49-2c52-7a54-9f80-a4fb82cea464`: Nvidia Q2 FY2027 Earnings - Rubin Ultra / Kyber NVL144 Timeline Update (Aug 25, 2026) - Future, cannot resolve. - `019efe0e-ea69-7ac4-9af7-50d0d58f3a57`: Micron Quarterly Capex FY2027 Track - Ongoing. - `019ef3bf-47e0-7795-ba54-6be76a9a84e7`: First Commercial Deployment of Nvidia-LG Co-Developed Smart Factory Standard - Future. - `019ee44e-b944-74a7-bc73-b87fb1bf1ac5`: Aggregate Hyperscaler Free Cash Flow Crossover (Q3 2026) (Oct 31, 2026) - Future. - `019edf29-c3f7-7561-bccb-ef09f3a7fb09`: Nvidia Q2 FY2027 Revenue Guidance Cross - Future (August 2026). - `019ed4d9-3d3f-7390-b1e8-77f8753e9989`: Kazakhstan Data Center Valley Phase 1 Groundbreaking - Future. - `019ed4d9-3256-78b5-94ac-9e710b346c1d`: Helix Digital Infrastructure Project Commitments Track - Ongoing. - `019ecfb3-79b3-76db-aeec-0bc8d0635c41`: TSMC Monthly Revenue YoY Growth Rate Track - Ongoing. - `019ec568-ed82-7d7b-a8d0-36e0e5717e8a`: Ornn Compute Price Index Blackwell GPU Rental Rate Track - Ongoing. - `019ec040-d231-7dc2-9387-902ca1d82c1d`: Nvidia Gross Margin Track - Ongoing. - `019ebb1c-c3cb-7db0-b267-77ac380cc943`: Regulatory Scrutiny of Off-Balance-Sheet AI SPVs - Ongoing. - `019eb0cd-20a0-71c6-b91e-92d528f6792e`: Hyperscaler Capex Guidance Track (July 2026 Earnings) (July 31, 2026) - Future. - `019eaba5-b004-719f-8669-34a8aae85368`: Nvidia Standalone Vera CPU Revenue Track - Ongoing. - `019e9c33-1f68-7b87-b126-0aa020f67c7f`: Apple Revamped Siri Launch with Blackwell Workloads (September 2026) - Future. So none of the watches have fired yet since they are all forward-looking, but we are tracking them closely. Let's write a comprehensive update to `nvidia-ai-capex-cycle-sustainability`. Wait, let's verify if there is any other company or sector we should pull `market_view` for. What about TSMC? TSM is a public company. Let's pull `market_view` for TSM to get its latest figures so we can ground our TSMC-related discussion in hard numbers.

This is highly valuable data. The TSM market view shows: - Market Cap: $2.27T (NYSE: TSM) - Gross margin: 61.9% - Latest quarter revenue (for period ending March 31, 2026): NT$1.13 trillion (wait, note that the Alpha Vantage numbers are in New Taiwan Dollars for the quarterly trajectory for TSM as it is based in Taiwan, but let's make sure we cite it properly as per `/markets/TSM/2026/07/11`). Wait, the TSM market view shows: - Q1 2026 (2026-03-31): rev NT$1.13T (T$1.13T) | op NT$657.29B | net NT$572.48B. Let's make sure we quote this accurately. Let's write our updated finding for `nvidia-ai-capex-cycle-sustainability`. Wait, let's also look at `hyperscaler-capex-surge-2026` to see if we should write a separate note or update it. Let's see if we can do an update on `hyperscaler-capex-surge-2026` as well, or if we should focus on updating `nvidia-ai-capex-cycle-sustainability` as our primary note. Let's do both if they represent separate coherent themes! - `nvidia-ai-capex-cycle-sustainability`: This tracks Nvidia's product roadmap, the Kyber delay debate, the physical and engineering challenges of the 78-layer PCB, and why the capex story remains intact because Nvidia will sell more Oberon racks to cover any Kyber shortfall. - `hyperscaler-capex-surge-2026`: This tracks the massive capital expenditures of the hyperscalers themselves, including the updated Goldman Sachs estimates ($725B in 2026, $5.3T through 2030), and the upcoming earnings prints. Wait, let's see if we have enough fresh information to update `hyperscaler-capex-surge-2026`. Yes, the Goldman Sachs $5.3T projection and the $725B 2026 projection are spectacular fresh numbers that we can use to update `hyperscaler-capex-surge-2026`. Wait! Let's read the guidelines for `write_note`: - "Every substantive finding MUST cite at least one real source you actually consulted this cycle... carry 1-2 REPRESENTATIVE VERBATIM QUOTES, copied exactly, each kept with the link it came from... then say what it means." - "Slugs are stable identities, not per-cycle labels — this is what makes findings a living document." - "No YAML frontmatter — the title is the first `# ` heading and the slug is the slug parameter you pass." - "To link ANOTHER finding from a note body, write its slug as a wikilink — `[[its-slug]]`." Let's carefully construct the markdown body for `nvidia-ai-capex-cycle-sustainability`. Wait, let's check what the previous content of `nvidia-ai-capex-cycle-sustainability` was. The preview is: `The structural demand driving Nvidia’s artificial intelligence dominance remains robustly intact in July 2026, even as the company navigates a fresh wave of product roadmap debates. On July 5,…` Since we are doing an *upsert* (update), we should write a comprehensive, beautifully structured note that covers the entire current state of this topic, blending the previous context (which we know was about the July 5/6 SemiAnalysis report on Kyber delay and Nvidia's denial) with our rich new engineering details (78 layers, M9-grade CCL, Q-glass, NVL72x2 back-to-back cancellation, and TSMC's role as the packaging bottleneck). Let's draft the body for `nvidia-ai-capex-cycle-sustainability`. Wait, let's list the sources we will cite: 1. `https://www.datacenterdynamics.com/en/news/nvidia-pushes-kyber-release-to-2028-following-manufacturing-concerns-report/` (DCD, July 6, 2026) 2. `https://www.viksnewsletter.com/p/kyber-midplane-design-making-nvl72x2` (Vik's Newsletter, July 6, 2026) 3. `https://www.techtimes.com/articles/320142/20260711/tsmc-q2-earnings-july-16-three-cowos-signals-that-test-ais-spending-ceiling.htm` (Tech Times, July 11, 2026) 4. `/markets/NVDA/2026/07/11` (Nvidia public markets view) 5. `/markets/TSM/2026/07/11` (TSMC public markets view) Let's select 1-2 representative verbatim quotes for each source and keep them with their links. For DCD: > "Nvidia’s alternative plan, which involved placing two Oberon racks back-to-back in order to achieve a similar power output, has also been scrapped after cloud companies and hyperscalers pushed back against its “odd design and heavy operational burden.”" > "In response to a request for comment from DCD regarding the Kyber claims, a spokesperson for Nvidia said: 'Our roadmap is intact.'" For Vik's Newsletter: > "The midplane PCB makes the equivalent of 20,000 connections that would otherwise need to be copper cables which would be too many to fit in the spine. These boards are built on separate 26 layer boards each nearly a square meter in size and then laminated together to form a massive PCB required for Kyber midplane." > "To keep the midplane PCB losses in check, it is built on an advanced high-frequency material consisting of an M9-grade copper clad laminate (CCL) with Quartz glass (Q-glass) reinforcement." > "The Oberon rack will remain the mainstay, and Nvidia will sell more of them in the Rubin era to cover the Kyber shortfall. SemiAnalysis still sees data center compute revenue running 20 percent above consensus in the second half of FY27, so views are not uniformly bearish." For Tech Times: > "A typhoon that shut down Taiwan's financial markets on Friday forced TSMC to postpone its June 2026 monthly sales data release — a routine disclosure that investors had been counting on as their last clean read of the quarter before the company reports its second-quarter 2026 earnings in six days. That data now arrives July 13, giving investors roughly 72 hours to absorb it before Thursday's print." > "At the June 4 annual shareholders meeting, CEO C.C. Wei told shareholders that CoWoS capacity remains 'extremely tight and sold out through 2026.' Nvidia's management has described CoWoS assembly as 'oversubscribed through at least mid-2026,' and the Silicon Analysts CoWoS allocation tracker found lead times of 52 to 78 weeks at TSMC's advanced-packaging backend facilities, with bookings extending into 2027." Let's write this beautiful, comprehensive, and highly detailed note now! I will call `write_note` with slug `nvidia-ai-capex-cycle-sustainability`.

This is a highly comprehensive update to `nvidia-ai-capex-cycle-sustainability`. Now, let's look at `hyperscaler-capex-surge-2026`. Let's see if we can update this note to reflect the massive, upgraded capex projections reported by Goldman Sachs and others, and tie it to the upcoming Q2 2026 earnings season. Wait, let's look at the existing note slug: `hyperscaler-capex-surge-2026`. Let's see: `read_notes` lists a note with slug `hyperscaler-capex-surge-2026` but wait, let's check its title: `KKR, Nvidia, and Vistra Launch $10 Billion Helix Venture to Bypass Grid Bottlenecks` (slug: `hyperscaler-capex-surge-2026`). Ah! The note with slug `hyperscaler-capex-surge-2026` actually tracks the KKR, Nvidia, and Vistra Helix venture. Let's look at the other notes: - `enterprise-ai-roi-gap-revenue-reckoning`: "The Hyperscaler Capital Crossover: AI Capex Outstrips Cash Flow, Triggering Historic Debt and Equity Waves" Let's see if we should update `enterprise-ai-roi-gap-revenue-reckoning` or `hyperscaler-capex-surge-2026` with the updated Goldman Sachs numbers. Wait, the Goldman Sachs numbers are about the incredible size of the hyperscaler capex ($725 billion in 2026, $5.3 trillion through 2030 across Meta, Microsoft, Amazon, Alphabet). This fits perfectly under `enterprise-ai-roi-gap-revenue-reckoning` which tracks the massive capex cycle outstripping cash generation and triggering major funding waves! Let's review the preview of `enterprise-ai-roi-gap-revenue-reckoning`: `The massive capital expenditure cycle driving the artificial intelligence boom has entered a critical phase where infrastructure investments are outstripping corporate cash generation. Rather than…` Yes! This is the perfect place to add the updated Goldman Sachs figures and the upcoming Q2 earnings context. Let's write an updated note for `enterprise-ai-roi-gap-revenue-reckoning`! Wait, let's list the sources we will cite for this: 1. `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` (Yahoo Finance, June 3, 2026) 2. `https://www.techtimes.com/articles/320142/20260711/tsmc-q2-earnings-july-16-three-cowos-signals-that-test-ais-spending-ceiling.htm` (Tech Times, July 11, 2026) Let's select 1-2 representative verbatim quotes: For Yahoo Finance: > "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. Prior to the start of first quarter earnings, this figure stood at $4.5 trillion." > "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." > "Amazon is projecting $200 billion in capital expenditures, Alphabet is targeting $175 billion to $185 billion, Meta is guiding $115 billion to $135 billion, and Microsoft is tracking toward $190 billion for the calendar year." Let's write this updated note now! I will call `write_note` with slug `enterprise-ai-roi-gap-revenue-reckoning`.

This is excellent. We have updated both key notes: 1. `nvidia-ai-capex-cycle-sustainability` (focusing on Nvidia's product roadmap, the Kyber 78-layer PCB midplane technical bottleneck, the NVL72x2 backup plan cancellation, and why the capex story remains intact because Nvidia will sell more Oberon racks to cover any Kyber shortfall). 2. `enterprise-ai-roi-gap-revenue-reckoning` (focusing on the hyperscalers' capex projections, Goldman Sachs' upgraded $5.3T multi-year and $725B 2026 budgets, the individual company figures, and the upcoming Q2 earnings litmus test). Wait, let's look at the threads and watches again. Is there any other watch we should record, or any thread we should resolve? Let's see: we didn't resolve any existing watch because none of them has fired yet (TSMC earnings is on July 16, TSMC June monthly sales is on July 13, Hyperscaler earnings are in late July, Nvidia earnings are on August 25). But wait! Let's check if there are any threads we can resolve. Wait, let's look at the list of threads again. The list of threads returned: - All of them are of kind `watch`! Let's check if there are any other threads of kind `question`, `investigation`, or `contradiction`. Wait, in the output of `list_threads`, there were no threads of those kinds, only `watch` threads. Let's verify: indeed, the threads list contains only watches. Wait, let's write a quick update to our long-term memory file `/var/lib/hey-lefty/notes/019e89b4-58b3-7e3b-a0c2-6a68bd3242db/019e95a0-30d1-7d08-ae5b-10ac242b9d5c/memory/MEMORY.md` if it exists. Let's check if we can write to memory or if we should just call `finish_cycle`. Wait, the Guidelines say: "Before modifying a file, read it first. Do not assume files or directories exist." And: "Your findings are a LIVING SET, not a fresh page... Call finish_cycle to end the cycle." Let's check if there are any files we need to read or write. Oh, wait! We don't have a bash tool or file system tools in our declaration list! The only tools we have are the ones listed above. Ah, so we cannot read or write `/var/lib/hey-lefty/notes/019e89b4-58b3-7e3b-a0c2-6a68bd3242db/019e95a0-30d1-7d08-ae5b-10ac242b9d5c/memory/MEMORY.md` directly because we don't have terminal tools or file system tools in our tool declaration! We only have `list_threads`, `market_view`, `read_notes`, `record_event`, `record_thread`, `record_watch`, `resolve_thread`, `web_fetch`, `web_search`, `write_note`, and `finish_cycle`. That's perfectly fine! We can do all our work through the dedicated research API tools. Let's do a quick final check of our notes and make sure everything is in order. We have successfully updated: - `nvidia-ai-capex-cycle-sustainability` - `enterprise-ai-roi-gap-revenue-reckoning` These updates are highly grounded, cite real sources we fetched this cycle, carry verbatim quotes kept with their links, and use stable slugs. We are now ready to call `finish_cycle`. Let's do that!

I have completed the research cycle for topic `019e95a0-30d1-7d08-ae5b-10ac242b9d5c` ("Is Nvidia's AI capex story still intact?"). ### Executive Summary of Findings During this cycle, I investigated the recent debates surrounding Nvidia's next-generation hardware roadmap and the broader hyperscaler capital expenditure trajectory. Here is the summary of what was uncovered and updated in the living findings: 1. **The Kyber NVL144 Rack Delay and the 78-Layer PCB Bottleneck** (`[[nvidia-ai-capex-cycle-sustainability]]`): * **The Snag:** On July 5–6, 2026, reports surfaced that Nvidia’s upcoming liquid-cooled **Kyber NVL144 rack-scale architecture** (designed to house the 2027 Rubin Ultra GPUs) has been delayed by over 12 months to 2028. * **The Engineering Physics:** The delay is rooted in severe manufacturing challenges with the massive **78-layer PCB midplane**. To prevent signal losses over distance, the board is constructed from advanced high-frequency materials (M9-grade copper clad laminate with Quartz glass reinforcement) to keep the dissipation factor at a microscopic ~0.0007. However, Quartz glass is highly abrasive, making via drilling extremely difficult, and lamination defects across 78 layers mean any flaw scraps the entire square-meter board. * **Scrapped Back-up Plan:** Nvidia's stopgap design—the **NVL72x2 back-to-back rack configuration**—was rejected and cancelled after hyperscalers pushed back against its awkward footprint and heavy operational burden on standard data hall layouts. * **Story Remains Intact:** Nvidia has denied the delay, stating its "roadmap is intact." More importantly, even if Kyber is delayed, hyperscalers will simply buy more Blackwell-era **"Oberon" NVL72 racks** to cover the shortfall. SemiAnalysis still projects Nvidia's data-center compute revenue to run 20% above Wall Street consensus in H2 FY2027. 2. **Parabolic Hyperscaler Capex Escalation** (`[[enterprise-ai-roi-gap-revenue-reckoning]]`): * **Upgraded Projections:** Goldman Sachs has significantly raised its multi-year capex forecast for the four largest hyperscalers (Meta, Microsoft, Amazon, Alphabet) to **$5.3 trillion** from FY2025 to FY2030 (up from $4.5 trillion). * **2026 Budgets:** For calendar year 2026, these four giants plan to allocate **$725 billion** to capital expenditures (up 77% YoY). Individually, Amazon is projecting **$200 billion**, Microsoft is tracking toward **$190 billion**, Alphabet is targeting **$175B–$185 billion**, and Meta is guiding **$115B–$135 billion**. This massive runway reinforces the sustainability of the near-term AI capex story. 3. **TSMC as the Ultimate Bottleneck and Upcoming Q2 Earnings** (`[[nvidia-ai-capex-cycle-sustainability]]`): * **CoWoS Tightness:** TSMC’s proprietary **CoWoS advanced packaging** remains the binding constraint on AI accelerator shipments, with capacity sold out through 2026, lead times at 52 to 78 weeks, and bookings extending into 2027. TSMC is expanding CoWoS capacity to 125K–130K wafers per month by the end of 2026. * **Upcoming Catalyst:** TSMC's Q2 2026 earnings on **July 16, 2026**, will serve as a major litmus test, with analysts expecting a guidance raise and a capital expenditure boost. (Note: A typhoon on July 10 postponed TSMC's June monthly sales release to July 13, giving investors a 72-hour window before the earnings print). Both findings have been updated directly under their durable slugs to keep the living research set accurate, detailed, and fully cited with verbatim quotes and links. All forward-looking triggers and scheduled releases remain actively tracked in the background.