Cycle on Jul 22, 2026

I will start by checking the open threads for this topic to understand what specific research questions or watches have been established, and then I will fetch the current public-market view for Nvidia (`NVDA`) to get up-to-date financial metrics, valuation, and news.

I will call `market_view` for Nvidia (`NVDA`) to get the latest public-market data, financial metrics, and recent news regarding Nvidia's AI capex story.

This is a very helpful baseline. We see Supermicro (SMCI) stock jumped 18% in after-hours trading on July 21, 2026, after raising its gross margin forecast and reporting a record backlog exceeding $60 billion. We also see Corning Inc (GLW) shares rising on surging demand for fiber-optic solutions driven by generative AI infrastructure expansion. These are strong external indicators that the AI data center capex boom remains highly active in mid-July 2026. Let's do some targeted web searches. I want to search for recent reports on Nvidia's AI capex story, cloud hyperscalers' capex forecasts (e.g., Google/Alphabet, Microsoft, Meta, Amazon), and general market sentiment as we enter the Q2 2026 earnings season. Alphabet's earnings are coming up around July 25, 2026, and Microsoft/Meta/Amazon should also be reporting or about to report soon. Let's run a search for "Nvidia AI capex July 2026" or "AI capex hyperscaler spend Q2 2026".

Let's search for "Supermicro gross margin raise backlog 60 billion July 2026" or "Supermicro SMCI July 2026 earnings" to fetch details on SMCI's announcement. SMCI is a key proxy for Nvidia's AI capex story because they assemble Nvidia's GPU-based servers (including HGX and Blackwell/Rubin systems). Let's see what the latest news about SMCI says.

Let's fetch the actual press release or news about Supermicro's fiscal Q4 2026 preliminary update from July 21, 2026. This is a very recent and critical catalyst. I will fetch the Yahoo Finance page or the Supermicro IR page to read the exact details of the gross margin raise and $60B backlog. Let's fetch `https://ir.supermicro.com/news/news-details/2026/Supermicro-Provides-Fourth-Quarter-of-Fiscal-Year-2026-Preliminary-Business-Update/default.aspx` or `https://finance.yahoo.com/markets/stocks/articles/supermicro-shares-surge-20-robust-204607039.html` first. Let's try Yahoo Finance since it's a general news source.

This is a massive piece of evidence. - Supermicro's GAAP/non-GAAP gross margins for the quarter ended June 30, 2026 (fiscal Q4 2026) are estimated to land between **15% and 17%**, compared to the previous guidance/expectations of **8.2% to 8.4%**. - Management attributes this to a highly favorable product and customer mix. - Revenue is expected to be near the low end of its $11.0 billion to $12.5 billion guidance. - Order backlog reached an all-time high, with **$60 billion in new orders** raked in during the fourth quarter alone! - The full Q4 and fiscal 2026 results will be released on **August 11, 2026**. Let's do some more search on what this means for Nvidia and the broader AI capex cycle. Is the capex cycle still intact? Yes, a $60B backlog for a single server assembler (SMCI) in a single quarter is jaw-dropping, and shows that the demand for AI infrastructure is not slowing down. Let's look at what other tech giants (hyperscalers) are saying or what analysts are saying about this. Let's do a web search for "Supermicro 60 billion backlog Nvidia Blackwell" or similar terms to see what analysts are writing. Let's also check if there are any recent articles about the Big Tech AI capex sprint in July 2026. Let's search for "Big Tech AI capex July 2026" or "hyperscaler capex Microsoft Google Meta 2026".

Let's fetch `https://finance.yahoo.com/technology/ai/articles/super-micro-says-fourth-quarter-202430991.html` which seems to be the main Reuters/Yahoo article about Supermicro's preliminary update. Let's read it to get more details and quotes.

This is incredibly strong evidence of sustained AI capex. Let's see: Super Micro got orders in excess of $60 billion in a *single quarter* (the fourth quarter of fiscal 2026, ending June 30, 2026). Let's compare this to Supermicro's revenue: its revenue for the quarter is expected to be near the low end of its $11.0B to $12.5B guidance. This means their quarterly order intake is roughly **5x** their actual quarterly revenue, creating a massive, multi-quarter backlog that secures future revenues. Let's search for "Nvidia AI capex intact" or "Nvidia AI capex 2026" to see what other developments there are regarding Nvidia's supply chain, Blackwell, and Rubin platforms. For instance, what about the "Vera Rubin platform broader commercial shipments in H2 2026" watch or "Nvidia Q2 FY2027 Earnings Release" watch? Let's check if there are updates on Blackwell and Rubin. Let's do a web search for "Nvidia Blackwell Rubin shipments H2 2026" or "Nvidia Blackwell delays 2026".

This is a gold mine of extremely fresh and relevant information from July 2026! Let's analyze the details here: 1. **The SemiAnalysis Report and CNBC Article (July 6, 2026):** - SemiAnalysis reported that Nvidia's next-generation AI rack-scale architecture, the **Kyber NVL144** (designed to house its 2027 Rubin Ultra chips), has been delayed by more than 12 months to 2028. - The report also claims that Nvidia has cancelled the 4-compute-die Rubin Ultra GPU, citing packaging issues such as substrate warping. - To offset these setbacks, Nvidia is reportedly planning to increase shipments of its more readily manufacturable Oberon-based Rubin and Rubin Ultra racks, supporting near-term AI infrastructure deployments. - NVL576 (a larger system linking eight racks via optical connections) is also likely delayed or limited to small volumes. 2. **Jensen Huang's Response (around July 17, 2026):** - Nvidia CEO Jensen Huang dismissed reports of a major delay in the Vera Rubin hardware rollout, affirming that Nvidia is on track to produce Vera Rubin systems in "giant" volumes. - He stated that reports of a delay in Vera Rubin's rollout are "not true." 3. **KeyBanc Analyst John Vinh's Note (July 2026):** - John Vinh reported a "slight delay" in the ramp of Nvidia's next-generation Vera Rubin hardware due to thermal heat lid issues in SK Hynix's HBM4 memory qualification. - However, Vinh noted that the thermal lid issues "have been resolved," and that indications show Rubin will start ramping in July 2026. - Vinh still forecasts Nvidia will ship between 1.7 million and 1.8 million Rubin chips this year (2026), alongside 5.5 million to 6.0 million units of its current-generation Blackwell hardware. Let's fetch the CNBC article `https://www.cnbc.com/2026/07/06/nvidia-kyber-rack-system-delays-manufacturing-taiwan-rubin-chips-.html` to get exact quotes and details about the Kyber rack system delays and the SemiAnalysis report. This is extremely important because we have an active watch on this: `Nvidia Kyber NVL144 rack commercial shipment verification` (ID: `019f6d77-03de-7a2b-bf6d-aacefbf423de`). Let's fetch this URL.

This is a highly detailed, excellent piece of research! Let's analyze what this CNBC article (published July 6, 2026) reveals: - **Kyber NVL144 delayed to 2028:** The rack-scale architecture designed to house Nvidia's 2027 Rubin Ultra chips is pushed to 2028 due to manufacturing challenges with its "PCB midplane" (printed circuit board midplane). It connects modules vertically to boost density and reduce latency. - **NVL576 likely delayed or limited:** The larger system linking eight racks via optical connections is also likely delayed or limited to small volumes. - **Backup plan cancelled:** Bolting two current-generation racks together was cancelled due to heavy pushback from CSPs/hyperscalers over its "odd design and heavy operational burden." - **Rubin Ultra 4-compute-die cancelled:** (From the previous summary, but let's double check if there are more details. SemiAnalysis claims Nvidia cancelled the 4-compute-die Rubin Ultra GPU due to packaging issues like substrate warping.) - **Nvidia's response:** Nvidia rejected the report, stating "Our roadmap is intact." - **Current-generation Rubin systems are in full production:** They begin shipping this fall (H2 2026) to eight cloud partners (AWS, Microsoft Azure, Google Cloud, etc.). - **Revenue outperformance projected:** SemiAnalysis projects Nvidia's data-center compute revenue will run **20% above Wall Street consensus** in the second half of fiscal 2027. - **Impact on long-term criticality:** Paul Triolo notes that Nvidia has faced these challenges before and will work with vendors to overcome them. Plus, power supply is the major constraint on AI data centers anyway, so delays might match up with power grid timelines. Let's do some more web search on the Rubin Ultra 4-compute-die cancellation or other aspects of the SemiAnalysis report. Is there a SemiAnalysis post on this? Let's search for "SemiAnalysis Kyber NVL144" or "SemiAnalysis Rubin Ultra cancelled".

Let's fetch `https://www.datacenterdynamics.com/en/news/nvidia-pushes-kyber-release-to-2028-following-manufacturing-concerns-report/` to understand the details of the cancellation of the 4-die Rubin Ultra and the delay of Kyber to 2028. This is highly relevant and detailed.

This is extremely helpful! Let's see: 1. **The 4-die Rubin Ultra cancelled:** - SemiAnalysis reported that Nvidia canceled the original four-die Rubin Ultra design in favor of a two-chiplet design, due to "manufacturing execution concerns" (announced at GTC 2026, and canceled just 3 months later). - This two-die offering will deliver around half the performance of the originally planned four-die Rubin Ultra. 2. **Kyber NVL144 delayed to 2028:** - Pushed from 2027 to 2028 due to issues manufacturing the PCB midplane. - Alternate plan (placing two Oberon racks back-to-back) was scrapped due to customer pushback on its "odd design and heavy operational burden." 3. **NVL576 delayed or limited:** - Connecting 8x Oberon racks over CPO (Co-Packaged Optics) between NVSwitches will likely be delayed or limited to small volumes. 4. **Overall impact:** - Leaves Nvidia with "no proven solution to expand the scale-up world size for Rubin Ultra." - However, this is a delay of the *ultra-advanced scale-up architecture* (Kyber/Rubin Ultra), not the *core* Blackwell and Rubin lines. - In fact, current-generation Rubin systems are in full production and begin shipping this fall (H2 2026) to eight cloud partners, and SemiAnalysis projects Nvidia's data-center compute revenue will run **20% above Wall Street consensus** in the second half of fiscal 2027 (which corresponds to late 2026 / early 2027). - Also, Supermicro's preliminary update on July 21, 2026, showing a **$60 billion backlog** (orders in excess of $60B in Q4 fiscal 2026 alone) and gross margins soaring to **15% to 17%** (vs. 8.2% to 8.4% forecast) proves that near-term demand for AI infrastructure is running at an absolute fever pitch. Let's check if there are other threads we can resolve or record. We have: - `AMD Advancing AI 2026 Keynote and Helios Platform Details` (fire_date: July 23, 2026): This is tomorrow! (Today is July 22, 2026). This watch is still open and should be kept open for tomorrow's cycle or future tracking, but we can note its proximity. - `Nvidia Kyber NVL144 rack commercial shipment verification` (fire_date: late 2027): This is exactly what the SemiAnalysis report is about! The watch is to "Verify if Nvidia begins commercial shipments of the Kyber NVL144 rack-scale systems by late 2027 or if SemiAnalysis's predicted delay to 2028 is borne out." Since the fire date is 2027-12-31, we can keep it open but record that a major development (the SemiAnalysis report and Nvidia's denial) has occurred. - `Nvidia Q2 FY2027 Earnings Release` (fire_date: late August 2026): This is also coming up next month. Let's do some research on other hyperscalers' capex plans to see if they are continuing to spend, or if there is any sign of a slowdown. Let's search for "Google Q2 2026 capex" or "Microsoft Q2 2026 capex" or "Meta Q2 2026 capex". Note that today is July 22, 2026, and Alphabet's Q2 2026 earnings are expected on July 25, 2026 (based on the watch: `Alphabet Q2 2026 Earnings and Dilution Impact Tracking`). Let's see if there are any preliminary reports or expectations for Big Tech's capex in Q2 2026.

This is incredibly useful. Let's look at the dates: - **Alphabet (Google)** reports **today, July 22, 2026 (after US close)**! - **Microsoft** and **Meta Platforms** report on **July 29, 2026 (after US close)**. - **Apple** and **Amazon** report on **July 30, 2026 (after US close)**. This is a perfect setup. Alphabet's earnings report is literally happening TODAY after the close. Let's see if there is any early news or if Alphabet has already reported. It's Wednesday, July 22, 2026. The current time is 00:22 UTC on July 22, so the US market hasn't opened yet for July 22, and Alphabet will report after the close today. Let's verify the current capex projections: - Combined AI capex for Amazon, Google, Meta, and Microsoft is projected to be around **$725 billion in 2026**, which is up **~77% year-over-year** from ~$410 billion in 2025 (some sources say $690B to $750B, but the consensus is massive growth to over $700 billion). - Let's check the CreditSights note on raising hyperscaler capex 2026 estimates. Let's do a quick search on "CreditSights hyperscalers issuance capex mid-year update" or "hyperscaler capex 2026 estimates raised".

This is spectacular! Let's synthesize these findings. 1. **Hyperscaler Capex Estimates Upgraded to ~$750 Billion for 2026:** - CreditSights (Jordan Chalfin and Michael Pugh) has officially upgraded its 2026 capex projection for the top 5 hyperscalers (Amazon, Microsoft, Alphabet, Meta, Oracle) to **~$750 billion**. - This is up massively from their ~$620 billion estimate in January 2026 and ~$600 billion in November 2025. - This represents a **~67% YoY increase** (or up ~$300 billion YoY), marking the third consecutive year of capex growth exceeding 60% YoY. - CreditSights describes the capex-to-revenue ratios as "seemingly untenable" for some of these firms: **86% for Oracle, 54% for Meta, 47% for Microsoft, 46% for Alphabet (Google), and 25% for Amazon**. - Approximately **75%** of this aggregate capex is estimated to be AI-related, meaning more than **$560 billion** is flowing into AI infrastructure in 2026 alone. 2. **Supermicro (SMCI) Blockbuster Preliminary Update (July 21, 2026):** - Supermicro announced orders exceeding **$60 billion** in its fiscal fourth quarter ending June 30, 2026, driven by unprecedented demand for its advanced AI servers (many of which house Nvidia's GPUs). - This is roughly **5x** its expected quarterly revenue of $11B–$12.5B, locking in a massive, multi-quarter delivery pipeline. - Supermicro's gross margins are expected to land between **15% and 17%**, completely resetting the narrative around low-margin server assembly (which previously saw gross margins bottoming at 6.3% in fiscal Q2 2026). This indicates high pricing power and a highly favorable product/customer mix. 3. **Nvidia's Product Roadmap Strains (SemiAnalysis Report):** - On July 6, 2026, SemiAnalysis reported that Nvidia's next-generation **Kyber NVL144** rack-scale architecture (designed to house its 2027 Rubin Ultra chips) has been delayed by more than 12 months to **2028** due to manufacturing challenges with its vertical PCB midplane. - SemiAnalysis also reported that Nvidia has cancelled its original 4-compute-die Rubin Ultra GPU design in favor of a 2-chiplet design due to manufacturing execution concerns, which reduces its performance by half. - The backup plan of bolting two Oberon racks back-to-back was also scrapped due to customer pushback on operational complexity. - NVL576 (linking 8x Oberon racks via optical connections) is also expected to be delayed or limited. - **Nvidia's Response:** Nvidia rejected the report, stating "Our roadmap is intact." - **John Vinh (KeyBanc):** Vinh reported a slight thermal lid delay in the ramp of Nvidia's next-generation Vera Rubin hardware, but noted the issue has been resolved and Rubin is starting to ramp in July 2026. Vinh still forecasts Nvidia shipping 1.7M–1.8M Rubin chips and 5.5M–6.0M Blackwell chips in 2026. - **Near-term Impact:** Despite the long-term Kyber/Rubin Ultra delays, Nvidia's near-term revenue holds extremely strong. SemiAnalysis itself projects Nvidia's data-center compute revenue will run **20% above Wall Street consensus** in the second half of fiscal 2027, driven by the massive rollout of Oberon-based Rubin and Rubin Ultra racks. Let's check if there are any other specific companies we should pull or if we can write up these findings. Let's review the existing notes to see which slugs to update: - `nvidia-record-q1-results-and-trillion-dollar-order-book` is perfect for the overall Nvidia AI capex story, and we should update it with the new details. - `big-tech-capex-reaches-710b-in-2026` can be updated to reflect the new ~$750 billion projection from CreditSights and the upcoming Q2 earnings. - `custom-silicon-and-datacenter-hardware-competition` can be updated or we can keep it as is. - `gpu-specialized-clouds-and-strategic-co-investment` is about neoclouds. Wait! Let's write two highly detailed, evidence-grounded findings: 1. **Nvidia's AI Capex Story Remains Intact Supported by Record Earnings and $1 Trillion Pipeline** (`nvidia-record-q1-results-and-trillion-dollar-order-book`): We will update this note with the Supermicro $60B backlog and gross margin blowout, the CreditSights capex estimate upgrade to $750B, and the details of the SemiAnalysis Kyber NVL144/Rubin Ultra delays vs. Nvidia's pushback and KeyBanc's bullish shipment forecasts. This is a comprehensive update on the overall Nvidia capex story. 2. **Big Tech's AI Capex Reaches Historic Scale as 2026 Sprint Approaches $750 Billion and Cannibalizes Software Budgets** (`big-tech-capex-reaches-710b-in-2026`): We will update this note with CreditSights' upgraded estimate of ~$750 billion for the top 5 hyperscalers (up from $620B in Jan 2026), the "seemingly untenable" capex-to-revenue ratios (86% Oracle, 54% Meta, 47% Microsoft, 46% Alphabet, 25% Amazon), and the timing of the Q2 2026 earnings kick-off with Alphabet reporting today (July 22, 2026). Let's verify if there is any other specific watch we can resolve or update. Wait, let's look at the active watches: - `Nvidia Kyber NVL144 rack commercial shipment verification` (ID: `019f6d77-03de-7a2b-bf6d-aacefbf423de`): This watch is about verifying if the Kyber NVL144 rack ships by late 2027 or is delayed to 2028 as SemiAnalysis predicted. Since this is an active watch for a future date (2027-12-31), we can record an event on it to update the status, but since it's a one_shot watch that hasn't officially resolved yet (it's only July 2026), we should NOT resolve it yet. Wait, `resolves: false` can keep an ongoing watch open, but for a `one_shot` watch, we shouldn't record an event unless it is resolved, or we can record it with `resolves: false` if we want to keep it open? No, the tool description says: "Set resolves=true if this watch is now done (a one_shot catalyst), false if it's an ongoing watch that should keep tracking." Since this watch is a one_shot, we should keep it open and not resolve it yet (or we can record an intermediate event with `resolves: false` if allowed? Let's check: the tool says "resolves: true closes a one_shot watch; false keeps an ongoing one open". For a one_shot watch, if we call it with `resolves: false`, it might fail or behave weirdly. Let's just keep the watch open without recording an event, or if we want to update it we can do so in the findings. Actually, let's look at the watch details again. It's a one-shot watch. Let's leave it open since the final shipment verification is in late 2027.) Let's double-check if we can update the first note `nvidia-record-q1-results-and-trillion-dollar-order-book` or if we should create a new note if the subject is different. The subject is "Is Nvidia's AI capex story still intact?", which is exactly the continuation of the `nvidia-record-q1-results-and-trillion-dollar-order-book` note! So updating it is perfect. Let's read the current contents of the notes we want to update using `read_notes` first to make sure we don't overwrite anything important and can cleanly integrate the new findings.

Let's fetch the full content of `nvidia-record-q1-results-and-trillion-dollar-order-book` and `big-tech-capex-reaches-710b-in-2026` so we can update them seamlessly. Wait, there's no direct tool to read the full body of a note in the declarations, only `read_notes` which returns the previews. Ah! The declarations list `read_notes` which returns slug, title, preview, updated_at, and source_count for each note. Wait, let's see if we can read the full body of a note. There is no `read_note` (singular) tool. But wait! The prompt says: "Before modifying a file, read it first. Do not assume files or directories exist." Wait, are these notes stored as files in the workspace? Let's check the workspace directory: `/var/lib/hey-lefty/notes/019e89b4-58b3-7e3b-a0c2-6a68bd3242db/019e95ad-bd47-7cf5-bd57-eb159d26bbbe`. Let's list the files in that directory using standard shell commands? No, we don't have shell tools declared in the tool declarations! The only tools we have are: - `default_api:list_threads` - `default_api:record_thread` - `default_api:resolve_thread` - `default_api:record_watch` - `default_api:record_event` - `default_api:read_notes` - `default_api:write_note` - `default_api:finish_cycle` - `default_api:web_search` - `default_api:web_fetch` - `default_api:market_view` Ah! So we do not have a shell tool. We only have these specific default_api tools. Since we don't have a tool to read the full body of a note, and `read_notes` only returns a preview, we can write a fresh, comprehensive note under the same slug. The `write_note` tool description says: "reusing a slug updates the existing note and snapshots the prior content into note_revisions." This means we can write the updated full note using the same slug, and it will overwrite/update it. Wait! Let's examine the previews of the two notes we want to update: 1. `nvidia-record-q1-results-and-trillion-dollar-order-book` - Title: "Nvidia's AI Capex Story Remains Intact Supported by Record Earnings and $1 Trillion Pipeline" - Preview: "Nvidia's artificial intelligence capital expenditure story remains remarkably robust in mid-July 2026, anchored by blowout financial results, massive supplier commitments, and solid forward guidance..." 2. `big-tech-capex-reaches-710b-in-2026` - Title: "Big Tech's AI Capex Reaches Historic Scale as 2026 Sprint Approaches $710 Billion and Cannibalizes Software Budgets" - Preview: "The capital expenditure boom supporting artificial intelligence infrastructure has reached unprecedented heights in mid-2026. The world's largest technology companies—Alphabet, Microsoft, Meta, and..." Let's write updated versions of these notes. Wait, let's look at `big-tech-capex-reaches-710b-in-2026`. We can change the title to: `Big Tech's AI Capex Reaches Historic Scale as 2026 Sprint Approaches $750 Billion and Cannibalizes Software Budgets` Or keep the title exactly the same but update the content to reflect the new $750B figure from CreditSights. Let's see: the prompt says "When writing a finding that refines one of the above, reuse its slug to update." Let's design the body for `nvidia-record-q1-results-and-trillion-dollar-order-book`. We want to cover: - How Nvidia's AI capex story remains incredibly robust, supported by the latest blockbuster preliminary fourth-quarter update from Supermicro (SMCI) on July 21, 2026, which revealed over $60 billion in new orders in Q4 alone, representing roughly 5x its expected quarterly revenue. - Supermicro's gross margin guidance being raised to 15%-17% (vs. 8.2%-8.4% prior), signaling immense pricing power and a highly favorable product/customer mix for Nvidia-based AI servers. - The latest details on Nvidia's next-generation Vera Rubin and Rubin Ultra roadmaps. On July 6, 2026, SemiAnalysis reported that Nvidia's vertical rack-scale architecture, the Kyber NVL144, has been delayed by more than 12 months to 2028 due to manufacturing challenges with its vertical PCB midplane. SemiAnalysis also reported that the original 4-compute-die Rubin Ultra has been cancelled in favor of a 2-chiplet design due to packaging concerns. - Nvidia's official response rejecting the report, stating "Our roadmap is intact." - KeyBanc analyst John Vinh's note confirming a minor thermal lid delay in the Rubin ramp (now resolved), with volume shipments beginning in July 2026. Vinh still forecasts Nvidia shipping 1.7 million to 1.8 million Rubin chips and 5.5 million to 6.0 million Blackwell chips in 2026. - SemiAnalysis itself projecting Nvidia's data-center compute revenue will run 20% above Wall Street consensus in the second half of fiscal 2027, driven by massive shipments of Oberon-based Rubin and Rubin Ultra racks. - Grounding the numbers in the public markets view of Nvidia (`NVDA`), which shows TTM revenue of $253.49 billion (+85.2% YoY), gross margin of 74.1%, and quarterly net income of $58.32 billion as of April 30, 2026. Let's draft the body for `nvidia-record-q1-results-and-trillion-dollar-order-book`: ```markdown # Nvidia's AI Capex Story Remains Intact Supported by Record Earnings and $1 Trillion Pipeline Nvidia's artificial intelligence capital expenditure story remains remarkably robust in late July 2026, anchored by blowout financial results, massive supplier commitments, and solid forward guidance. The company’s financial trajectory continues to defy skepticism, supported by a historic order pipeline and unprecedented demand for its next-generation hardware architectures. ### Record Financial Trajectory and Market Dominance According to public market data as of July 22, 2026, Nvidia maintains a $4.92 trillion market capitalization, driven by a trailing twelve-month (TTM) revenue of $253.49 billion, representing an 85.2% year-over-year growth rate, and an earnings growth of 214.5% YoY. For its most recent quarter ending April 30, 2026, Nvidia reported a record $81.61 billion in revenue, a 74.1% gross margin, and $48.59 billion in free cash flow, exceeding analyst earnings estimates for the fourth consecutive quarter with an EPS of $1.87 (vs. $1.77 estimated). ### Supermicro’s Blockbuster $60 Billion Backlog Signals Unprecedented Demand A primary proxy for Nvidia's hardware delivery, AI server maker Super Micro Computer (SMCI), released a blockbuster preliminary fourth-quarter business update on July 21, 2026, which sent its shares surging 20% in after-hours trading. Supermicro revealed that it received orders in excess of **$60 billion** in its fiscal fourth quarter ending June 30, 2026, driven by growing demand for its advanced AI servers housing Nvidia's GPUs. This quarterly order intake is roughly **5x** Supermicro's expected quarterly revenue of $11.0 billion to $12.5 billion, locking in a massive, multi-quarter delivery pipeline. Furthermore, Supermicro completely reset the market's profitability narrative by raising its estimated GAAP and non-GAAP gross margins for the quarter to **15% to 17%** (significantly outpacing its previous guidance of 8.2% to 8.4%). Management directly attributed this expansion to a highly favorable product and customer mix, confirming the immense pricing power of high-end Nvidia-based server configurations. ### Manufacturing Challenges and Vera Rubin Roadmap Adjustments While near-term demand is running at a fever pitch, Nvidia's breakneck annual release cadence has met manufacturing constraints. On July 6, 2026, research firm SemiAnalysis reported that Nvidia's next-generation **Kyber NVL144** rack-scale architecture—designed to house its 2027 Rubin Ultra chips—has been delayed by more than 12 months to 2028 due to manufacturing challenges with its vertical printed circuit board (PCB) midplane. According to the report, Nvidia's backup plan of bolting two Oberon racks back-to-back was scrapped due to customer pushback over its "odd design and heavy operational burden." SemiAnalysis also reported that Nvidia canceled the original four-die Rubin Ultra GPU design in favor of a two-chiplet design due to "manufacturing execution concerns," which will deliver roughly half of the originally planned performance. In response to these reports, Nvidia officially rejected the claims, stating, "Our roadmap is intact." CEO Jensen Huang dismissed reports of a major delay, affirming that Nvidia is on track to produce Vera Rubin systems in "giant" volumes. ### Near-Term Shipments and Financial Resilience Despite long-term roadmap adjustments, Nvidia's near-term revenue trajectory remains highly secure. KeyBanc analyst John Vinh reported in July 2026 that a slight delay in the Rubin ramp caused by thermal lid issues in SK Hynix's HBM4 memory qualification has been fully resolved, with volume shipments beginning in July 2026. Vinh still forecasts that Nvidia will ship **1.7 million to 1.8 million Rubin chips** and **5.5 million to 6.0 million Blackwell chips** in 2026. Crucially, SemiAnalysis itself projects that Nvidia's data-center compute revenue will run **20% above Wall Street consensus** in the second half of fiscal 2027 (late 2026 / early 2027), as Nvidia offsets any Kyber delays by scaling up shipments of its more readily manufacturable Oberon-based Rubin and Rubin Ultra racks. This underscores that even with high-end packaging constraints, the sheer volume of hyperscaler capex guarantees sustained financial outperformance. ## Sources - [NVIDIA Corporation (NVDA) Market View](/markets/NVDA/2026/07/22) — Financial metrics, valuation, and recent news as of July 22, 2026. - [Supermicro shares surge 20% as robust margins, $60B backlog eclipse soft revenue](https://finance.yahoo.com/markets/stocks/articles/supermicro-shares-surge-20-robust-204607039.html) — "Supermicro announced that its GAAP and non-GAAP gross margins for the quarter ended June 30, 2026, are estimated to land between 15% and 17%." - [Super Micro says fourth-quarter orders topped $60 billion](https://finance.yahoo.com/technology/ai/articles/super-micro-says-fourth-quarter-202430991.html) — "Super Micro Computer said on Tuesday it has received orders in excess of $60 billion in the fourth quarter, helped by growing demand for its advanced AI servers..." - [Nvidia’s next-gen AI rack system delayed to 2028 on manufacturing snags, SemiAnalysis says](https://www.cnbc.com/2026/07/06/nvidia-kyber-rack-system-delays-manufacturing-taiwan-rubin-chips-.html) — "The Kyber NVL144 has been pushed to 2028 due to difficulties in manufacturing a key circuit board: SemiAnalysis." - [Nvidia pushes Kyber release to 2028 following manufacturing concerns – report](https://www.datacenterdynamics.com/en/news/nvidia-pushes-kyber-release-to-2028-following-manufacturing-concerns-report/) — "The news comes a week after SemiAnalysis separately reported that Nvidia had canceled the four-die Rubin Ultra design in favor of the two-chiplet design, also due to 'manufacturing execution concerns.'" - [Nvidia Stock Rises Despite 'Slight Delay' in Rubin Chip Rollout](https://www.barrons.com/articles/nvidia-stock-price-rubin-chips-1988209e) — "Vinh expects the Rubin ramp to begin in July and still forecasts Nvidia will ship 1.7 million to 1.8 million Rubin chips this year." ``` This is an incredibly detailed and well-sourced note that directly answers the user's brief. Now let's design the body for `big-tech-capex-reaches-710b-in-2026` (we can keep the slug exactly the same, but we can change the title to reflect $750B if we want, or keep the title close to it). Let's see: CreditSights upgraded its 2026 capex projection for the top 5 hyperscalers to **~$750 billion** (up from $620B in January 2026). Let's draft the body for `big-tech-capex-reaches-710b-in-2026`: ```markdown # Big Tech's AI Capex Reaches Historic Scale as 2026 Sprint Approaches $750 Billion and Cannibalizes Software Budgets The capital expenditure boom supporting artificial intelligence infrastructure has reached unprecedented heights in mid-2026. The world's largest technology companies—Alphabet, Microsoft, Meta, Amazon, and Oracle—are engaged in a historic infrastructure sprint, raising their combined financial commitments to levels that exceed the budgets of major national megaprojects. ### Capex Estimates Upgraded to ~$750 Billion In a mid-year update published in July 2026, credit research firm CreditSights raised its 2026 capital expenditure projection for the top five hyperscalers (Amazon, Microsoft, Alphabet, Meta, and Oracle) to **~$750 billion**. This represents a massive increase from CreditSights' estimate of ~$620 billion in January 2026 and ~$600 billion in late 2025. The revised estimate implies that 2026 capex will grow by **~$300 billion YoY, or ~67% YoY**. This marks the third consecutive year in which hyperscaler capex growth has exceeded 60% year-over-year. CreditSights estimates that approximately **75%** of this aggregate spend (~$560 billion) is dedicated exclusively to AI infrastructure, including graphics processing units (GPUs), custom accelerators, high-bandwidth memory, optical networking, and specialized liquid-cooled data center facilities. ### "Seemingly Untenable" Capital Intensity and the Shift to Debt This breakneck spending pace is consuming an unprecedented share of corporate resources. CreditSights describes the projected 2026 capex-to-revenue ratios across these tech giants as "seemingly untenable" under traditional software business models: - **Oracle (ORCL):** Capex is projected to consume **86%** of revenue. - **Meta Platforms (META):** Capex is projected to consume **54%** of revenue. - **Microsoft (MSFT):** Capex is projected to consume **47%** of revenue. - **Alphabet (GOOGL):** Capex is projected to consume **46%** of revenue. - **Amazon (AMZN):** Capex is projected to consume **25%** of revenue. With capital expenditures now consuming up to 94% of operating cash flows (after accounting for dividends and buybacks) for some hyperscalers, these historically cash-rich giants are increasingly turning to debt markets. They are issuing USD bonds in unprecedented sizes to bridge the gap between rapidly rising AI capex budgets and internal free cash flow, transforming their capital structures to support the AI supercycle. ### Q2 2026 Earnings Kick-Off and Capex Scrutiny The sustainability of this capex sprint is the central question of the Q2 2026 earnings season, which kicks off in late July 2026. Investors are closely monitoring whether AI monetization is beginning to catch up with infrastructure spend. - **Alphabet (Google):** Reports Q2 2026 earnings on **July 22, 2026** (after the US close). Investors are highly focused on Google's capital intensity, cash balances, and potential share count dilution following its massive capital raise. - **Microsoft and Meta Platforms:** Scheduled to report on **July 29, 2026**. - **Amazon and Apple:** Scheduled to report on **July 30, 2026**. Despite concerns about the long-term return on investment (ROI) of generative AI, the short-term competitive dynamics of the "AI arms race" prevent any single hyperscaler from slowing down. The risk of under-investing and falling behind in AI capabilities continues to outweigh the risk of over-building, cementing Nvidia's position as the primary beneficiary of this historic capital reallocation. ## Sources - [Tech: Raising Hyperscaler Capex 2026 Estimates](https://know.creditsights.com/insights/tech-raising-hyperscaler-capex-2026-estimates/) — "We are now projecting ~$750 bn in capex for the top 5 hyperscalers, which is up massively from our ~$620 bn estimate in January 2026..." - [Hyperscaler CapEx Hits $600B in 2026](https://introl

I will write the first updated research finding regarding Nvidia's AI capex story, incorporating the latest July 2026 developments (such as Supermicro's $60B backlog, SemiAnalysis's Kyber NVL144 delay report, and KeyBanc's shipment estimates). This will update the existing note under the slug `nvidia-record-q1-results-and-trillion-dollar-order-book`.

I will now write the second updated research finding regarding the broader hyperscaler AI capex sprint, incorporating CreditSights' upgraded ~$750 billion projection for the top five hyperscalers, their "seemingly untenable" capital intensity ratios, and the timeline for Q2 2026 earnings starting with Alphabet today. This will update the existing note under the slug `big-tech-capex-reaches-710b-in-2026`.

I will now call `finish_cycle` to complete this research cycle, as I have updated both key notes on Nvidia's AI capex story and Big Tech's $750B capex trajectory with high-quality, up-to-date, cited evidence.

I have completed the research cycle on the topic: **"Is Nvidia's AI capex story still intact?"** ### Summary of Key Findings 1. **Supermicro's Blockbuster $60B Backlog Signals Unprecedented Demand** - On July 21, 2026, AI server assembler Super Micro Computer (SMCI) reported receiving over **$60 billion in new orders** during its fiscal fourth quarter ending June 30, 2026. This quarterly order intake is roughly **5x** its expected quarterly revenue of $11B–$12.5B, locking in a massive multi-quarter delivery pipeline. - SMCI raised its GAAP/non-GAAP gross margins estimate to **15% to 17%** (vs. previous guidance of 8.2% to 8.4%), which management directly attributed to a highly favorable product and customer mix. This demonstrates immense pricing power for Nvidia-based AI server configurations and completely resets the low-margin narrative around server assembly. 2. **Hyperscaler Capex Estimates Upgraded to ~$750 Billion for 2026** - Credit research firm CreditSights raised its 2026 capital expenditure projection for the top five hyperscalers (Amazon, Microsoft, Alphabet, Meta, and Oracle) to **~$750 billion** (up from ~$620 billion in January 2026). - This represents a massive **~67% YoY increase** (~$300 billion YoY), with approximately **75%** of the spend (~$560 billion) dedicated exclusively to AI infrastructure. - CreditSights describes the projected 2026 capex-to-revenue ratios across these tech giants as "seemingly untenable" under traditional software business models: **86% for Oracle, 54% for Meta, 47% for Microsoft, and 46% for Alphabet**. This has forced historically cash-rich giants to turn to debt markets in unprecedented size to fund the buildout. - The Q2 2026 earnings season kicks off **today, July 22, 2026, with Alphabet (Google) reporting after the US close**, putting these capex budgets and early AI monetization under intensive investor scrutiny. 3. **Manufacturing Constraints and Vera Rubin Roadmap Adjustments** - On July 6, 2026, SemiAnalysis reported that Nvidia's vertical rack-scale architecture, the **Kyber NVL144** (designed to house its 2027 Rubin Ultra chips), has been delayed by more than 12 months to **2028** due to manufacturing challenges with its vertical printed circuit board (PCB) midplane. - SemiAnalysis also reported that Nvidia canceled its original 4-compute-die Rubin Ultra GPU design in favor of a 2-chiplet design due to "manufacturing execution concerns," cutting its performance by half. - While Nvidia rejected these reports (stating "Our roadmap is intact"), and KeyBanc analyst John Vinh reported that minor thermal lid issues in Rubin have already been resolved with volume shipments starting in July 2026, the long-term roadmap adjustments represent a minor near-term bump. - In fact, SemiAnalysis projects that Nvidia's data-center compute revenue will run **20% above Wall Street consensus** in the second half of fiscal 2027 (late 2026 / early 2027), as Nvidia offsets Kyber delays by scaling up shipments of its more readily manufacturable Oberon-based Rubin and Rubin Ultra racks. ### Action Taken - Updated the finding: `[[nvidia-record-q1-results-and-trillion-dollar-order-book]]` (Nvidia's overall AI capex and roadmap status). - Updated the finding: `[[big-tech-capex-reaches-710b-in-2026]]` (Hyperscaler capex trajectory upgraded to $750B). - Both notes have been grounded in concrete public-market data, analyst reports, and press releases with exact verbatim quotes and stable URLs.