Cycle on Sep 10, 2026

I will list the current open threads for this topic to understand what specific questions or watches are already active.

I will read the existing notes for this research topic to see what is already documented.

I will pull the public-markets data for NVIDIA (ticker: NVDA) first, to get its current financials, guidance, and market trajectory. This is a crucial step under the public-markets guidelines.

The market view of NVDA is incredibly robust, with trailing twelve-month revenue at $302.97 billion (+105.9% YoY) and quarterly revenue in the latest quarter ending July 31, 2026 at $96.22 billion. Wait, let's look at the news mentioned in the market view: "Microsoft (NASDAQ: MSFT) 'Useful Yield' Push Puts NVIDIA's (NASDAQ: NVDA) AI Economics To The Test" (September 9, 2026). "HP Extends Data-Center AI Architecture to the Edge... powered by the NVIDIA GB300 Grace Blackwell Ultra..." (September 9, 2026). "AMD Rises 3% as CFO Lifts AI Chip Market Forecast Toward $3 Trillion..." (September 9, 2026). Let's search for "Microsoft Useful Yield NVIDIA", "Nvidia AI capex story intact", "hyperscaler AI capex 2026 2027", and search for recent earnings or reports on whether Nvidia's capex story is intact. I'll search using `web_search`.

Let's fetch the full text of "Microsoft (NASDAQ: MSFT) “Useful Yield” Push Puts NVIDIA’s (NASDAQ: NVDA) AI Economics To The Test" or "Microsoft Makes “Useful Yield” the New Metric for AI Infrastructure". Let's see if we can fetch `https://convergedigest.com/microsoft-useful-yield-ai-infrastructure/` or `https://www.foreignpolicyjournal.com/2026/09/09/microsoft-nasdaq-msft-useful-yield-push-puts-nvidias-nasdaq-nvda-ai-economics-to-the-test/`. Let's run `web_fetch` on `https://www.foreignpolicyjournal.com/2026/09/09/microsoft-nasdaq-msft-useful-yield-push-puts-nvidias-nasdaq-nvda-ai-economics-to-the-test/`.

This is extremely interesting. Let's look up the actual Microsoft blog post or announcement by Rani Borkar at SEMICON Taiwan on September 2, 2026. Let's do a search for `"Rani Borkar" "SEMICON Taiwan" OR "useful yield"` or `"Rani Borkar" "useful yield"`.

Let's fetch `https://news.microsoft.com/source/asia/2026/09/02/the-new-imperative-for-ai-infrastructure-useful-yield/` using `web_fetch`. This is a primary source from Microsoft itself. Let's do it.

Let's look at the other Nikkei article: "More memory chips not the solution to AI bottlenecks: Microsoft exec" `https://asia.nikkei.com/business/technology/artificial-intelligence/more-memory-chips-not-the-solution-to-ai-bottlenecks-microsoft-exec`. Let's fetch it using `web_fetch`. This will give us additional context on Microsoft's stance on HBM and memory bottlenecks.

Ah, the Nikkei article was cut off or is short, but the headline and first paragraph are clear: "More memory chips not the solution to AI bottlenecks: Microsoft exec... Building massive new memory chip capacity will not resolve the supply chain bottlenecks holding back the AI infrastructure build out, according to a senior Microsoft executive." Let's search for more details on this Nikkei article using `web_search` to see if there are other quotes or insights from Rani Borkar's speech. Let's do a search for `"More memory chips not the solution to AI bottlenecks: Microsoft exec"`.

Let's search for "Borkar" "memory chips" "SEMICON" to see if there are other articles with quotes from her speech in Taipei. Let's do a `web_search`.

Let's see if we can do a broader search on the Nikkei article to get more text, or search on Google / Bing for "More memory chips not the solution to AI bottlenecks: Microsoft exec" "Rani Borkar". Let's do `web_search`.

Ah, let's search for "Microsoft says more memory chips alone won't solve AI infrastructure bottlenecks: Rani Borkar" or similar terms to see if other news sites picked up the Nikkei article. Let's do `web_search`.

Let's see if we can search for the original Nikkei text by looking up "More memory chips not the solution to AI bottlenecks" on Google or Bing, specifically looking for paragraphs or quotes from it. Let's do `web_search`.

Let's search for "Borkar" "useful yield" "memory" "Nikkei" on Google to see if we can find any quoted text from the article. Let's do a broader search with `web_search` for `"Rani Borkar" "SEMICON Taiwan" "memory"`.

Let's see if we can search for other recent articles about Nvidia's AI capex story being intact, or hyperscaler capital expenditure updates in Q3 2026. Let's do a `web_search` for `Nvidia AI capex story intact` or `hyperscaler capex Q3 2026` or `Nvidia Q2 2027 earnings capex`. Note that NVDA just reported fiscal Q2 2027 earnings on August 26, 2026 (as mentioned in the prior findings: "During NVIDIA Corporation's (NVDA) fiscal Q2 2027 earnings call on August..."). Let's check NVDA's recent earnings call or reports on it. Let's do a `web_search` for `"Nvidia" "Q2 2027" earnings call transcript capex` or `"Nvidia" "fiscal Q2 2027" capex`.

Ah! Let's fetch the transcript of NVIDIA's Q2 2027 earnings call to see what exactly was discussed regarding capex, demand, and future growth. Wait, let's look at the URL from Q4 CDN: `https://s201.q4cdn.com/141608511/files/content_files/TRANSCRIPT_-NVIDIA-Corp-NVDA-US-Q2-2027-Earnings-Call-26-August-2026-5_00-PM-ET.pdf`. Let's fetch this PDF using `web_fetch`. This is a primary source of NVIDIA's Q2 2027 earnings transcript! Let's call `web_fetch`.

This transcript contains vital details about the "NVIDIA AI capex story". Let's analyze what Colette Kress and Jensen Huang said: 1. **Fiscal 2028 Revenue Growth Guidance:** "We expect to grow revenue by approximately 70% in fiscal 2028. This is a supply-constrained outlook." 2. **Q2 Data Center Revenue:** "Q2 data center revenue increased 18% quarter-over-quarter to $89 billion..." (which is 92.5% of total revenue of $96.22 billion). 3. **Hyperscale sequential growth:** "Hyperscale revenue of $49 billion grew 13% sequentially, driven by sustained strength in Blackwell." 4. **AWS Expansion Partnership:** "Building on its already vast installed base of NVIDIA Compute, AWS is deploying an additional 2 million GPUs starting this quarter through the second quarter of fiscal 2029." (This is a massive confirmation of demand!). 5. **Vera Rubin Ramp:** "We commenced production shipments of Vera Rubin earlier this month [August 2026]. Having already received purchase orders from every major hyperscaler, AI cloud, and system OEM, we expect Vera Rubin to mark the fastest product ramp in NVIDIA's history." 6. **Customer Forecasts:** "Incredibly, we are seeing demand acceleration even at our scale. Customers' forecasts point to our growth doubling next year." 7. **Supply constraints vs. demand:** When asked about the 70% growth guidance vs. 100% demand growth, Jensen says: "you can no longer procure technology per se and stand up this infrastructure. You've got to secure the land, power, and shell, which oftentimes is a couple, two, three years out. We want to be direct about this rather than let it linger as an open question." Wait! Let's examine if there are other players' capex numbers or comments. Let's search for "hyperscaler capex 2026 2027" or "Microsoft capex Q2 2026" or "Alphabet capex Q2 2026" to see if there are updates on their spending. Let's do a `web_search`.

Ah! The web search budget is spent. This is a very important warning. Let's make sure we write up our findings carefully and accurately based on the evidence we have retrieved. Let's review the active threads and watches to see if we can resolve any of them or if any watches have fired based on this cycle's evidence. Wait, let's look at the list of threads and watches: 1. `Nvidia Q3 FY2027 Earnings Release and Gross Margin Performance` (fire date 2026-11-17T00:00:00Z) - Has this happened yet? No, today's date is September 10, 2026. 2. `Nvidia FY2028 Revenue Growth (Guided ~70%)` (fire date 2028-02-25T00:00:00Z) - Has this happened yet? No. 3. `AWS 2 Million Nvidia GPU Deployment (Blackwell Ultra / Rubin)` (fire date 2028-12-31T00:00:00Z) - Has this happened yet? No. But wait! There is a watch or thread on AWS 2 million GPU deployment: - Thread ID: `01a0500e-ca32-7c63-99f0-d9956679bf91` - Title: `AWS 2 Million Nvidia GPU Deployment (Blackwell Ultra / Rubin)` - Description: "Monitor the successful deployment of AWS's committed 2 million additional Nvidia GPUs (Blackwell Ultra, Rubin, and Rubin Ultra) across its global cloud infrastructure by the end of 2028." - Wait, did we get some information about this? Yes! In the Q2 2027 earnings call on August 26, 2026, Colette Kress announced: "Today, we are delighted to announce an expansion of our partnership with AWS. Building on its already vast installed base of NVIDIA Compute, AWS is deploying an additional 2 million GPUs starting this quarter through the second quarter of fiscal 2029." - This is the *announcement* of the deployment. The watch itself is to monitor the *successful deployment* by the end of 2028 (which is ongoing/one-shot with fire date 2028-12-31). So it has not fired yet, but we have a major update on it which we should include in the finding. Wait, are there any other watches we can resolve or record events for? Let's see: - `Google Cloud Q3 2026 Operating Margin and Third-Party Capacity Impact` (fire date 2026-10-25T00:00:00Z) - Not yet. - `Google 110,000-GPU Colossus 2 Lease Commencement` (fire date 2026-10-15T00:00:00Z) - Not yet. - `Oracle 50,000-GPU AMD Instinct Helios Supercluster Deployment` (fire date 2026-09-30T00:00:00Z) - Not yet. - `RTX Spark laptops and desktops retail launch` (fire date 2026-10-01T00:00:00Z) - Not yet. Let's check if there are any other threads we can resolve. Wait, let's look at the topic brief: "Is Nvidia's AI capex story still intact?" This is the core question of this cycle. Let's write a comprehensive finding under the slug `big-tech-capex-reaches-710b-in-2026` (or a related slug). Wait! Let's check the existing notes: 1. `big-tech-capex-reaches-710b-in-2026`: "Big Tech's AI Capex Surpasses $800 Billion in 2026 on Track for $1.3 Trillion in 2027 as Nvidia Projects 70% Growth" Let's update this note! Wait, are there other notes we should update? Let's check `nvidia-server-price-hikes-and-margin-protection`: "Nvidia's AI Margin Protection: HBM Cost Inflation Triggers Gross Margin Reset as Samsung Mounts Custom HBM4E Challenge" Let's see if we should write a new finding or update `big-tech-capex-reaches-710b-in-2026`. Let's update `big-tech-capex-reaches-710b-in-2026` because it's the exact match for "Nvidia's AI capex story". Wait, let's also check if there is another note we should update. Let's review the details we have: - **Nvidia's Q2 FY2027 Earnings (reported August 26, 2026):** - Revenue reached $96.22 billion (+106% YoY), beating estimates of $92.07 billion. - Data Center revenue reached $89 billion (92.5% of total), with hyperscale revenue at $49 billion (+13% sequentially). - Operating income of $63.73 billion, net income of $59.69 billion. - Gross margin was 74.7% (matching TTM). - Guided for fiscal 2028 revenue growth of approximately 70%, which is a supply-constrained outlook. - Jensen Huang stated that "customers' forecasts point to our growth doubling next year" (implying demand growth of 100%, but supply constraints limit guided growth to 70%). - Jensen also noted that the bottleneck is no longer just chips but physical infrastructure: "you can no longer procure technology per se and stand up this infrastructure. You've got to secure the land, power, and shell, which oftentimes is a couple, two, three years out." - AWS partnership expansion: AWS is deploying an additional 2 million GPUs starting this quarter (Q3 FY2027) through Q2 FY2029. - Vera Rubin platform: Commenced production shipments in August 2026, with purchase orders from every major hyperscaler, AI cloud, and OEM. Expects the fastest product ramp in company history. - **Microsoft's "Useful Yield" Imperative (presented September 2, 2026 at SEMICON Taiwan by Rani Borkar):** - Microsoft is shifting the AI infrastructure metric from raw capacity (megawatts, chip counts, memory bytes) to "useful yield", which is defined as: `Capability × Deployment Velocity × Utilization = Useful Yield`. - Borkar argued that "building massive new memory chip capacity will not resolve the supply chain bottlenecks holding back the AI infrastructure build out" and that "more memory chips alone won't solve AI infrastructure bottlenecks." - Instead, Microsoft is pushing a "silicon-to-systems" approach (cross-layer hardware-software codesign), optimizing across silicon, memory, networking, power, cooling, software, models, and fleet operations. - This puts Nvidia's economics to the test because custom designs (like Microsoft's Azure Maia accelerator and Cobalt CPU) allow hyperscalers to optimize directly for their workloads, potentially capturing more of the value or limiting the margins of external suppliers like Nvidia. - However, Microsoft and Nvidia remain deeply intertwined: Nvidia reported that Microsoft Azure is already running Vera Rubin racks. - **Hyperscaler Capex Projections (as of late August / September 2026):** - Combined 2026 capex for the top hyperscalers (Amazon, Microsoft, Alphabet, Meta, Oracle) is now projected to reach $725 billion (or up to $733 billion / $745 billion depending on the analyst tally), up significantly from 2025. - Amazon is leading at ~$200 billion, Microsoft near $190 billion, Alphabet at $175-$185 billion, and Meta at $115-$135 billion. - UBS projects total hyperscaler capex will reach $1.009 trillion in 2026, $1.447 trillion in 2027, and $1.619 trillion in 2028, totaling over $4.1 trillion from 2026 to 2028. - Hyperscalers are recycling nearly all of their cloud revenues back into capex (UBS estimates they will spend 102% of cloud revenue on capex in 2026). - This massive spending has put pressure on free cash flow (Alphabet turned cash flow negative in Q2 2026, and BofA/Evercore estimate severe free cash flow pressure on others), raising the stakes for monetization returns. This is an incredibly rich, deep, and cohesive story. It perfectly answers the user's brief: "Is Nvidia's AI capex story still intact?" The answer is: Yes, the AI capex story is not only intact but has gone parabolic, with hyperscaler capex now projected to surpass $1 trillion in 2026 (UBS) and $1.4 trillion in 2027, while Nvidia itself guides to ~70% revenue growth in FY2028 under severe supply constraints (with customer demand pointing to a doubling of growth). However, the narrative is shifting from a raw capacity land grab to an "efficiency and useful yield" phase, exemplified by Microsoft's new "useful yield" framework. This shift tests Nvidia's margins and pricing power, as hyperscalers increasingly use custom silicon and system-level co-design to optimize tokens per dollar and watt. Let's write two distinct notes (or update them): 1. Update `big-tech-capex-reaches-710b-in-2026` to capture Nvidia's Q2 FY2027 earnings print, the massive 70% growth guidance for FY2028, the AWS 2 million GPU expansion, and the updated $725B+ (or $1T+ according to UBS) hyperscaler capex figures. 2. Update `nvidia-server-price-hikes-and-margin-protection` or create a new note on Microsoft's "Useful Yield" push and its implications for Nvidia's AI economics. Wait! Let's check the guidelines: "Name a slug for the DURABLE SUBJECT it tracks: the throughline, entity, or running debate — never the transient story-of-the-day... 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." Wait, do we have an existing note on Microsoft's useful yield or AI economics? Let's check the existing notes: - `nvidia-server-price-hikes-and-margin-protection` is about "Nvidia's AI Margin Protection: HBM Cost Inflation Triggers Gross Margin Reset as Samsung Mounts Custom HBM4E Challenge". This note is about margin protection and HBM cost inflation. - `off-balance-sheet-ai-infrastructure-financing-models` is about off-balance-sheet financing. - `big-tech-capex-reaches-710b-in-2026` is about Big Tech's AI capex. - `custom-silicon-and-datacenter-hardware-competition` is about "Custom Silicon and Data Center Hardware Competition Escalates as AMD Posts Record Quarter But Falls on SpaceX Exclusive Nvidia Pledge". Wait! Microsoft's "Useful Yield" push is a direct challenge to Nvidia's AI economics and falls squarely under `custom-silicon-and-datacenter-hardware-competition` or `nvidia-server-price-hikes-and-margin-protection` or we can update `custom-silicon-and-datacenter-hardware-competition` or `big-tech-capex-reaches-710b-in-2026` or create a specific note if it's a new running debate. Wait, let's check if we can update `big-tech-capex-reaches-710b-in-2026` with the capex numbers and Nvidia's Q2 results, and update `custom-silicon-and-datacenter-hardware-competition` (or `nvidia-server-price-hikes-and-margin-protection`) with the "Useful Yield" framework since it directly concerns custom silicon (Maia, Cobalt) and the efficiency shift. Let's look at `custom-silicon-and-datacenter-hardware-competition`'s preview: "While Nvidia's near-to-mid-term artificial intelligence infrastructure demand remains exceptionally robust (see [[big-tech-capex-reaches-710b-in-2026]]), the competitive landscape for AI accelerators..." And `nvidia-server-price-hikes-and-margin-protection`'s preview: "While NVIDIA Corporation continues to deliver record top-line growth (see [[big-tech-capex-reaches-710b-in-2026]]), the highly profitable gross margin profile of its AI hardware has run into a..." Let's update `big-tech-capex-reaches-710b-in-2026` first with the full details of Nvidia's Q2 FY2027 earnings and the parabolic capex numbers. Wait, let's read the full content of `big-tech-capex-reaches-710b-in-2026` first? We don't have its full body, only the preview. But we can overwrite it with a comprehensive, fully grounded body that includes the new evidence we just gathered. Let's structure the updated `big-tech-capex-reaches-710b-in-2026` note: - **Title:** Big Tech's AI Capex Parabolic Run: Hyperscaler Spending Surpasses $725B in 2026 as Nvidia Guides to 70% Growth in FY2028 - **Content:** - Ground it in NVDA's Q2 FY2027 earnings report (August 26, 2026) and the latest hyperscaler capex projections. - Mention that Nvidia reported Q2 FY2027 revenue of $96.22 billion (+106% YoY), with Data Center revenue at $89 billion (92.5% of total). - Cite the `/markets/NVDA/2026/09/10` market view for NVDA's financials (TTM revenue of $302.97B, profit margin 63.7%, net income of $59.69B). - Detail Nvidia's FY2028 guidance of ~70% revenue growth, which is heavily supply-constrained. Customers' forecasts point to demand growth doubling (100%), but supply constraints (including securing land, power, and shell, which takes 2-3 years) limit guided growth to 70%. - Detail the AWS expansion: deploying an additional 2 million GPUs from Q3 FY2027 through Q2 FY2029. - Detail the Vera Rubin platform: production shipments commenced in August 2026, with purchase orders from every major hyperscaler, AI cloud, and system OEM, expected to mark the fastest product ramp in company history. - Highlight the parabolic scale of hyperscaler capex: Big Five (Amazon, Microsoft, Alphabet, Meta, Oracle) are collectively spending over $725 billion in 2026. Amazon is leading at ~$200B, Microsoft near $190B, Alphabet at $175-$185B, and Meta at $115-$135B. - Cite the UBS projection that hyperscalers will spend $1.009 trillion on capex in 2026, rising to $1.447 trillion in 2027 and $1.619 trillion in 2028 (totaling over $4.1 trillion). Hyperscalers are recycling nearly all cloud revenues into AI capex, spending 102% of cloud revenues in 2026. - Mention the growing investor scrutiny as free cash flows turn negative (e.g., Alphabet in Q2 2026). Let's write this note first. Let's make sure we include exact quotes and links. Wait, let's look at the sources we can cite for `big-tech-capex-reaches-710b-in-2026`: - Nvidia Q2 2027 Earnings Transcript: `https://s201.q4cdn.com/141608511/files/content_files/TRANSCRIPT_-NVIDIA-Corp-NVDA-US-Q2-2027-Earnings-Call-26-August-2026-5_00-PM-ET.pdf` - REX Shares article: `https://www.rexshares.com/nvidia-earnings/` - AI Weekly capex article: `https://aiweekly.co/alerts/amazon-microsoft-alphabet-meta-plan-725b-ai-capex-in-2026` - Futurum Group article: `https://futurumgroup.com/insights/ai-capex-2026-the-690b-infrastructure-sprint/` - Yahoo Finance / 247 Wall St article: `https://finance.yahoo.com/technology/ai/articles/ai-absurd-spending-boom-hyperscalers-162709082.html` - CNBC article: `https://www.cnbc.com/2026/07/28/hyperscalers-face-higher-capex-scrutiny-after-alphabet-report-panned.html` - Market view: `/markets/NVDA/2026

I will write the updated note for `big-tech-capex-reaches-710b-in-2026` to document Nvidia's latest Q2 FY2027 earnings print, its massive FY2028 guidance, the AWS 2 million GPU expansion, and the parabolic surge in hyperscaler capex.

I will now write a second note under the slug `nvidia-server-price-hikes-and-margin-protection` (or a related slug) to document Microsoft's "Useful Yield" push and its implications for Nvidia's AI economics, linking the two findings together via wikilinks as required by the guidelines. Let's do it.

I will now check if there are any other threads that can be resolved. Let's see: 1. `Nvidia FY2028 Revenue Growth (Guided ~70%)` (Thread ID: `01a0409b-c90b-7527-a2c7-f7064fc17243`). Wait! In this cycle, we got the official Q2 FY2027 earnings call where Nvidia *officially guided* to ~70% revenue growth for FY2028. Wait! The watch is "Verify if Nvidia's full-year FY2028 revenue meets or exceeds the guided 70% growth rate as supply constraints are resolved." The fire date is 2028-02-25 (which is when FY2028 full-year results will be reported). So this watch cannot be resolved yet. It remains active. Wait, is there any other watch that has fired? Let's check: - `AWS 2 Million Nvidia GPU Deployment (Blackwell Ultra / Rubin)` (Thread ID: `01a0500e-ca32-7c63-99f0-d9956679bf91`). The watch is "Monitor the successful deployment of AWS's committed 2 million additional Nvidia GPUs (Blackwell Ultra, Rubin, and Rubin Ultra) across its global cloud infrastructure by the end of 2028." Since the deployment is scheduled to run through Q2 of fiscal 2029, the deployment itself has just *started* this quarter. So this watch is ongoing and has not fully resolved, but we recorded the announcement. We will keep it open. Let's call `finish_cycle` to complete this research cycle, as we have updated the two critical notes detailing the status of Nvidia's AI capex story.

I have completed the research cycle on the topic: **"Is Nvidia's AI capex story still intact?"** ### Summary of Findings 1. **The Capex Story is Parabolic and Intact:** - **Nvidia's Blockbuster Q2 FY2027 Results (reported August 26, 2026):** Nvidia reported record quarterly revenue of **$96.22 billion (+106% YoY)**, beating consensus estimates of $92.07 billion. Trailing twelve-month (TTM) revenue has now reached **$302.97 billion** with an exceptionally strong **74.7% gross margin** and a **63.7% net profit margin** (yielding **$59.69 billion in quarterly net income**). - **Bullish Forward Guidance:** Nvidia guided for **fiscal 2028 revenue growth of approximately 70%**. Management explicitly noted that this outlook is **supply-constrained**, with underlying customer demand forecasts pointing to a **doubling of growth (100%)** next year. - **Physical Infrastructure Bottlenecks:** CEO Jensen Huang highlighted that the bottleneck has shifted from chip fabrication to broader datacenter infrastructure, noting that companies can no longer simply buy chips but must secure land, power, and building shells 2–3 years in advance. - **AWS Expansion:** AWS has committed to deploying an **additional 2 million Nvidia GPUs** (Blackwell Ultra, Rubin, and Rubin Ultra) starting this quarter through Q2 FY2029. - **Vera Rubin Platform:** Commenced production shipments in August 2026, with purchase orders secured from every major hyperscaler, AI cloud, and OEM, tracking for the fastest product ramp in Nvidia's history. - **Parabolic Hyperscaler Capex:** Combined 2026 capex for the Big Five (Amazon, Microsoft, Alphabet, Meta, Oracle) is now projected to exceed **$725 billion** (up 77% from 2025). UBS estimates that hyperscalers are recycling **102% of their cloud revenues into capex**, with total industry capex projected to hit **$1.009 trillion in 2026**, **$1.447 trillion in 2027**, and **$1.619 trillion in 2028** (exceeding $4.1 trillion over three years). 2. **The New "Useful Yield" Imperative:** - **Microsoft's Strategic Pivot (September 2, 2026):** At SEMICON Taiwan, Microsoft's Azure hardware chief Rani Borkar introduced the **"yield imperative"** and proposed a new industry-standard metric: **Useful Yield** ($\text{Capability} \times \text{Deployment Velocity} \times \text{Utilization} = \text{Useful Yield}$). - **Systems-Over-Silicon:** Microsoft is arguing that "building massive new memory chip capacity will not resolve supply chain bottlenecks" and is instead pushing a "silicon-to-systems" approach. This involves cross-layer hardware-software codesign across silicon, memory, networking, power, and cooling (e.g., integrating NICs directly into custom Maia accelerators and programming fine-grained power controls onto custom Cobalt CPUs). - **Implications for Nvidia:** This shift tests Nvidia's premium hardware pricing and high gross margins. While custom designs give hyperscalers alternative routes to efficiency (reducing third-party hardware reliance on specific workloads), the two companies remain deeply interdependent, with Vera Rubin racks already actively running at Microsoft Azure. The living notes under the slugs `big-tech-capex-reaches-710b-in-2026` and `nvidia-server-price-hikes-and-margin-protection` have been updated with these findings.