Cycle on Jul 17, 2026

I am starting the research cycle by listing the current threads and watches for this topic, which will help identify any existing questions or catalysts we need to address.

I will call `market_view` for Nvidia (ticker: NVDA), TSMC (ticker: TSM), and the semiconductor sector to pull the latest quantitative data and market context. This will ground our assessment of whether the AI capex story remains intact.

I will also call `market_view` for TSMC (TSM) to check their latest earnings and revenue growth, which was mentioned in the news summary as having "record earnings" but experiencing a chip selloff. This will give us a strong leading indicator of AI hardware demand.

Now let's search for the latest news on TSMC's July 16, 2026 earnings report and guidance. Since today is July 17, 2026, TSMC's earnings were released yesterday (July 16, 2026). This is an incredibly fresh and vital data point for the AI capex story. I will use `web_search` with a focus on TSMC's Q2 2026 results and forward guidance.

This is massive news! TSMC has raised its 2026 capital expenditure guidance to $60 billion–$64 billion, up from $52 billion–$56 billion. This is a huge increase (about 15% midpoint hike, or around $8 billion to $10 billion addition). Let's fetch the full transcript of TSMC's Q2 2026 earnings call, or some of the definitive articles that detail why they raised capex, what their revenue outlook is, and how this relates to Nvidia's supply. First, I will fetch `https://investor.tsmc.com/chinese/encrypt/files/encrypt_file/reports/2026-07/57b65edbfe6e480e74abe202be983ecbde79e934/TSMC%202Q26%20Transcript.pdf` (Wait, this is in Chinese or requires encryption? Let's check the English transcript or Yahoo Finance's transcript first). Let's check the Seeking Alpha / Investing.com or Yahoo Finance URLs that have the transcript or analysis. Let's do a search to get a clean English transcript or detailed article from Seeking Alpha or similar. I can search `TSMC Q2 2026 earnings call transcript seeking alpha` or similar, or fetch the URL from the search results. Let's look at the Seeking Alpha article: `"Nvidia: The TSMC Earnings Release Was A Warning Shot (NASDAQ:NVDA)"` which was published 19 hours ago. Let's fetch that page or do a search for the exact URL to fetch. Let's fetch the actual URL of that article. But wait, I don't have the exact URL from the `/goto?url=...` yet, so let's fetch it by searching for its title to get the clean URL or fetch the redirect. Wait, let's search for the Seeking Alpha article title: `Nvidia: The TSMC Earnings Release Was A Warning Shot` or `TSMC Q2 2026 Results: Beat & Raise Guidance`. Let's do a search to find direct URLs of these articles.

Let's fetch `https://seekingalpha.com/article/4922631-nvidia-the-tsmc-earnings-release-was-a-warning-shot` to see why the TSMC earnings release was considered a "warning shot" for Nvidia and what it reveals about the AI capex story. It might discuss TSMC's capex raise, customer demand, or potential risks. Let's fetch it.

Ah, the article content is a bit brief in its markdown preview, so let's check some other sources on the TSMC Q2 earnings to get the concrete numbers, quotes, and details. Let's do a search for the TSMC Q2 earnings call transcript or a detailed analysis of the call. I will search for `"TSMC" "60" "64" "capex" "July 16" 2026` or `"TSMC" "Arizona" "100"` to get the specific quotes and details. Let's run a search.

Let's fetch `https://www.investing.com/news/transcripts/earnings-call-transcript-tsmc-lifts-2026-outlook-as-ai-demand-stays-hot-in-q2-2026-93CH-4794777` or `https://www.benzinga.com/news/26/07/60504620/taiwan-semiconductor-reports-q2-2026-results-full-earnings-call-transcript` to read the actual words spoken by TSMC's management during their earnings call yesterday (July 16, 2026). This will give us direct, high-quality quotes on AI demand, capex projections, custom silicon, and Rubin/Blackwell. Let's fetch the Benzinga transcript first as it is likely to have the full text.

This is an incredible, first-hand source transcript! We have exact verbatim quotes from C.C. Wei (Chairman and CEO of TSMC) and Wendell Huang (CFO) from their earnings call on July 16, 2026. Let's look at the key details from yesterday's call: 1. **Capex Increase**: TSMC raised its full-year 2026 capital budget to **$60 billion to $64 billion** (up from $52 billion to $56 billion previously). This is a massive increase, driven by "continued strong structural demand from our customers, including the newly emerging agentic AI market." 2. **Arizona Investment**: TSMC announced an **additional $100 billion investment in Arizona** to build several more logic wafer fabs (for 2-nanometer and below) and advanced packaging fabs. This brings the total Arizona investment to **$265 billion**. 3. **AI Demand**: C.C. Wei states that "AI-related demand continues to be extremely robust" and "our conviction in the multi-year AI megatrend remains very high." He notes that the AI CAGR is "stronger than what we said before" (which was mid-to-high 50s). He also states that the supply-demand gap is "really big." 4. **Agentic AI and CPUs**: C.C. Wei explains that "The emergence of agentic AI is leading to a resurgence in the role of CPUs in AI data centers, which drive more silicon demand in addition to AI accelerators." This helps explain why CPU customers are adding demand. 5. **No Inventory Overhang**: TSMC is actively checking "AI data center's progress. The building, the location, the demand, the racks. We are checking all that to make sure that TSMC's chips will not be put in inventory." 6. **A14 Technology**: TSMC announced that its next-generation A14 (1.4nm equivalent node) technology is on track, with pre-production starting in 2027 and volume production in 2028. They also introduced A13 and A12 for 2029 volume production. Let's check the watches we listed earlier: 1. `Hyperscaler Capex Guidance Track (July 2026 Earnings)` (`019eb0cd-20a0-71c6-b91e-92d528f6792e`): Ongoing watch. The fire date is July 31, 2026. This is still active as hyperscalers (MSFT, META, GOOGL, AMZN) will report in late July. 2. `Nvidia Gross Margin Track` (`019ec040-d231-7dc2-9387-902ca1d82c1d`): Ongoing watch. Our current market view shows Nvidia's gross margin is 74.1%. 3. `TSMC Monthly Revenue YoY Growth Rate Track` (`019ecfb3-79b3-76db-aeec-0bc8d0635c41`): Ongoing watch. 4. `Nvidia Q2 FY2027 Revenue Guidance Cross` (`019edf29-c3f7-7561-bccb-ef09f3a7fb09`): One-shot watch. Firing date is August 25, 2026. 5. `Nvidia Q2 FY2027 Earnings - Rubin Ultra / Kyber NVL144 Timeline Update` (`019f4b49-2c52-7a54-9f80-a4fb82cea464`): One-shot watch. Firing date is August 25, 2026. Wait, the watch `TSMC Monthly Revenue YoY Growth Rate Track` (`019ecfb3-79b3-76db-aeec-0bc8d0635c41`) is ongoing. Let's check if there are any other watches. Wait! Let's check TSMC's revenue growth. In the market view, TSM's TTM revenue growth is +35.1% YoY. In yesterday's earnings call, TSMC raised its full-year 2026 revenue growth outlook to "slightly above 40% year-over-year in US dollar terms" (up from previous guidance). And they reported Q2 revenue grew 36% YoY (or 37% at the midpoint of Q3 guidance). This means the monthly revenue YoY growth is well above 20%. Let's write a note updating the Nvidia AI capex story. The previous note was: `Nvidia's AI Capex Story Remains Robustly Intact Amid Soaring Projections and Kyber Rack Delay Debates` (slug: `nvidia-ai-capex-cycle-sustainability`). Let's reuse this slug to update this note in-place with the fresh TSMC earnings data from July 16, 2026! Wait, let's verify if there are other public players or news that we should search for to see if there are any other recent updates on Nvidia's capex story. Let's search for "Nvidia capex July 2026" or "hyperscaler capex July 2026" to see if any other news has come out in the last week. Let's do a search.

This is a goldmine of very recent articles! Let's examine some of these: 1. `"Among AI crowd, some investors position for slower hyperscaler spending growth"` (Reuters/WTVB, July 17, 2026): - "But that spending now looks set to slow, with UBS estimating hyperscalers' capex will rise 76% this year to $673 billion, but will increase by only 25% next year to $842 billion." - Let's fetch this Reuters article to see what it says about the capex story and the broader market's position. 2. `"Top Wall Street Analyst Says AI Spending Is Delivering Real Returns: 'When We Build a Data Center, It’s Already Pre-Sold'"` (AOL, July 16, 2026 / 247wallst, July 10, 2026): - This discusses how Microsoft, Amazon, and Alphabet claim that data centers are "pre-sold" or already matched to customer commitments. - Let's fetch this article too to understand the return on investment (ROI) debate. 3. `"Amazon Is Spending 200M In AI This Year: Is that Good News For Investors?"` (247wallst, July 14, 2026): - Wait, "Amazon Is Spending 200M" or maybe "$20 billion" or "$200 billion"? Let's fetch the actual article to check the numbers. Let's fetch `https://wtvbam.com/2026/07/17/analysis-among-ai-crowd-some-investors-position-for-slower-hyperscaler-spending-growth/` first to see the UBS estimates and the shifting expectations.

This is another extremely rich and relevant article! It discusses the critical debate: - On one side, some investors are positioning for slower hyperscaler spending growth (e.g. UBS estimates hyperscaler capex will rise 76% this year to **$673 billion** but only increase by 25% next year to **$842 billion** and just 6% in 2028). - On the other side, TSMC just reported record profits and hiked its own 2026 capex guidance by ~$10 billion to **$60-$64 billion** (up from $52-$56 billion), citing "continued strong structural demand... including the newly emerging agentic AI market." - There is a notable state-level regulatory development: "New York on Tuesday became the first U.S. state to halt construction of large new data centers, imposing a one-year moratorium as concerns grow that the facilities driving the AI boom are raising power costs, straining water supplies and burdening local communities." This is a key tripwire/metric that we should monitor! Let's fetch the other article: `https://www.aol.com/articles/top-wall-street-analyst-says-190912183.html` to read about the "pre-sold" data centers comment and the ROI debate. This is very relevant to our `enterprise-ai-roi-gap-revenue-reckoning` slug or our `nvidia-ai-capex-cycle-sustainability` slug. Let's fetch it.

This is incredibly deep data! We have specific, named sources and extremely clear numbers. Let's analyze what we have: 1. **TSMC's Blockbuster Q2 2026 Earnings & Raised Guidance (July 16, 2026)**: - Raised full-year 2026 capital budget to **$60-$64 billion** (up from $52-$56 billion). - Raised full-year 2026 revenue growth outlook to **slightly above 40% YoY** in USD terms. - Announced an **additional $100 billion investment in Arizona** (bringing total to $265 billion) to build several more 2nm and below logic & advanced packaging fabs. - C.C. Wei (CEO) noted that "AI-related demand continues to be extremely robust" and "the gap [between supply and demand] is really big." - Noted a resurgence in CPUs in AI data centers due to **agentic AI**, which drives CPU demand in addition to accelerators. He notes: "The emergence of agentic AI is leading to a resurgence in the role of CPUs in AI data centers, which drive more silicon demand in addition to AI accelerators. We believe this is positive for TSMC as no matter what CPU approach is taken, whether it's x86, ARM-based, or RISC-V architecture, they are almost all TSMC customers." - TSMC is actively checking "AI data center's progress. The building, the location, the demand, the racks. We are checking all that to make sure that TSMC's chips will not be put in inventory." 2. **The Hyperscaler ROI & Capex Debate (July 2026)**: - Gil Luria (D.A. Davidson) points out that hyperscalers say their data center investments are already generating attractive returns because the capacity is "pre-sold" before construction is complete. - OpenAI and Anthropic's combined run rate has surged from under **$20 billion to over $75 billion** in just six months. - Hyperscaler capex figures are staggering: - **Microsoft**: Q3 FY26 capex was **$30.88 billion** (up 84.39% YoY), while operating margin held at 46.3% and AI business reached a $37 billion run rate (up 123% YoY). Commercial backlog (RPO) is **$627 billion**. - **Amazon**: Q1 2026 capex was **$44.203 billion**, with full-year 2026 capex guided at roughly **$200 billion**. AWS grew 28% to $37.587 billion. Custom chips (Trainium) topped a $20 billion revenue run rate. Anthropic committed to up to 5 GW of Trainium capacity and OpenAI to roughly 2 GW starting in 2027. - **Alphabet**: Q1 2026 capex was **$35.67 billion**, and 2026 capex is guided at **$175-$185 billion**. Google Cloud grew 63% to $20.03 billion, with backlog nearly doubling QoQ to over **$460 billion**. Free cash flow dropped 47% YoY to $10.12 billion. - UBS estimates that hyperscalers' capex will rise 76% this year to **$673 billion** (up from prior projections), but will increase by only 25% next year to **$842 billion** and just 6% in 2028. - On the financing side, Torsten Slok (Apollo) notes that cover ratios for Big Tech corporate bonds have fallen to below 2x in July 2026 from nearly 5x in February 2026, suggesting that capital market pressures may start to constrain spending growth. 3. **Regulatory Tripwire (July 14, 2026)**: - "New York on Tuesday [July 14, 2026] became the first U.S. state to halt construction of large new data centers, imposing a one-year moratorium as concerns grow that the facilities driving the AI boom are raising power costs, straining water supplies and burdening local communities." This is a major physical constraint tripwire! Let's check our existing findings and slugs: - `nvidia-ai-capex-cycle-sustainability` (Nvidia's AI Capex Story Remains Robustly Intact Amid Soaring Projections and Kyber Rack Delay Debates): This is the perfect slug to update with TSMC's blockbuster Q2 earnings, raised capex, and the agentic AI CPU resurgence narrative. - `enterprise-ai-roi-gap-revenue-reckoning` (The Hyperscaler Capital Crossover: AI Capex Outstrips Cash Flow, Triggering Historic Debt and Equity Waves): This is the perfect slug to update with the new hyperscaler capex numbers (Amazon $200B, Alphabet $175B-$185B, Microsoft $30.88B/quarter), the "pre-sold" data center debate, OpenAI/Anthropic's $75B run rate, and the New York moratorium on large data centers. Let's read our current notes first to see what's in them, using `read_notes` for topic `019e95a0-30d1-7d08-ae5b-10ac242b9d5c`.

This is an excellent set of notes! We have very clear, specific notes. Let's update: 1. `nvidia-ai-capex-cycle-sustainability`: We will update this note to include the TSMC Q2 earnings and raised capex. 2. `enterprise-ai-roi-gap-revenue-reckoning`: We will update this note to include the new hyperscaler capex figures, the "pre-sold" data center debate, OpenAI/Anthropic's $75B run rate, and the New York moratorium on large data centers. Let's read the current contents of `nvidia-ai-capex-cycle-sustainability` and `enterprise-ai-roi-gap-revenue-reckoning` to see what's in them before we overwrite them. Oh wait, I don't need to read them if I can just write the updated markdown, but wait, the guidelines say: "Before modifying a file, read it first." Wait, `read_notes` already gave us a preview, but wait, do we have a tool to read the full note body? No, we don't have a direct `read_note_body` tool, but wait, `read_notes` returns the slug, title, preview, updated_at, and source_count. We don't have a way to read the full body of a note except if we overwrite it or if we search. Since we are updating the notes with the latest evidence from yesterday and today (July 16-17, 2026), we can write the updated comprehensive notes. Let's carefully construct the body for `nvidia-ai-capex-cycle-sustainability`. The note should focus on: - How Nvidia's AI capex story remains robustly intact, strongly validated by TSMC's blockbuster Q2 2026 earnings and its massive capital budget hike to $60-$64 billion (up 15% at the midpoint from $52-$56 billion). - C.C. Wei's comments on the "really big" supply-demand gap and the "insatiable" demand for leading-edge silicon. - The acceleration of agentic AI leading to a resurgence of CPUs in AI data centers, which benefits TSMC and Nvidia (especially with their Grace and upcoming Vera CPUs). - TSMC's additional $100 billion investment in Arizona to build several more 2nm and below logic and advanced packaging fabs. - TSMC's active monitoring of AI data center progress to ensure no inventory buildup. - Real market data from Nvidia's `/markets/NVDA/2026/07/17` and TSMC's `/markets/TSM/2026/07/11` (including TSM's record Q2 results on July 16). Let's write the note for `nvidia-ai-capex-cycle-sustainability` now. Let's make sure it contains direct verbatim quotes and links to the sources. Let's double check if we have the correct links: - TSMC Earnings Call Transcript: `https://www.benzinga.com/news/26/07/60504620/taiwan-semiconductor-reports-q2-2026-results-full-earnings-call-transcript` - Seeking Alpha analysis: `https://seekingalpha.com/article/4922631-nvidia-the-tsmc-earnings-release-was-a-warning-shot` - Market views: `/markets/NVDA/2026/07/17` and `/markets/TSM/2026/07/11` Let's draft the note body for `nvidia-ai-capex-cycle-sustainability`. No YAML frontmatter. First heading should be `# Nvidia's AI Capex Story Remains Robustly Intact Amid TSMC's Historic Capex Hike and Agentic AI CPU Resurgence`. Let's write the note.

Now let's update the second note, `enterprise-ai-roi-gap-revenue-reckoning` (The Hyperscaler Capital Crossover: AI Capex Outstrips Cash Flow, Triggering Historic Debt and Equity Waves). Let's review the new data points we have for this note: 1. **Staggering Hyperscaler Capex Figures**: - **Amazon**: Guided full-year 2026 capex is roughly **$200 billion** (Q1 capex was **$44.203 billion**). Custom chips (Trainium) topped a $20 billion run rate. Anthropic committed to up to 5 GW of Trainium capacity and OpenAI to roughly 2 GW starting in 2027. - **Alphabet**: Guided full-year 2026 capex is **$175-$185 billion** (Q1 capex was **$35.67 billion**). Google Cloud backlog nearly doubled QoQ to over **$460 billion**. Free cash flow fell 47% YoY to $10.12 billion. - **Microsoft**: Q3 FY26 capex was **$30.88 billion** (up 84.39% YoY). Commercial backlog (RPO) is **$627 billion**. 2. **The ROI Debate & "Pre-Sold" Data Centers**: - Gil Luria (D.A. Davidson) highlights the disconnect: investors see cratering free cash flow margins, but hyperscalers argue that data centers are "pre-sold" before construction is complete. - OpenAI and Anthropic's combined run rate has surged from under **$20 billion to over $75 billion** in just six months, providing concrete evidence of real demand. 3. **Slowing Capex Growth Projections (UBS)**: - UBS estimates hyperscaler capex will rise 76% this year to **$673 billion**, but spending growth looks set to slow to a 25% increase next year to **$842 billion** and just 6% in 2028. - This moderating trajectory creates a potential mismatch with the lofty revenue expectations priced into chipmaker stocks. 4. **Financing Pressures**: - Torsten Slok (Apollo) reports that corporate bond cover ratios for Big Tech have fallen to below 2x in July 2026 from nearly 5x in February 2026, indicating that capital markets may eventually constrain unlimited spending. 5. **Physical & Regulatory Constraints (New York Moratorium)**: - On July 14, 2026, New York became the first U.S. state to halt construction of large new data centers, imposing a one-year moratorium due to concerns about power costs, water supply, and local grid strain. This is a critical physical bottleneck. Let's write this updated note under the stable slug `enterprise-ai-roi-gap-revenue-reckoning`. I will make sure it is fully grounded in the sources, contains verbatim quotes, and is structured perfectly. No YAML frontmatter. First heading is `# The Hyperscaler Capital Crossover: Presold Backlogs, $200B Amazon Capex, and the First State-Level Data Center Moratorium`. Let's call `write_note`.

We have successfully written and updated both key notes! Let's review the active threads and watches to see if we can resolve any or register new ones. 1. **Hyperscaler Capex Guidance Track (July 2026 Earnings)** (`019eb0cd-20a0-71c6-b91e-92d528f6792e`): This watch tracks whether major hyperscalers (Microsoft, Meta, Alphabet, Amazon) update their FY2026/FY2027 capex outlook during their July 2026 earnings calls. This watch has a fire date of July 31, 2026, which is in the future (today is July 17, 2026). So it should remain open. 2. **First State-Level Data Center Moratorium**: This is a major new physical/regulatory tripwire that we should track. We can record a new watch or thread if we want to follow other states adopting similar moratoriums. Let's register a new watch for additional state-level data center construction halts or federal grid regulations. Wait, let's see if we have any other threads to resolve. Let's look at the open threads we retrieved from `list_threads`: - `019f650b-f8e3-7329-abca-aadef5844003`: `AMD Instinct MI500 Launch and Rubin Ultra Comp` (dated: 2027-06-30) - `019f650b-eeac-73e7-8bdf-58497b3b458f`: `TSMC CoPoS Packaging Pilot Line Launch` (dated: 2026-12-31) - `019f5599-dcfb-788b-a861-4a9c86598587`: `Low-Cost Chinese Model Breakthrough` (one-shot) - `019f4b49-2c52-7a54-9f80-a4fb82cea464`: `Nvidia Q2 FY2027 Earnings - Rubin Ultra / Kyber NVL144 Timeline Update` (dated: 2026-08-25) - `019f2c65-b0bd-7794-add6-7b6b15ef3fff`: `First Commercial Shipment of Nvidia Rubin GPUs` (dated: 2026-12-31) - `019f17cc-fae4-7cda-927b-fb7b20231712`: `Official Launch of ICE GPU Compute Futures Contracts` (one-shot) - `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` (one-shot) - `019ee44e-b944-74a7-bc73-b87fb1bf1ac5`: `Aggregate Hyperscaler Free Cash Flow Crossover (Q3 2026)` (dated: 2026-10-31) - `019edf29-c3f7-7561-bccb-ef09f3a7fb09`: `Nvidia Q2 FY2027 Revenue Guidance Cross` (dated: 2026-08-25) - `019ed4d9-3d3f-7390-b1e8-77f8753e9989`: `Kazakhstan Data Center Valley Phase 1 Groundbreaking` (dated: 2027-01-01) - `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)` (ongoing, dated: 2026-07-31) - `019eaba5-b004-719f-8669-34a8aae85368`: `Nvidia Standalone Vera CPU Revenue Track` (ongoing) - `019e9c33-1f68-7b87-b126-0aa020f67c7f`: `Apple Revamped Siri Launch with Blackwell Workloads` (dated: 2026-09-15) Wait, can we record an event for any of these? - `TSMC Monthly Revenue YoY Growth Rate Track` (`019ecfb3-79b3-76db-aeec-0bc8d0635c41`): TSMC's Q2 revenue was $40.2 billion (+36% YoY) and Q3 revenue guidance is $44.6B-$45.8B (representing ~37% YoY growth). So the YoY growth rate remains strongly above 20%. Let's record an ongoing event for this to show it's tracking well above the 20% threshold. Wait, the trigger is: "Fires if TSMC's monthly revenue year-over-year growth drops below 20%." Since it has not dropped below 20% (it's actually at 35%-40%), the watch has not been triggered to fire. It is an ongoing watch that should keep tracking without firing an event unless the trigger is met. Let's register a new watch for the data center moratoriums. We saw that New York State just imposed a one-year moratorium on large data center construction on July 14, 2026. This is a massive physical tripwire. Let's record a watch to track if other states (like Virginia, Georgia, or Ohio) or federal agencies impose similar moratoriums or restrictive grid capacity regulations on AI data centers. Let's call `record_watch`.

Excellent! Now we have recorded the watch. Let's make sure our memories are updated if necessary. Wait, let's look at the Guidelines: - Long-term memory: `/var/lib/hey-lefty/notes/019e89b4-58b3-7e3b-a0c2-6a68bd3242db/019e95a0-30d1-7d08-ae5b-10ac242b9d5c/memory/MEMORY.md` (write important facts here) - History log: `/var/lib/hey-lefty/notes/019e89b4-58b3-7e3b-a0c2-6a68bd3242db/019e95a0-30d1-7d08-ae5b-10ac242b9d5c/memory/HISTORY.md` Let's see if we should write to the memory or history log. No, wait, since we are running a standard research cycle, we should focus on the research tools and call `finish_cycle` at the end of the cycle. Let's do a quick double-check of our written notes to ensure everything is correct and matches our guidelines. Our notes: 1. `nvidia-ai-capex-cycle-sustainability`: Up-to-date with TSMC's Q2 earnings, capex hike to $60B-$64B, and agentic AI. 2. `enterprise-ai-roi-gap-revenue-reckoning`: Up-to-date with hyperscaler capex figures, Gil Luria's "pre-sold" data center debate, OpenAI/Anthropic's $75B run rate, and New York's moratorium. Both notes are fully grounded in the actual sources fetched during this cycle (the Benzinga transcript, Seeking Alpha article, and Reuters analysis), carry verbatim quotes with their links, and use wikilinks to link other notes (e.g., `[[hyperscaler-capex-surge-2026]]` in `enterprise-ai-roi-gap-revenue-reckoning`). We are ready to finish the cycle. I will call `finish_cycle`.

I have completed this research cycle on the sustainability of Nvidia’s AI capex story. Below is a detailed summary of the findings and updates made during this cycle. ### Executive Summary The structural demand driving **Nvidia’s (NVDA)** artificial intelligence dominance and the broader AI hardware capex cycle has been aggressively validated as of mid-July 2026. This is strongly anchored in two monumental developments: 1. **TSMC’s Historic Q2 2026 Earnings & Raised Capex (July 16, 2026):** TSMC raised its full-year 2026 capital budget to **$60 billion–$64 billion** (up ~15% at the midpoint from $52 billion–$56 billion) and announced an additional **$100 billion investment in Arizona** (bringing its total Arizona commitment to **$265 billion**) to build several more 2nm and advanced packaging fabs. 2. **The "Pre-Sold" Data Center Backlog Defense:** Despite widening concerns over hyperscaler free cash flow compression, cloud giants are defending their historic spending—such as **Amazon’s $200 billion** and **Alphabet’s $175–$185 billion** guided 2026 budgets—by pointing to massive, pre-sold customer backlogs and the explosive growth of customers like **OpenAI and Anthropic**, whose combined run rate surged from under **$20 billion to over $75 billion** in just six months. --- ### Key Findings & Note Updates #### 1. Nvidia's AI Capex Story Remains Robustly Intact * **Slug updated:** `nvidia-ai-capex-cycle-sustainability` * **The Supply-Demand Gap:** TSMC CEO Dr. C.C. Wei described the gap between supply and demand as "really big," reinforcing that physical hardware bottlenecks (rather than demand softening) remain the primary constraint. * **Inventory Security:** To appease fears of a "double-ordering" bubble, TSMC confirmed it is actively monitoring physical data center deployments and rack installations to ensure that chips are going directly into service rather than sitting in inventory: > *"We actually are checking the AI data center's progress. The building, the location, the demand, the racks. We are checking all that to make sure that TSMC's chips will not be put in inventory."* * **The Agentic AI Resurgence:** The emergence of complex agentic workflows has triggered an unexpected resurgence in the role of high-performance CPUs in AI data centers to handle orchestration and branching logic alongside accelerators. This system-level shift directly plays into Nvidia's Grace/Vera CPU lineup and full-system architecture. #### 2. The Hyperscaler Capital Crossover & Physical Limits * **Slug updated:** `enterprise-ai-roi-gap-revenue-reckoning` * **Pre-Sold Backlogs:** Hyperscalers are protecting their operating margins by cutting costs elsewhere and pointing to massive, pre-signed customer agreements. Microsoft boasts a commercial remaining performance obligation (RPO) of **$627 billion**, while Alphabet's cloud backlog has nearly doubled quarter-on-quarter to over **$460 billion**. * **Physical Tripwire (New York's Historic Moratorium):** Physical constraints are transitioning from theoretical risks to hard regulatory realities. On **July 14, 2026**, New York became the first U.S. state to halt construction of large new data centers, imposing a **one-year moratorium** due to concerns over local power costs, water supplies, and grid strain. This regulatory freeze highlights that power availability has become the ultimate speed limit for the AI capex boom, forcing hyperscalers to seek alternative private-power JVs like `[[hyperscaler-capex-surge-2026]]`. --- ### Future Outlook and Registered Watches To ensure we stay ahead of this evolving narrative, we registered a new forward-looking watch: * **State-Level AI Data Center Moratorium Expansion Track (`019f6f55-61e5-71a9-898a-2a6088793817`):** This tracks whether other major data center states (such as Virginia, Georgia, Texas, or Ohio) follow New York's lead in imposing construction freezes or grid capacity restrictions on AI infrastructure.