Cycle on Aug 20, 2026

I will begin by listing the open threads on this topic to understand what has been tracked or requested previously.

I will fetch the current public-market view for Nvidia Corporation (NVDA) to see if there is any fresh earnings data or news, especially since today (August 20, 2026) is the scheduled fire date for Nvidia's Q2 FY2027 earnings release.

The market view lists quarterly earnings up to April 30, 2026. Since the current date is August 20, 2026, let's search the web to see if Nvidia has released its Q2 FY2027 (ended July 2026) results yet, or if it is scheduled for later today or next week. Let's do a search on Nvidia Q2 FY2027 earnings date or results.

Ah, the earnings release is scheduled for August 26, 2026! Today is August 20, 2026, so the earnings release has not happened yet. It is in 6 days. Let's make sure we note this. Wait, let's look at the topic brief: "Is Nvidia's AI capex story still intact?" Let's see what has been happening in the market recently regarding AI capex, Nvidia's demand, hyperscalers, and any recent news. Let's do some searches on AI capex trends in Q2/Q3 2026, hyperscaler capex updates (Microsoft, Alphabet, Meta, Amazon, Oracle), and Nvidia's supply chain or backlog. Let's search for "AI capex" "Nvidia" "2026" and see what recent articles or discussions say. Let's do a search.

Let's fetch several of these highly relevant recent articles to understand the exact mechanics of what's happening. Particularly: 1. `https://www.rexshares.com/nvidia-earnings/` - "NVIDIA Earnings Q2 FY27: Revenue, Data Center, AI Capex" (published very recently, talks about Q2 2026 hyperscaler capex and relationship to Nvidia) 2. `https://www.cnbc.com/2026/08/14/ai-infrastructure-debt-leverage-risks.html` - "AI’s infrastructure boom is getting more leveraged — and harder to track" (August 14, 2026, very fresh) 3. `https://www.uncoveralpha.com/p/nvidia-and-the-hyperscalers-high` - "Nvidia and the Hyperscalers High Stakes Poker Game" (August 13, 2026, details Nvidia's counter-move on July 1, 2026, and "AI Compute Partnership") 4. `https://www.fool.com/investing/2026/08/13/nvidia-stock-investors-good-news-from-wall-street/` - "Nvidia Stock Investors Just Got Good News From Wall Street" (August 13, 2026) Let's fetch these URLs. I'll start with `https://www.uncoveralpha.com/p/nvidia-and-the-hyperscalers-high` to understand the "AI Compute Partnership" and the "High Stakes Poker Game" between Nvidia and the hyperscalers.

This is an incredibly rich and detailed article with deep insights on Nvidia's "AI Compute Partnership" (formalized July 1, 2026) and its massive $500B Wall Street compute financing platform announced on Monday (which would be August 10, 2026, since the article is dated August 13, 2026). Let's search for the CNBC article from August 14, 2026, to see what details it has on the debt, leverage, and the $500B Wall Street deal. Let's fetch `https://www.cnbc.com/2026/08/14/ai-infrastructure-debt-leverage-risks.html` or search for it. Let's do a search first to get the exact URL or details. Wait, the URL is already in the search results: `https://www.cnbc.com/2026/08/14/ai-infrastructure-debt-leverage-risks.html`. Let's fetch it.

Wow, this is incredibly valuable. Let's look at the details. - "Nvidia’s plan to develop platforms for AI infrastructure in partnership with Apollo, Blackstone, BlackRock, Brookfield, KKR and Goldman Sachs could involve private asset-like structures and asset-based financing. Nvidia CEO Jensen Huang told CNBC Monday that Nvidia’s chips are now an “investable infrastructure asset.”" (This was on Monday, August 10, 2026). - "Some tech giants are using joint ventures and other leasing vehicles to borrow money for AI data center spending — without the debt appearing on their balance sheets until the leases begin." - "Goldman Sachs analysts estimated that hyperscalers have combined lease commitments for data centers, R&D facilities, offices and equipment of $1.5 trillion, up from about $200 billion five years ago. This includes about $1 trillion of “uncommenced” lease commitments, which are not yet shown in financial statements but will result in future payments." - "Lotfi Karoui, multi-asset credit strategist at PIMCO, said the AI capex cycle is, adjusted for inflation, on track to be the largest investment cycle since the 19th-century railway construction." - "consensus forecasts that hyperscaler capital spending alone will surpass $1 trillion per year from 2027 onward, “with no clear signs of moderation.”" - "Situational Awareness’ collapse showed how vulnerable crowded, leveraged AI trades can be to the sharp sell-offs and rallies the sector has seen recently." "The fund was unable to meet a series of margin calls from lenders after its heavily concentrated portfolio — which included names such as SK Hynix and CoreWeave — suffered during a recent tech sell-off and its assets fell from $45 billion to about $10 billion. Ken Griffin’s larger, multi-strategy hedge fund Citadel later stepped in to buy Situational Awareness’ publicly listed positions at a discount. SK Hynix and CoreWeave have since rallied." Let's do a search on "Situational Awareness" hedge fund collapse or "Situational Awareness" SK Hynix CoreWeave to see if there are more details. This is a very interesting development under the theme of "Nvidia's AI Moat Expands to $500 Billion Wall Street Private Credit Partnership Amid Circular Financing Concerns" (slug: `nvidia-record-q1-results-and-trillion-dollar-order-book`) or "Off-Balance-Sheet Joint Ventures and the New AI Infrastructure Financing Model" (slug: `off-balance-sheet-ai-infrastructure-financing-models`). Let's search for "Situational Awareness" hedge fund.

This is amazing! Let's review what we have: 1. Leopold Aschenbrenner's AI-focused hedge fund **Situational Awareness LP** nearly imploded in July 2026. It suffered heavy losses (losing 67% in July 2026, dropping from $45 billion in assets/NAV on July 1 to around $10 billion), leading to a massive margin call. 2. The fund was heavily leveraged (up to 4x / 400%). Its prime brokers (Bank of America, Goldman Sachs, JPMorgan) issued margin calls. 3. On July 30, 2026, Ken Griffin's **Citadel** (specifically its Wellington fund) stepped in to buy the bulk of Situational Awareness's public equity portfolio (valued at approximately $16 billion) in a quick-fire block trade over less than 24 hours at a ~10% discount. Citadel competed with Jane Street and Millennium. 4. The positions included highly concentrated bets on AI hardware and semiconductor names like **SK Hynix**, **Nebius**, and **Micron**. 5. This swoop by Citadel helped stem a broader $3 trillion AI stock rout, and many of the AI-related stocks have since rebounded (gaining nearly 30% since the liquidation). Citadel's Wellington fund gained 5.9% in July, bringing its year-to-date return to 12% as of early August. 6. Situational Awareness remains active with about $10 billion in assets, largely consisting of its private holdings, including its stake in Anthropic. This is a fantastic development that directly relates to the broader AI infrastructure financing model and credit/leverage risks of the AI boom, which is highly relevant to Nvidia's AI capex story. Wait! Let's check the other key elements of the Nvidia capex story: 1. **Hyperscaler capex trends**: The top hyperscalers (Amazon, Google, Microsoft, Meta) reported Q2 2026 capex. Let's look at the numbers. According to the Rexshares article: "Q2 2026 combined hyperscaler capex: $166.0B (+87% YoY, +27% QoQ)." Also, "consensus forecasts that hyperscaler capital spending alone will surpass $1 trillion per year from 2027 onward, with no clear signs of moderation" (PIMCO's Lotfi Karoui). Wait, the Fool article notes: "consensus estimate now says their capex spending will total $733 billion in 2026" (Wall Street raised 2026 capex forecast). "consensus estimate currently says capex spending among the top five hyperscalers will grow 28% to $939 billion next year [2027]." 2. **Nvidia's counter-moves / AI Compute Partnership**: On July 1, 2026, Nvidia formalized the "AI Compute Partnership" model where Nvidia acts as a financial backstop, agreeing to rent back unused GPUs at a fixed rate so lenders will finance neocloud buildouts, and in exchange, Nvidia takes a share of the neocloud's cloud revenue. First partners: Firmus (up to 170,000 GPUs in Batam, Indonesia) and Sharon AI (40,000 GB300s in Australia). 3. **Nvidia's $500B Wall Street Compute Financing Platform**: On Monday, August 10, 2026, Nvidia announced a partnership with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to raise more than $500B in third-party capital for AI infrastructure, with compute itself used as collateral in SPVs. Jensen Huang told CNBC that Nvidia's chips are now an "investable infrastructure asset." Nvidia retains the option to backstop up to $125B (25%) of the deals. 4. **Hyperscaler tensions and ASIC programs**: The hyperscalers are trying to commoditize Nvidia's layer by designing their own custom silicon (ASICs) for inference: - Google produces +3M TPUs annually, Broadcom deal for 3.5GW next-gen TPU capacity starting 2027. - Amazon's chip business (Trainium + Graviton) runs at a $25B annualized revenue rate, growing triple digits. Anthropic is scaling Trainium chips from 500k to 1M. - Microsoft is unveiling Maia 300 in September 2026, negotiating for 300,000+ chips for delivery in 2027 with TSMC, aiming for over 1M units. - Hyperscalers are framing capex as "options" on Q2 calls, separating long-lead capex (land, power, shells) from short-lead capex (chips). They buy chips only months before data centers go live. Let's do some more search on "Nvidia" "$500B" "Apollo" "BlackRock" to see if we can find more details or official announcements about this massive $500 billion financing platform. This is a very big deal that directly answers "Is Nvidia's AI capex story still intact?" Let's search.

This is amazing. Let's fetch the Global Data Center Hub article: `https://www.globaldatacenterhub.com/p/nvidias-500b-ai-infrastructure-financing` to understand the exact structure of this $500B financing deal (which was announced on August 10, 2026). It discusses "the 25 percent residual support, the unsigned agreements, and the Goldman mandate". This will be highly specific and precise, which is exactly what we need.

This is an incredibly detailed and precise analysis of the $500B Wall Street Compute Financing Platform. Let's review what we have: 1. **The $500B Wall Street Compute Financing Platform**: - Announced on **August 10, 2026**, Nvidia signed Memorandums of Understanding (MOUs) with six financial partners: **Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR**, targeting over $500 billion of third-party capital over time. - The platforms are designed to mobilize third-party capital for Nvidia's customers (frontier AI labs, enterprises, and AI clouds) to build data centers and acquire Nvidia hardware. - **Residual-Value Support**: Nvidia may provide a residual-value support mechanism for **up to 25% of an opportunity**, assessed project-by-project. This is not a single fund or direct cash commitment from Nvidia, and is still a term-sheet concept until final agreements are executed. - **Goldman Sachs' Dual Role**: Goldman is the only traditional bank in the group, playing both investment (providing junior capital/private credit via its asset management arm) and distribution roles (placing debt with private-credit funds and eventually public markets). - **Credit Market Reactions**: Nvidia's 5-year credit default swaps (CDS) narrowed to 72.11 bps on August 11, 2026, after CEO Jensen Huang clarified the limited scope of the support. Yield spreads on Nvidia's 5.625% 2056 bonds narrowed by 2 bps to 113 bps over Treasuries. - **Ohio Project Revision**: Just four days after the announcement, the Wall Street Journal reported that Nvidia cut its planned guarantee for OpenAI's SB Energy campus in southern Ohio from roughly $250 billion to under $120 billion (limited to the first 5 gigawatts of a 10GW program). - No memorandum has converted into a binding final agreement yet. No partner allocation or first project has been named. Let's check the other threads and watches we can update or resolve! Let's list the open threads and watches that this information touches: 1. `019fee38-1dbd-73a3-8fa2-113b552feff5` (Nvidia $500B Wall Street Compute Financing Platform Capital Deployment) - This is an ongoing watch. The watch description is: "Monitor the official launch, terms, and capital deployment of the first compute financing transaction under Nvidia's $500 billion partnership with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR." - Wait, since the announcement happened on August 10, 2026, we can record an event for this watch! The event is that the partnership was officially announced on August 10, 2026, via MOUs with these six financial partners. But since no binding agreement has been executed and no first project has been named, the watch is ongoing (resolves: false). - Let's check: can we record an event on `019fee38-1dbd-73a3-8fa2-113b552feff5`? Yes! Let's do that. 2. `01a00d1e-7c86-7d8e-9b6d-3a56ead1f70b` (SB Energy IPO and Nvidia $1.5B Anchor Investment) - This is a watch for when SB Energy prices its IPO. The fire date is September 30, 2026. This is still in the future. 3. `01a007f9-8536-7aa4-97a1-5fc02943ef9a` (Hyperscaler SPV and Corporate Bond Spread Widening) - Let's see if we have any updates on hyperscaler debt or credit spreads. - The "Global Data Center Hub" article mentions that Goldman Sachs analysts estimated that hyperscalers have combined lease commitments of $1.5 trillion, including about $1 trillion of "uncommenced" lease commitments. - PIMCO's Lotfi Karoui highlighted that consensus forecasts for hyperscaler capital spending alone will surpass $1 trillion per year from 2027 onward, with "no clear signs of moderation." - The Sage Advisory article mentioned: "Since September, the hyperscalers... have more than doubled their collective U.S. dollar debt... Hyperscaler Debt Deluge: The New Driver of IG Spread Pressure." - Let's do a search on "hyperscaler bond spreads" or "AI infrastructure debt" to see if there is any sign of widening spreads or credit rating downgrades. Let's do a search.

This is incredibly valuable and highly detailed! Let's examine the facts on bond spreads: - Since September 2025, hyperscalers (Microsoft, Meta, Google, Oracle, Amazon, and Nvidia) have more than doubled their collective U.S. dollar debt footprint to over **$360 billion**. - Amazon issued **$25 billion** in debt in July 2026, marking the ninth deal this year larger than $20 billion (with seven of those linked to hyperscalers). - This "debt deluge" has pressured spreads wider. For example, the Amazon offering pushed its existing 30-year bond about **20 basis points wider** as investors demanded additional concession. - According to a Reuters analysis of LSEG data from late July 2026, **78 out of 91 hyperscaler bonds** issued in 2026 were trading at higher yields than at issuance (meaning spreads have widened). - Specifically, the median spread on 2- to 4-year bonds rose to **40 basis points** (from 30 basis points in 2025). - The median spread on 5- to 7-year debt increased to **60 basis points** (from 50 basis points). - The median spread on bonds maturing in more than 20 years rose to **118 basis points** (from 108.5 basis points). - This is a very clear and concrete set of metrics that we should track for our watch on "Hyperscaler SPV and Corporate Bond Spread Widening" (ID: `01a007f9-8536-7aa4-97a1-5fc02943ef9a`). Wait, the watch trigger is: "Fires if hyperscaler or SPV borrowing spreads widen by >=25 bps over the broader investment-grade market, signaling severe credit market repricing of AI debt." - Amazon's 30-year bond spread widened by 20 basis points, and overall spreads rose by 10-15 basis points across various tenors. This is close to the 25 bps threshold but has not quite hit it systemically across the broad investment-grade corporate market in a single quarter yet, though it is showing clear pressure. Let's record an event on this watch since we have a major update! The watch is ongoing, so `resolves: false`. Let's check other threads and watches: 1. `019ff883-8436-7932-a2f9-bc957bd0b14d` (Aggregate Hyperscaler Free Cash Flow Trajectory (2026–2028)) - This is an ongoing watch. We can update it with the latest projections and current situation (deteriorating cash flows, debt issuance, negative cash flows at Google/Amazon, etc.). 2. `019f9bd1-265c-7c53-ac9b-daf15136e97e` (Google Cloud Q3 2026 Operating Margin and Third-Party Capacity Impact) - Scheduled for October 25, 2026. Still in the future. 3. `019f9185-7240-73e6-897f-7a8c1b51e4a8` (Google 110,000-GPU Colossus 2 Lease Commencement) - Scheduled for October 15, 2026. Still in the future. 4. `019f8c5c-970d-7ba8-a4f1-d61379b41103` (Oracle 50,000-GPU AMD Instinct Helios Supercluster Deployment) - Scheduled for Q3 2026. Let's search to see if there are any updates on Oracle's Helios supercluster or AMD Instinct MI450/MI455X. Let's do a search on Oracle Helios supercluster.

This is scheduled for Q3 2026, which is ongoing. No official launch has been reported yet (today is August 20, 2026). Let's keep a watch. Let's check `019ed2f6-f136-7051-a5b0-fa3b26ec02e0` (Nvidia Q2 FY2027 Earnings Release) - This is scheduled for August 26, 2026. The fire date was set as August 20, 2026, in the system, but our search showed the actual date is August 26, 2026. Since the earnings release is in 6 days, we should update the fire date for this watch to August 26, 2026, or let it ride. Wait, since it hasn't fired yet, we don't resolve it. Let's make sure we record that it is scheduled for August 26, 2026. Let's check `019ef70b-6b49-7831-893f-6d78e9b2099e` (Oracle $40B Debt and Equity Financing Plan for FY2027) - This is an ongoing watch. We can see if there are any updates on Oracle's $40B debt/equity financing or ATM program. Let's search for Oracle $40B ATM or debt.

This is excellent context. No new major filing or issuance has occurred yet for Oracle beyond what was announced, so the watch is ongoing. Let's double check if we can write several highly detailed findings to update the existing notes: 1. **Nvidia's AI Moat Expands to $500 Billion Wall Street Private Credit Partnership Amid Circular Financing Concerns** (slug: `nvidia-record-q1-results-and-trillion-dollar-order-book`) - We must update this note with the massive **August 10, 2026** announcement of the **$500B Wall Street Compute Financing Platform** (MOUs with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR). - Detail the 25% residual support mechanism, Goldman Sachs' dual role (junior capital/private credit and debt placement), the SEC's July 29, 2026 regulatory relief on data center securities, and the narrowing of Nvidia's CDS to 72.11 bps on August 11, 2026. - Detail the reduction of Nvidia's planned guarantee on the OpenAI southern Ohio campus from $250B to under $120B (limited to the first 5 gigawatts of a 10GW program). - Discuss how this platform attempts to turn Nvidia compute into an "investable asset class," and the risk that all projects are collateralized by hardware from one manufacturer. 2. **Off-Balance-Sheet Joint Ventures and the New AI Infrastructure Financing Model** (slug: `off-balance-sheet-ai-infrastructure-financing-models`) - Update this with the Goldman Sachs estimate of **$1.5 trillion** in combined lease commitments for data centers, R&D, and equipment among hyperscalers (up from $200B five years ago), which includes **$1 trillion** in "uncommenced" lease commitments that do not yet show up in financial statements but represent future contractual obligations. - Discuss PIMCO's Lotfi Karoui's commentary (Aug 11, 2026) stating the AI capex cycle is on track to be the largest investment cycle since the 19th-century railway construction, with hyperscaler capex projected to surpass **$1 trillion per year** from 2027 onward. - Detail the collapse of Leopold Aschenbrenner's **Situational Awareness LP** in July 2026. It lost 67% in July, dropping from $45B in assets to $10B, triggered by 4x leverage and falling semiconductor stocks. Discuss the fire sale of its $16B public equity book (including SK Hynix, Nebius, and Micron) to Ken Griffin's **Citadel** at a ~10% discount on July 30, 2026. This swoop helped stem a $3 trillion AI stock rout. 3. **Is Nvidia's AI capex story still intact?** - Let's address the core question of the brief: Yes, the capex story is intact but transitioning from pure organic demand to a highly leveraged, financialized "documentation race" and "high-stakes poker game" between Nvidia and hyperscalers. - Hyperscalers are framing capex as "options" on Q2 calls, buying long-lead assets (land, shell, power) years in advance but delaying chip (short-lived asset) purchases until months before launch. - Hyperscalers are scaling their custom ASIC programs (Google's TPUs at +3M/yr; Amazon's Trainium/Graviton at $25B run rate; Microsoft's Maia 300 launching in Sept 2026 with TSMC allocation for 300k chips in 2027). - In response, Nvidia is bypassing hyperscalers by financing neoclouds via the **"AI Compute Partnership"** (formalized July 1, 2026) and the **$500B Wall Street platform** (announced August 10, 2026) to manufacture alternative demand, acting as a demand backstop and taking a share of neocloud revenues. - Let's check if we have a note on this or if we should write a new finding or update an existing one. We have an existing note: **Custom Silicon and Data Center Hardware Competition Escalates as Amazon's Custom Chip Run Rate Hits $25 Billion** (slug: `custom-silicon-and-datacenter-hardware-competition`). Let's update that note with the latest custom silicon updates (Google TPUs, Amazon Trainium, Microsoft Maia 300, Anthropic's chip usage, etc.). - Let's check if we have a note specifically on the hyperscalers' capex trajectory. We have: **Big Tech's AI Capex Surpasses $730 Billion in 2026 on Track for $1 Trillion in 2027 as Capacity Constraints Persist** (slug: `big-tech-capex-reaches-710b-in-2026`). Let's update this note with the latest Q2 2026 combined capex ($166B, up 87% YoY, 27% QoQ), the Wall Street forecast of $733B for 2026 and $939B for 2027, the "debt deluge" pushing hyperscaler debt footprint to over $360B, and the widening bond spreads (Amazon's $25B bond widening by 20 bps). Wait, let's call `read_notes` first to check the exact titles, slugs, and contents of the existing notes to make sure we integrate everything perfectly. This is a crucial step! Let's do it.

Now I will read the contents of the existing notes we want to update to see how to integrate our new findings seamlessly. First, let's fetch the contents of `nvidia-record-q1-results-and-trillion-dollar-order-book`. Wait, we don't have a specific tool to read the full body of a single note directly except that `read_notes` returns a preview. Ah, we can just rewrite/update the note by using `write_note` with the same slug. Since we are doing a revision, we should make sure we carry forward the core thesis and add the new details. Let's see: what are the key notes we want to write or update? 1. `nvidia-record-q1-results-and-trillion-dollar-order-book` (Nvidia's AI Moat Expands to $500 Billion Wall Street Private Credit Partnership Amid Circular Financing Concerns) 2. `off-balance-sheet-ai-infrastructure-financing-models` (Off-Balance-Sheet Joint Ventures and the New AI Infrastructure Financing Model) 3. `big-tech-capex-reaches-710b-in-2026` (Big Tech's AI Capex Surpasses $730 Billion in 2026 on Track for $1 Trillion in 2027 as Capacity Constraints Persist) 4. `custom-silicon-and-datacenter-hardware-competition` (Custom Silicon and Data Center Hardware Competition Escalates as Amazon's Custom Chip Run Rate Hits $25 Billion) Wait, let's fetch the full text of these notes if they are cached, or write them based on the preview + our new evidence. Let's do a search or check if we can fetch them. Ah, we don't have a tool to read a specific note's full body, but we can write a highly comprehensive note that fully covers the topic. Let's look at the preview of `nvidia-record-q1-results-and-trillion-dollar-order-book`: "The artificial intelligence capital expenditure boom supporting NVIDIA Corporation (NVDA) has entered a highly financialized phase. Nvidia is transitioning from a pure-play hardware supplier into the..." And `off-balance-sheet-ai-infrastructure-financing-models`: "As the capital expenditures required to build frontier AI infrastructure balloon toward $1 trillion (see [[big-tech-capex-reaches-710b-in-2026]]), the world's leading technology companies and cloud..." Let's carefully construct our updated notes. ### Note 1: `nvidia-record-q1-results-and-trillion-dollar-order-book` Let's update this note to detail Nvidia's massive $500B Wall Street Private Credit Partnership announced on August 10, 2026, which represents the pinnacle of this "highly financialized phase." We will cite: - The press release: `https://nvidianews.nvidia.com/news/nvidia-partners-with-apollo-blackrock-blackstone-brookfield-goldman-sachs-and-kkr-to-establish-ai-compute-infrastructure-financing-platforms-to-mobilize-over-500-billion-of-third-party-capital` - CNBC: `https://www.cnbc.com/2026/08/10/nvidia-wall-street-asset-managers-500-billion-ai-push.html` - Global Data Center Hub: `https://www.globaldatacenterhub.com/p/nvidias-500b-ai-infrastructure-financing` - UncoverAlpha: `https://www.uncoveralpha.com/p/nvidia-and-the-hyperscalers-high` - Market view: `/markets/NVDA/2026/08/20` Let's outline the contents: - **Title**: Nvidia's AI Moat Expands to $500 Billion Wall Street Private Credit Partnership Amid Circular Financing Concerns - **Introduction**: Nvidia is transitioning from a pure hardware supplier into a financial architect of the AI boom. - **The $500 Billion Wall Street Alliance**: Detailed breakdown of the August 10, 2026 announcement with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. Stated goal of mobilizing over $500B of third-party capital over time. - **The 25% Residual Support Mechanism**: Explain the "up to 25% of an opportunity" support mechanism, assessed project-by-project. Discuss the lack of definition around the base, waterfall position, or valuation method, creating a "documentation race" to set the market template. - **Goldman's Dual Role**: Goldman Sachs' mandate to provide junior capital/private credit via its asset management arm while placing senior debt with private-credit funds and public markets. - **Credit Market Response**: CDS narrowing to 72.11 bps on August 11, 2026, and the 2056 bond spread narrowing to 113 bps. - **The Southern Ohio Project Revision**: WSJ report that Nvidia slashed its planned guarantee for OpenAI's SB Energy campus in southern Ohio from $250B to under $120B (limited to the first 5 gigawatts of a 10GW program) within four days of the $500B announcement. - **The Circularity and Collateral Risk**: All projects in the program are collateralized by hardware from one manufacturer (Nvidia), meaning the recovery value of every lender depends on a single variable: the future market-clearing price of Nvidia compute. Let's write this note first. I will state my intent and write it using `write_note`. This is a substantive and deeply grounded note.I will now write the updated note for `nvidia-record-q1-results-and-trillion-dollar-order-book` to reflect the massive $500 billion Wall Street private credit alliance announced on August 10, 2026, and its structural implications.

This is excellent. Now let's update the second note: `off-balance-sheet-ai-infrastructure-financing-models` (Off-Balance-Sheet Joint Ventures and the New AI Infrastructure Financing Model). Let's review the new facts we have for this note: - **Hyperscaler lease commitments**: Goldman Sachs analysts estimated (Aug 6, 2026) that hyperscalers have combined lease commitments for data centers, R&D, and equipment of **$1.5 trillion** (up from $200 billion five years ago). This includes **$1 trillion of "uncommenced" lease commitments**, which are not yet recognized in financial statements but represent future contractual liabilities that will come due as facilities go live. - **PIMCO's Lotfi Karoui commentary**: On August 11, 2026, Karoui stated that the AI capex cycle is on track to be the largest investment cycle since the 19th-century railway construction, with consensus forecasts projecting hyperscaler capital spending alone to surpass **$1 trillion per year** from 2027 onward. - **The Leopold Aschenbrenner / Situational Awareness LP Collapse**: - Situational Awareness, an AI-focused hedge fund run by 24-year-old former OpenAI researcher Leopold Aschenbrenner, nearly imploded in late July 2026. - The fund was heavily leveraged (up to 4x / 400%) and had concentrated long positions in AI hardware, memory, and specialized cloud stocks including **SK Hynix**, **Micron**, and **Nebius**. - Ramped up by a 439% net return through June 2026, the fund's assets peaked at **$45 billion** on July 1, 2026. - A sharp correction in AI hardware stocks (e.g., SK Hynix tumbling almost 50% from its June peak) triggered massive margin calls from prime brokers (Bank of America, Goldman Sachs, JPMorgan). - On **July 30, 2026**, the fund was forced to liquidate its entire public equity book in a quick-fire block trade. Ken Griffin's **Citadel** (Wellington fund) bought the bulk of the book (valued at **$16 billion**) at a **~10% discount**, beating out Jane Street and Millennium. - This swoop by Citadel helped stem a broader $3 trillion AI stock rout, and the fund's remaining assets sit at $10 billion (mostly private holdings, including Anthropic). - This event highlights the high degree of leverage, concentration, and systemic volatility embedded in the public-equity side of the AI infrastructure trade. Let's write this updated note for `off-balance-sheet-ai-infrastructure-financing-models`. I will state my intent and write it using `write_note`.I will now write the updated note for `off-balance-sheet-ai-infrastructure-financing-models` to integrate the $1.5 trillion in lease commitments ($1 trillion uncommenced) and the dramatic collapse of Leopold Aschenbrenner's Situational Awareness LP.

This is fantastic. Now let's update the third note: `big-tech-capex-reaches-710b-in-2026` (Big Tech's AI Capex Surpasses $730 Billion in 2026 on Track for $1 Trillion in 2027 as Capacity Constraints Persist). Let's review the new facts we have for this note: - **Combined Q2 2026 Hyperscaler Capex**: Combined hyperscaler capex reached **$166.0 billion** in Q2 2026. This represents an **87% year-over-year increase** and a **27% quarter-over-quarter jump**. - **Consensus Forecast Upgrades**: Consensus estimates for hyperscaler capex have been raised significantly. Consensus now projects that 2026 spending will total **$733 billion**, and next year (2027) spending will grow 28% to **$939 billion** (approaching the $1 trillion mark). - **The "Debt Deluge" Footprint**: Since September 2025, the six major hyperscalers (Microsoft, Meta, Google, Oracle, Amazon, and Nvidia) have more than doubled their collective U.S. dollar debt footprint to over **$360 billion**. - **Bond Spread Pressure**: This massive issuance is putting pressure on investment-grade credit spreads. - In July 2026, Amazon issued a **$25 billion** jumbo bond (the ninth deal larger than $20 billion in 2026, with seven linked to hyperscalers). This pushed its existing 30-year bond **20 basis points wider**. - A Reuters analysis of LSEG data from late July 2026 showed that **78 out of 91 hyperscaler bonds** issued in 2026 were trading at higher yields than at issuance. - Specifically, the median spread on 2- to 4-year bonds rose to **40 basis points** (from 30 bps in 2025). - The median spread on 5- to 7-year debt increased to **60 basis points** (from 50 bps). - The median spread on bonds maturing in more than 20 years rose to **118 basis points** (from 108.5 bps). - While net leverage remains healthy, the sheer scale of issuance is forcing spreads wider as investor demand cools, leading hyperscalers to supplement debt with equity issuance. Let's write this updated note for `big-tech-capex-reaches-710b-in-2026`. I will state my intent and write it using `write_note`.I will now write the updated note for `big-tech-capex-reaches-710b-in-2026` to reflect the latest Q2 2026 capex figures, upgraded 2026/2027 consensus spending forecasts, and the credit market spread widening from the hyperscaler debt footprint.

This is incredible. Now let's update the fourth note: `custom-silicon-and-datacenter-hardware-competition` (Custom Silicon and Data Center Hardware Competition Escalates as Amazon's Custom Chip Run Rate Hits $25 Billion). Let's review the new facts we have for this note: - **The Inference Shift**: The AI market is shifting from training (where Nvidia's moat is strongest due to CUDA, NVLink, and tight clustering) to inference. Inference workloads are massively parallel but do not require the same tight cluster coherence, making custom ASICs (application-specific integrated circuits) highly cost-competitive. - **Google's TPU Program**: Google produces **+3 million TPUs annually**. In April 2026, Google expanded its partnership with Broadcom to bring roughly **3.5 additional gigawatts** of next-gen TPU capacity online starting in 2027. Google is also actively selling TPU systems to external cloud customers, directly competing with Nvidia's merchant market. Anthropic is a major customer, with a deal for access to up to 1 million TPUs. - **Amazon's Chip Business**: Amazon's custom silicon business (Trainium + Graviton) runs at a **$25 billion annualized revenue rate**, growing triple digits. Graviton is used by 98% of the top 1,000 EC2 customers. Amazon's Project Rainier (for Anthropic) is scaling from 500,000 Trainium chips toward 1 million. Trainium is highly competitive on a cost-per-token basis for inference because Amazon is pushing for low margins to drive adoption. - **Microsoft's Maia Program**: Microsoft is scheduled to unveil its **Maia 300** chip in September 2026. Microsoft is negotiating with TSMC for capacity to deliver more than **300,000 chips in 2027**, with an eventual target of over 1 million units ("gigawatts' worth" of Maia). Maia 300 targets a 30% to 40% cost advantage over Nvidia chips, and Microsoft is courting Anthropic as an anchor customer. - **Nvidia's Counter-Strategy (The "AI Compute Partnership")**: To counter the hyperscalers' custom silicon threat and prevent them from commoditizing its margins, Nvidia is bypassing them. On **July 1, 2026**, Nvidia formalized the "AI Compute Partnership" model to finance specialized "neoclouds" (such as Firmus and Sharon AI) as alternative demand channels, acting as a financial backstop and taking a share of neocloud revenues. Let's write this updated note for `custom-silicon-and-datacenter-hardware-competition`. I will state my intent and write it using `write_note`.I will now write the updated note for `custom-silicon-and-datacenter-hardware-competition` to reflect the shifting dynamics from training to inference, the latest custom ASIC milestones from Google, Amazon, and Microsoft, and Nvidia's strategic counter-moves to bypass the hyperscalers.

This is incredible work. Let's review our active watches and record events for any that have fired or have major updates. Let's list the relevant watches: 1. `019fee38-1dbd-73a3-8fa2-113b552feff5` (Nvidia $500B Wall Street Compute Financing Platform Capital Deployment) - Stated trigger: "Fires when Nvidia or its financial partners announce the first finalized compute financing deal or pool deployment under the $500B MOU." - Stated status: The MOU was signed on August 10, 2026, but no binding final agreement or first project has been finalized yet. This is a major progress event, but the watch remains ongoing. Let's record an event with `resolves: false`. 2. `01a007f9-8536-7aa4-97a1-5fc02943ef9a` (Hyperscaler SPV and Corporate Bond Spread Widening) - Stated trigger: "Fires if hyperscaler or SPV borrowing spreads widen by >=25 bps over the broader investment-grade market, confirming systemic credit tightening." - Stated status: Amazon's $25B jumbo bond sale in July 2026 pushed its existing 30-year bond 20 bps wider. Overall, 78 of 91 hyperscaler bonds issued in 2026 are trading at higher yields than at issuance, and median spreads have risen by 10 bps across various tenors. This is a significant update, showing clear spread pressure but not yet crossing the systemic 25 bps threshold across the broad IG market. Let's record an event with `resolves: false`. 3. `019ed2f6-f136-7051-a5b0-fa3b26ec02e0` (Nvidia Q2 FY2027 Earnings Release) - Stated trigger: "Fires when Nvidia officially reports its Q2 FY2027 financial results. Check for revenue vs $91B guidance, data center growth, and Blackwell/Rubin ramp updates." - Stated status: The earnings release is scheduled for August 26, 2026, at 2:00 PM EDT. Today is August 20, 2026, so it has not fired yet. Let's keep it open. Let's record the events for the first two watches. I will call `record_event` for each of them. Let's state my intent and execute.I will record an event for the watch `019fee38-1dbd-73a3-8fa2-113b552feff5` (Nvidia $500B Wall Street Compute Financing Platform Capital Deployment) to capture the official MOU signing on August 10, 2026, and its structural terms. This watch remains ongoing as no final agreements or first projects have been executed.

I will now record an event for the watch `01a007f9-8536-7aa4-97a1-5fc02943ef9a` (Hyperscaler SPV and Corporate Bond Spread Widening) to reflect the widening bond spreads and the massive debt footprint of hyperscalers. This watch remains ongoing as the systemic widening has not yet crossed the 25 bps threshold across the broad investment-grade corporate market.

Now let's check if there are any other open threads or watches we should address. Wait, let's review our findings and the core brief: "Is Nvidia's AI capex story still intact?" Our findings show: 1. **Capex Volume is Intact and Growing**: Q2 2026 combined hyperscaler capex exploded by 87% YoY and 27% QoQ to $166B. Wall Street has upgraded full-year 2026 consensus capex to $733B, and 2027 to $939B. 2. **The Funding Model is Shifting**: Hyperscalers have doubled their USD debt footprint to over $360B since Sept 2025. Goldman Sachs estimates they have $1.5 trillion in lease commitments ($1 trillion of which is uncommenced and off-balance-sheet), which will come due as data centers go operational. 3. **Credit and Leverage Tensions are Rising**: The debt surge is putting pressure on investment-grade bond spreads (Amazon's $25B bond pushed spreads 20 bps wider, and 78 of 91 hyperscaler bonds are trading at higher yields than at issuance). The dramatic collapse of the highly leveraged ($45B peak, 4x levered) AI hedge fund **Situational Awareness LP** in late July 2026 required Ken Griffin's **Citadel** to step in and buy its $16B public book at a 10% discount, highlighting the leverage and concentration risks of the AI trade. 4. **The "High-Stakes Poker Game" and Custom Silicon**: Hyperscalers are trying to commoditize Nvidia's ~75% gross margins by shifting to inference-led workloads and scaling custom ASICs (Google's TPUs, Amazon's Trainium, Microsoft's Maia 300 launching in September). They are framing chip purchases as "options" on earnings calls. 5. **Nvidia's Financial Counter-Offensive**: Nvidia is bypassing hyperscalers by using its massive $100B+ annual free cash flow to backstop and co-invest in specialized "neoclouds" (July 1, 2026 "AI Compute Partnership") and establishing the **$500B Wall Street Compute Financing Platform** (announced August 10, 2026 with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR). This platform attempts to turn Nvidia compute into an "investable asset class," though it introduces systemic single-manufacturer collateral risk. This is an incredibly robust, deeply evidenced, and cohesive narrative that perfectly answers the brief. Let's check if we need to call `finish_cycle`. Since we have written and updated all our findings and recorded events on the active threads, we are ready to call `finish_cycle`. Let's do it! No findings reason is empty since we wrote multiple findings. Let's execute.

### Executive Summary: Is Nvidia's AI Capex Story Still Intact? Yes, the AI capital expenditure story supporting **NVIDIA Corporation (NVDA)** remains structurally intact and is accelerating in terms of absolute volume, but it has entered a highly financialized, leveraged, and competitive phase. The landscape has transitioned from simple organic product demand into a **"high-stakes poker game"** between Nvidia and its largest customers (the hyperscalers), characterized by massive off-balance-sheet debt accumulation, custom silicon escalation, and a historic private credit mobilization by Nvidia to bypass the cloud giants. --- ### Key Findings & Structural Developments #### 1. Capex Volume is Exploding but Shifting to a "Debt Deluge" * **Record Spending**: Combined hyperscaler capex reached **$166.0 billion** in Q2 2026, representing an **87% year-over-year increase** and a **27% quarter-over-quarter jump** (see [[big-tech-capex-reaches-710b-in-2026]]). * **Upgraded Consensus**: Wall Street has raised its 2026 consensus capex projection to **$733 billion**, with 2027 projected to grow 28% to **$939 billion**, putting the industry on track to surpass $1 trillion in annual capital expenditures by 2028. * **The Debt Footprint**: To fund this, hyperscalers have more than doubled their collective U.S. dollar debt footprint since September 2025 to over **$360 billion**. This massive supply has pressured investment-grade spreads wider; for example, Amazon’s **$25 billion** bond sale in July 2026 pushed its existing 30-year bond **20 basis points wider**. * **Off-Balance-Sheet Leases**: Goldman Sachs estimates that hyperscalers have accumulated **$1.5 trillion in combined lease commitments** (up from $200 billion five years ago), which includes **$1 trillion in "uncommenced" lease commitments** that do not yet appear on balance sheets but represent future contractual obligations (see [[off-balance-sheet-ai-infrastructure-financing-models]]). #### 2. The Leverage Rupture: Leopold Aschenbrenner's Situational Awareness LP Collapse * **The Collapse**: The high-leverage risk of the AI infrastructure trade was demonstrated in late July 2026 when the AI-focused hedge fund **Situational Awareness LP** (managed by 24-year-old former OpenAI researcher Leopold Aschenbrenner) nearly imploded. * **The Fire Sale**: After riding concentrated long positions in AI hardware (including SK Hynix, Micron, and Nebius) to a peak AUM of **$45 billion** on July 1, 2026, a sharp pullback in semiconductor stocks turned the fund's **400% (4x) leverage** against it. * **Citadel's Swoop**: To meet margin calls from prime brokers, the fund sold its entire public stock book—valued at **$16 billion**—to Ken Griffin's **Citadel** at a **~10% discount** in a quick-fire block trade on July 30, 2026. This massive swoop helped stem a broader $3 trillion AI market rout and fueled a subsequent 30% rebound in the sector. #### 3. The Custom Silicon Threat: Shifting from Training to Inference * **The Inference Shift**: As the AI market shifts from training (where Nvidia’s moat is strongest) to inference (running models), custom ASICs designed for specific workloads are proving highly cost-competitive on a cost-per-token basis (see [[custom-silicon-and-datacenter-hardware-competition]]). * **Hyperscaler Milestones**: * **Google**: Produces **+3 million TPUs annually** and expanded its Broadcom partnership in April 2026 to secure **3.5 gigawatts** of next-gen TPU capacity starting in 2027. * **Amazon**: Its custom silicon business (Trainium + Graviton) is running at an annualized revenue rate of **over $25 billion**, growing at triple digits. * **Microsoft**: Is scheduled to unveil its **Maia 300** chip in September 2026, negotiating with TSMC for capacity to deliver **300,000+ chips in 2027** with a targeted 30% to 40% cost advantage over Nvidia. * **Capex as an "Option"**: On Q2 2026 calls, hyperscalers explicitly framed chip purchases as short-lived "options," committing to long-lead assets (land, power, shells) years in advance but delaying chip purchases until months before data centers go live so they can scale back if demand softens. #### 4. Nvidia’s Financial Counter-Offensive: Bypassing the Cloud Giants * **AI Compute Partnership**: To counter hyperscaler commoditization and secure its own demand, Nvidia formalized a model on July 1, 2026, where it acts as a financial backstop for specialized "neoclouds" (e.g., Firmus, Sharon AI) in exchange for a direct share of their cloud revenue (see [[custom-silicon-and-datacenter-hardware-competition]]). * **The $500 Billion Wall Street Alliance**: On **August 10, 2026**, Nvidia signed MOUs with **Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR** to establish independent compute financing platforms designed to mobilize over **$500 billion of third-party capital** (see [[nvidia-record-q1-results-and-trillion-dollar-order-book]]). * **The 25% Residual Support**: Nvidia will provide a residual-value support mechanism for **up to 25% of an opportunity**, assessed project-by-project. This has triggered a "documentation race" among the asset managers to execute the first binding agreement and write the market template. * **Guarantees Downsized**: Highlighting its shift to offload concentrated liabilities onto Wall Street private credit, Nvidia cut its planned guarantee for OpenAI’s SB Energy campus in southern Ohio from $250 billion to **under $120 billion** (limited to the first 5 gigawatts of a 10GW program) within four days of the $500B announcement. * **The Systemic Collateral Risk**: While this platform disperses borrower risk across six independent asset managers, **the underlying collateral remains 100% concentrated in Nvidia hardware**. The recovery value of every lender depends entirely on a single variable: the future market-clearing price of a GPU hour. A systemic shock to GPU-hour pricing would impair the collateral value of all projects simultaneously. --- ### Key Watches & Catalysts 1. **Nvidia Q2 FY2027 Earnings (August 26, 2026)**: Expected to report Q2 FY2027 results and provide guidance for Q3. Key metrics to monitor include whether revenue meets/exceeds the guided $91 billion (+/- 2%) and if gross margins remain flat at ~74-75%. 2. **Microsoft Maia 300 Launch (September 2026)**: Monitor the official commercial release, TSMC 3nm allocation, and whether Anthropic is officially secured as an anchor customer. 3. **The First Compute-Backed Rated Tranche (Ongoing)**: Watch for the first rated securitization of compute-backed debt under the new SEC regulatory relief, which will convert this private origination business into a liquid, tradable public credit market.