Cycle on Aug 17, 2026

I am starting by listing the open threads for this topic to understand what has already been established and what specific questions or watches are currently outstanding.

I will now read the existing notes to see the current findings and slugs so I can update them or understand the exact context.

We have several interesting outstanding threads and watches. Specifically: - `019feabd-a6f0-75a7-8e81-6af5f9b659bd`: "Monitor Q2 2026 earnings for CoreWeave (CRWV) and Nebius (NBIS) to track debt growth (expected to hit $33B for CoreWeave), revenue trajectory, and competitive pressure from SpaceX's entry into the neocloud market." The fire date is August 12, 2026. This is in the past (today is Aug 17, 2026). Let's check CoreWeave and Nebius Q2 2026 earnings! - Let's check `market_view` for CoreWeave (CRWV) and Nebius (NBIS). Wait, are they public? Nebius (NBIS) is public. CoreWeave (CRWV) was planning an IPO or has filed. Let's run a search and fetch some market data first. Let's do a market view check on NBIS, CRWV, and also look for their Q2 2026 earnings. Let's first search the web for "CoreWeave Q2 2026 earnings" or "Nebius Q2 2026 earnings", and check market views. Let's start with market views for NBIS and CRWV to see if they are active.

I will now call `market_view` for CoreWeave (CRWV) to see if there is data. CoreWeave was expected to have its IPO or maybe it's listed. Let's see.

This is incredibly interesting. Let's look at the numbers. For CoreWeave (CRWV): - Total debt: $51.61 billion as of 2026-06-30! Wait, in the watch thread `019feabd-a6f0-75a7-8e81-6af5f9b659bd`, we had: "expected to hit $33B for CoreWeave". Instead, it has skyrocketed to **$51.61 billion**! - Let's look at the cash flow for Q2 (ending 2026-06-30): operating cash flow of $679.0M, capex of $6.42 billion, resulting in negative free cash flow of $5.74 billion for the quarter! - Net income for Q2 was -$626.0M, and EBITDA was $76.0M (down from $1.03B in the prior quarter, which is a massive drop!). - Total debt is $51.61B against Cash of $6.40B. - Debt/EBITDA is 13.6x! - Magnetar Financial (a 10% owner) has been selling substantial amounts of stock and warrants: $213.8M of stock on August 13, 2026, 1.92M shares, 200,000 warrants, and has entered call options. Let's look at Nebius (NBIS): - Total debt is $9.59B as of 2026-03-31, cash is $9.30B. Let's see if we can find more updated numbers for Q2 2026. - Let's do a web search on CoreWeave and Nebius Q2 2026 earnings to see what details have been reported and what comments or reports are available. Let's start with searching about CoreWeave and Nebius earnings.

Let's fetch the SEC filing or press release for CoreWeave's Q2 2026 earnings to get the exact numbers. Let's do a `web_fetch` on `https://www.sec.gov/Archives/edgar/data/1769628/000176962826000362/coreweave2q26earningspress.htm` or the CNBC article `https://www.cnbc.com/2026/08/11/coreweave-crwv-q2-earnings-report-2026.html`. Let's fetch both to be sure. Let's start with the SEC archive link as it contains the official press release.

Let's also look at the CNBC article about CoreWeave's Q2 earnings to see if there is any other context, especially about the debt and the "debt trap" or backlog details. Let's fetch `https://www.cnbc.com/2026/08/11/coreweave-crwv-q2-earnings-report-2026.html`.

This is extremely valuable! Let's analyze the details of CoreWeave's Q2 earnings: - **Revenue**: $2.575 billion vs. $1.212 billion in Q2 2025 (up 112.5% YoY). - **Net Loss**: $626 million (up from $290 million in Q2 2025). - **Operating Loss**: $49 million (compared to an operating income of $19 million in Q2 2025). - **Interest Expense, Net**: $640 million (up from $267 million in Q2 2025). This is massive! The interest expense of $640 million actually exceeds their operating loss of $49M, and is the primary driver of their huge net loss of $626 million. This is a classic "debt-trap" scenario where interest on their capital-intensive buildout consumes all operating profitability. - **Adjusted EBITDA**: $1.510 billion vs $753 million in Q2 2025. Wait, how do they get to $1.510 billion Adjusted EBITDA from an operating loss of $49 million? It's primarily because of massive depreciation and amortization (from their massive GPU and data center assets). - **Total Debt**: CNBC reports "As of quarter end, CoreWeave had $35 billion in debt on its balance sheet to cover the cost of Nvidia graphics processing units and other equipment." Wait, our market view says "Total debt $51.61B" as of 2026-06-30. Let's look at the difference. The market view might include lease liabilities (which are capitalized under ASC 842 and are technically debt-like, especially for data center leases and GPU leases), while CNBC might be referring specifically to funded debt (like loans, notes, convertibles). Or perhaps total liabilities or lease liabilities are very large. Let's check CoreWeave's balance sheet if we can find it, or let's search for "CoreWeave balance sheet Q2 2026" or "CoreWeave total debt". - Let's look at their **Capex Guidance**: CoreWeave raised its annual capex forecast to **$35 billion to $39 billion** (up from $31 billion to $35 billion in May)! This is an extraordinary amount of capex for a company with $7.59B TTM revenue. - **Revenue backlog**: $104 billion as of June 30, 2026. This excludes over $25 billion in new commitments from Q3. - **Active power**: 1.5 gigawatts (GW) of active power, with guidance of over 1.85 GW by year end. - **Competitors**: SpaceX has begun selling excess computing capacity, and Meta has considered launching a cloud business. - **Magnetar selling**: Magnetar Financial, a 10% owner, sold $213.8M in stock on August 13, 2026 (immediately after the earnings pop on August 11). Let's do a search on Nebius (NBIS) Q2 2026 earnings as well. The market view showed Q2 2026 revenue of $582.3 million (up 454% YoY) and a diluted loss of $0.68 per share. Let's see if we can find more details, especially about their debt, capex, and cash burn. Let's do a web search.

Let's fetch `https://mvcinvesting.substack.com/p/nebius-group-nbis-q2-2026-earnings` or `https://discussion.fool.com/t/my-nebius-q2-2026-notes/125967` to get the details of Nebius's Q2 cash flow, debt, and capex. Let's start with the substack link. This will give us a very clear picture of Nebius's Q2 cash flow, capex, and debt.

Ah, firecrawl timed out on that link. Let's try to fetch the other one: `https://discussion.fool.com/t/my-nebius-q2-2026-notes/125967`. Let's also do a search to gather more detailed textual information about Nebius's Q2 results, cash flow, debt, and capex. Let's start with fetching the Motley Fool discussion link.

This is incredibly rich and detailed. Let's look at the key points for Nebius: - **Revenue**: $575 million or $582.3 million (depending on GAAP/Non-GAAP, let's use the official $582.3 million reported in the market view). - **Capex**: $5.7 billion in Q2 (up from $2.5 billion in Q1), H1 capex is $8.1 billion. - **2026 Capex guidance**: $20 billion to $25 billion reiterated. - **Operating cash flow**: $2.3 billion in Q2, but due to $5.7 billion in capex, they have negative free cash flow. - **Funding / Financing**: - Sold 12.7 million Class A shares through its ATM (At-The-Market) program at an average price of $223.6, raising approximately $2.8 billion. 12.3 million shares remained available under the ATM at quarter-end. - In July 2026, Nebius raised $775 million through its first secured asset-backed facility. - **Michael Burry Short**: Michael Burry has publicized a large short on Nebius (NBIS) at $212, describing the opportunity as "shooting fish in a barrel." This is an incredibly interesting, high-profile short-seller bet against Nebius! - **Customer Prepayments**: Customer prepayments cover 50% to 70% of project capex. - **Average Contract Value (ACV)**: Reached roughly $40 million per MW. - **Flywheel / Circular Financing**: The poster mentions "the circular flywheel invested by NVIDIA in the neoclouds." Note that NVIDIA holds a 9.3% stake in Nebius (mentioned in the search results: "Nvidia's 9.3% stake"). Let's do a quick search on Michael Burry's short of Nebius (NBIS) to get the exact details, quotes, and sources. This is a brilliant piece of the exposure map and a direct signal of market skepticism.

This is extremely interesting! Let's fetch the Seeking Alpha article `https://seekingalpha.com/news/4632100-burry-adds-to-ai-shorts-says-nebius-is-what-the-top-of-a-boom-looks-like` and/or the Benzinga article `https://www.benzinga.com/markets/prediction-markets/26/08/61210492/michael-burry-ai-shorts-oracle-micron-nebius` to get the exact quotes. Let's start with Seeking Alpha.

This is a superb article! Let's fetch the Street article or the Yahoo Finance article to see if we can get Burry's actual quotes or tweets/posts about Nebius. Let's do a search for "Burry" "Nebius" "top of a boom" or "backwardation" or "depreciation" or "shooting fish". Let's search with a broader query.

Let's fetch `https://www.thestreet.com/investing/stocks/michael-burry-nebius-bearish-bet` to get the exact quotes from Michael Burry. This is incredibly juicy! Burry is arguing that the AI boom rests on aggressive accounting (extending depreciation useful lives of GPUs/servers from 4 to 5 years, or more) and that the high-pricing of shorter-term contracts is a sign of "backwardation," meaning that the market expects a massive glut/price drop in the future, which is exactly why customers are willing to pay double for 6-month contracts rather than lock in long-term contracts. This is a brilliant economic argument! Let's fetch the Street article first.

This is an incredible, masterfully written article that gives us precisely the details we need to construct a world-class note. Let's analyze the core components of this "backwardation" and "depreciation extension" argument: 1. **The "Backwardation" Argument in AI Compute Pricing**: - Nebius management noted that they could potentially sell all of their 2027 capacity via 1- to 3-year agreements but are deliberately keeping some supply available for shorter-term contracts (e.g. 6 months). - Short-duration contracts command **$40 million to $50 million per megawatt (MW)**. - Midterm (1-3 year) contracts are priced at **$20 million to $25 million per MW**. - This means short-duration contracts are priced at nearly double the rate of midterm agreements. - In commodities, this is called **backwardation**—where the spot price (or short-term price) is much higher than the future price. - **Burry's Interpretation**: Why are customers willing to pay double for short-term access instead of locking in the lower midterm price? It suggests they expect the value of future compute to drop dramatically (due to a future glut or rapid obsolescence of current chips). It signals extreme near-term urgency but a complete lack of belief in the long-term value of the infrastructure. 2. **The "Depreciation Extension" Accounting Maneuver**: - At the start of 2026, Nebius extended the depreciation useful life of its servers and network equipment from **four years to five years**. - **Burry's Interpretation**: This is an aggressive accounting shift that artificially flattens operating expenses and boosts reported earnings. It directly conflicts with the "backwardation" pricing signal. If future compute pricing is expected to drop sharply (implying rapid economic obsolescence), why is Nebius assuming its hardware will remain economically useful *longer*? Extending depreciation useful lives is a classic way to inflate current operating income and margins (which Nebius reported as 41% group EBITDA margin and 50% AI EBITDA margin) while masking the rapid decay of the physical assets. 3. **Enron Comparison**: - On August 12, 2026, Burry recommended rewatching the Enron documentary *"The Smartest Guys in the Room"*, stating that he feels "history is repeating." This suggests he believes there is mark-to-market or circular accounting manipulation happening in the neocloud/AI infrastructure space. 4. **Burry's Positions**: - Burry has added to his shorts in **Nebius (NBIS)** (at $247), **Micron (MU)** (at $924), **Oracle (ORCL)** (at $152), and the **iShares Semiconductor ETF (SOXX)**. - He also shorted **Oracle (ORCL)** at $145, citing concerns about ballooning debt, long-term commitments, and off-balance-sheet obligations. This is a brilliant development that we must write a note on. Let's check if there is an existing note we can update or if we should create a new note. We have: - `nebius-meta-megadeal-and-financing` (which is titled "Nebius and CoreWeave Face Severe Backlog and Demand Risks But Find an Alphabet Lifeline"). - `nvidia-circular-vendor-financing-loop` - Let's check if we have a note for Michael Burry or if we can update `nebius-meta-megadeal-and-financing` to reflect these Q2 2026 earnings and Burry's short thesis. Wait, let's look at the notes list again. Yes, `nebius-meta-megadeal-and-financing` is exactly about Nebius and CoreWeave. We can update `nebius-meta-megadeal-and-financing` or we can write a fresh note about Michael Burry's short thesis and the accounting maneuvers since it tracks a very specific theme: the "depreciation useful lives and backwardation of AI compute" which applies to Nebius, CoreWeave, and Oracle (all three of which Burry is shorting!). Let's see if we should create a new note or update `nebius-meta-megadeal-and-financing`. Wait! The guidelines say: "Name a slug for the DURABLE SUBJECT it tracks: the throughline, entity, or running debate (`ai-coding-backlash`, `nvidia-datacenter-margins`) — never the transient story-of-the-day" A durable subject like `nebius-meta-megadeal-and-financing` can be updated. But let's see if we can create a more durable slug like `neocloud-depreciation-and-compute-backwardation` or update the existing `nebius-meta-megadeal-and-financing` or `coreweave-liquidity-crisis-and-syndicated-debt`. Wait! We have two core entities: Nebius (NBIS) and CoreWeave (CRWV). Both reported Q2 2026 earnings. Let's update `nebius-meta-megadeal-and-financing` since it already tracks Nebius and CoreWeave. Let's look at the content of `nebius-meta-megadeal-and-financing` first to see how we can expand it. Let's fetch the note `nebius-meta-megadeal-and-financing` using `read_notes` or read it from our context. Oh, `read_notes` only gives previews. Let's fetch the actual note content if we can, or we can just overwrite it with a comprehensive update. Wait, is there a tool to read a specific note? No, `read_notes` returns slugs, titles, previews, updated_at, and source_count. Let's search if we can read the note body. No, there is no tool to read the note body directly. But we can just update `nebius-meta-megadeal-and-financing` with the latest Q2 earnings, Burry's short, and the structural risks of both companies. Let's look at the other watches and see if any other watches have fired: - `019feabd-a6f0-75a7-8e81-6af5f9b659bd`: "Monitor Q2 2026 earnings for CoreWeave (CRWV) and Nebius (NBIS) to track debt growth (expected to hit $33B for CoreWeave), revenue trajectory, and competitive pressure from SpaceX's entry into the neocloud market." This watch has DEFINITELY fired! CoreWeave reported Q2 2026 earnings on August 11, and Nebius reported on August 12. We should call `record_event` on this watch. - Let's check `019f1273-ce52-7485-a89d-8b5383b6d175`: "Monitor secondary market trading prices and credit spreads of CoreWeave's $3.1 billion syndicated DDTL 5.0 facility." Let's see if there is any news about CoreWeave's DDTL 5.0 secondary market pricing or credit spreads. Let's do a search.

This is incredibly helpful! Let's analyze the new debt developments for CoreWeave: - CoreWeave closed yet another massive delayed draw term loan: **DDTL 5.5 Facility for $2.6 billion**, closed on **August 10, 2026** (just one day before earnings!). This facility was issued through CoreWeave Financing DDTL V-V, LLC. - Fitch assigned a **'BB+' rating** to this proposed DDTL 5.5 facility on July 16, 2026, affirming the company's IDR at BB- (with positive outlook) and DDTL 5.0 at 'BB+'. - This brings their series of DDTLs to: - **DDTL 4.0**: $8.5 billion (closed March 31, 2026) borrowed by CoreWeave Compute Acquisition Co. (rated A3 by Moody's, interesting structure because it has investment-grade rated tranches or is investment-grade rated as a project finance vehicle!). - **DDTL 5.0**: $3.1 billion (closed May 18, 2026) through CoreWeave Financing DDTL V, LLC. This is the first publicly syndicated one (enabling secondary market trading). - **DDTL 5.5**: $2.6 billion (closed August 10, 2026) through CoreWeave Financing DDTL V-V, LLC. - This is a staggering amount of debt! No wonder their total debt is listed as **$51.61 billion** on their balance sheet as of 2026-06-30 (which likely includes the $8.5B DDTL 4.0, the $3.1B DDTL 5.0, earlier facilities like DDTL 3.0, and massive capitalized lease liabilities for their 1.5 GW of data center power). Let's check if there is any other major player we should analyze. The prompt mentions: - The chip layer (Nvidia, AMD, Broadcom, Marvell, TSMC) - The hyperscaler buyers (Microsoft, Amazon, Google, Meta, Oracle) and their capex guidance. - The power/physical buildout (Vertiv, Eaton, Constellation, Vistra). - Let's search for "Vistra Corp debt" or "Vistra total debt" to see if it crossed the $22 billion threshold specified in our watch thread `019eca5a-01d4-7c6d-ab25-99092c634700`: "Track Vistra Corp's (VST) quarterly total debt levels and leverage. The company's debt currently stands at $19.91 billion. We will monitor if total debt crosses $22 billion as it continues to finance its capital-intensive power buildout." - Let's run a `market_view` on VST first. Let's do that!

The total debt of VST is $19.89 billion as of June 30, 2026. This is very close to the $19.91 billion baseline but has NOT crossed $22 billion yet. So the watch is ongoing and does not fire this cycle. Let's keep it ongoing. Let's check if there are other watches that have fired: - `019feabd-a6f0-75a7-8e81-6af5f9b659bd`: "Monitor Q2 2026 earnings for CoreWeave (CRWV) and Nebius (NBIS) to track debt growth (expected to hit $33B for CoreWeave), revenue trajectory, and competitive pressure from SpaceX's entry into the neocloud market." This watch fired! CoreWeave's debt actually hit **$35 billion** (excluding capitalized leases, and total debt is $51.61B including leases) and Nebius's revenue soared to $582.3M. Let's record this event! - Let's see if there are other watches. - `019fc6b4-e2fe-75a5-ac8a-ba0bc82a9ccf`: "Monitor Q3 2026 earnings for Alphabet, Meta Platforms, and Microsoft to track updated 2026/2027 capex guidance..." The fire date is October 28, 2026. Today is August 17, 2026, so this is in the future. - `019fa2a6-4223-7d3d-9a92-29605f9d1266`: "Meta Platforms Iris AI Chip Production Start." Fire date is September 1, 2026. This is in the future. - `019feabd-f129-7a64-a91f-ee8a0bcca989`: "SpaceX Colossus Lease 90-Day Cancellation Provision Active." Fire date is December 31, 2026. In the future. - `019f1273-ce52-7485-a89d-8b5383b6d175`: "Monitor secondary market trading prices and credit spreads of CoreWeave's $3.1 billion syndicated DDTL 5.0 facility." This is ongoing. Let's do a quick search on other companies. The brief asks us to: - "Track who is financially exposed if the AI infrastructure buildout slows — the exposure map nobody has stitched together." - "Core entities: the chip layer (Nvidia, AMD, Broadcom, Marvell, TSMC); the hyperscaler buyers (Microsoft, Amazon, Google, Meta, Oracle) and their capex guidance; the "neocloud" layer (CoreWeave, Nebius, Lambda) and how it's financed; the power/physical buildout (Vertiv, Eaton, Constellation, Vistra); and the debt — who funds data centers with leverage, the vendor-financing loops, and the private-credit exposure." - "I want to track quarterly capex guidance and revisions, signals about demand sustainability and customer concentration, and the financing structures underneath (debt issuance, SPVs, circular supplier-funds-customer deals)." Let's see: we have some incredible findings to report for: 1. **CoreWeave's Q2 2026 Earnings & Escalating Debt Structure**: - Let's check CoreWeave's capex guidance: they raised their 2026 capex guidance to **$35 billion to $39 billion** (from $31B-$35B in May). This is an astonishing capex-to-revenue ratio, with TTM revenue of $7.59B and Q2 revenue of $2.58B. - Their total debt on the balance sheet is **$51.61 billion** as of June 30, 2026, and they have **$35 billion** in funded debt. - They just closed a **$2.6 billion DDTL 5.5 facility** on August 10, 2026, issued through CoreWeave Financing DDTL V-V, LLC. This follows their **$3.1 billion DDTL 5.0** in May and **$8.5 billion DDTL 4.0** in March. - Their interest expense for the quarter was **$640 million**, which completely wiped out their $1.51 billion in Adjusted EBITDA (which is heavily inflated by depreciation of short-lived GPUs) and led to a **$626 million net loss** for the quarter, compared to a $290 million net loss in the prior year. This highlights the severe "debt-trap" risk where interest payments consume all operational cash flows. - Magnetar Financial, a 10% owner, has been aggressively selling stock and warrants, offloading **$213.8 million** in stock on August 13, 2026, right after the stock popped 14% on earnings. 2. **Nebius Group's Q2 2026 Earnings & Michael Burry's Short Thesis**: - Nebius reported Q2 revenue of **$582.3 million** (up 454% YoY). Group EBITDA was $236 million (41% margin) and Nebius AI EBITDA margin was 50%. - However, their growth is extremely capital intensive: Q2 capex was **$5.7 billion** (up from $2.5 billion in Q1), bringing H1 capex to $8.1 billion. This means they are burning massive amounts of cash, resulting in deeply negative free cash flow. - They are funding this by selling shares through their At-The-Market (ATM) program, raising **$2.8 billion** by selling 12.7 million shares at an average price of $223.6 during Q2. - **Michael Burry** has disclosed a large short position in Nebius (NBIS) at $247 (doubling down on his short first entered at $211.77), alongside shorts in Oracle (ORCL) and Micron (MU). - **Burry's core thesis** reveals two deep structural contradictions: - **Compute "Backwardation"**: Short-term (6-month) contracts command **$40 million to $50 million per MW**, while midterm (1-3 year) contracts are priced at **$20 million to $25 million per MW**. Customers are paying double for short-term access. This is a classic commodities backwardation signal, meaning that the market expects a massive glut/price drop in future compute. - **Depreciation Extension**: Despite this pricing backwardation pointing to rapid economic obsolescence of current chips, Nebius extended the depreciation useful life of its servers and network equipment from **four years to five years** at the start of 2026. This aggressive accounting maneuver flattens current operating expenses and inflates margins, but creates a massive future write-down risk when the value of the physical assets decays faster than the accounting schedule assumes. - **Enron Comparison**: On August 12, Burry recommended rewatching the Enron documentary *"The Smartest Guys in the Room"*, warning that history is repeating in the AI infrastructure space. This is a phenomenal double-header of findings. Let's write these up! Wait, let's look at our prior findings to see if there are slugs we should reuse. - `nebius-meta-megadeal-and-financing` (Nebius and CoreWeave Face Severe Backlog and Demand Risks But Find an Alphabet Lifeline). This slug is perfect for the Nebius / Michael Burry story. Or we can update both `nebius-meta-megadeal-and-financing` and `coreweave-liquidity-crisis-and-syndicated-debt`. Let's see: - `coreweave-liquidity-crisis-and-syndicated-debt` is titled "CoreWeave's Extreme Leverage, Widening Losses, and the Bifurcated Cost of GPU Debt". We can update this slug with CoreWeave's Q2 2026 earnings, the new DDTL 5.5, the $640M interest expense, and the Magnetar selling! - `nebius-meta-megadeal-and-financing` is titled "Nebius and CoreWeave Face Severe Backlog and Demand Risks But Find an Alphabet Lifeline". We can update this slug to focus on Nebius's Q2 2026 earnings, Michael Burry's short thesis, the compute "backwardation" pricing anomaly, and the depreciation extension accounting shift! Let's check if we have other prior findings we can update. - `applied-digital-coreweave-exposure-and-debt` (Applied Digital Exposes Physical Buildout to High-Yield Leverage and Cash Burn) - `amd-anthropic-circular-financing-and-helios-buildout` (AMD and Anthropic Replicate NVIDIA's Circular Vendor-Financing Playbook) - `nvidia-circular-vendor-financing-loop` (The NVIDIA Circular Vendor-Financing Loop and Systemic Domino Risk) - `utility-leverage-and-power-buildout-exposure` (Power Buildout and Utility Exposure to AI Capital Spending Slowdown) Let's do a search on other major players in the exposure map to see if there are any other quarterly updates or earnings-call commentary that we should include. Let's search for "Nvidia" "vendor-financing" or "customer concentration" or "capex" in the last week. Let's see if there is any other major piece of the exposure map. Let's do a search.

This is an extraordinary analysis from Daloopa! Let's examine the key insights: - **Nvidia Customer Concentration**: Nvidia's customer concentration has become highly material, with **61% revenue concentration** now! - **Accounts Receivable (AR) Mix**: At year-end, the largest AR customer accounted for **25%** of Nvidia's $38.5 billion AR balance, which is roughly **$9.6 billion**! The second-largest accounted for **18%** (roughly **$6.9 billion**). This means just two customers account for **43%** of Nvidia's total accounts receivable, exposing Nvidia to massive credit risk if either of these players experiences a cash crunch or defaults. - **Custom Silicon Substitution**: The largest buyers (hyperscalers like Amazon, Google, Meta, and Microsoft) are the ones aggressively building their own custom silicon alternatives at scale (e.g., Amazon AWS Trainium, Google TPU v7/v8, Meta Iris AI chip, Microsoft custom silicon). This means Nvidia's biggest customers are also funding and designing their primary alternatives. - **Nvidia Data Center Revenue Share**: Data center revenue now represents **92%** of Nvidia's total revenue, up from 60% two years ago. This means concentration in data center customers is effectively concentration in the overall business. - **Capex-to-Revenue Spread**: Over the past eight quarters, Nvidia's data center revenue has accounted for between 39% and 47% of total hyperscaler capex, peaking at 47% in Q1 2025 and recently standing at 45%. This shows that the share of incremental capex flowing to Nvidia is no longer expanding, signaling that custom silicon and other suppliers are beginning to capture a larger share of the capex pie. Let's look at this! This is a massive structural risk in the AI infrastructure buildout. We should definitely update our finding on the Nvidia circular vendor-financing loop and systemic domino risk (`nvidia-circular-vendor-financing-loop`), or write a note that focuses specifically on Nvidia's extreme customer concentration, AR credit risk, and the custom silicon substitution threat. Wait, let's look at the existing note: `nvidia-circular-vendor-financing-loop`. Let's update `nvidia-circular-vendor-financing-loop` to include these incredibly precise metrics about accounts receivable concentration (25% and 18% for the top two customers), overall revenue concentration (61%), and the custom silicon substitution scorecard. Let's check if there are other companies we should pull `market_view` for. What about AMD? Let's check AMD. Let's call `market_view(kind=ticker, id=AMD)` to see their current market data and earnings.

This is fascinating! Let's examine AMD's news: - **Riot Platforms signs $9B AI compute deal with Anthropic**: Riot Platforms (a Bitcoin miner) has signed a significant **$9 billion, 20-year compute agreement** with AI company **Anthropic** involving 191 MW of capacity from Riot's Rockdale, Texas facility. This highlights the massive pivot of Bitcoin miners (like Riot Platforms, Applied Digital, CoreWeave originally) into AI compute hosting, and Anthropic's expanding multi-cloud/multi-host strategy. - Let's check AMD's Q2 2026 earnings: Revenue of $11.54B (up 50.1% YoY) and net income of $2.30B. That's strong growth, showing they are scaling their MI300/MI350 chips. Let's do a search on NVIDIA's Q2 2026 earnings date or latest results. Today is August 17, 2026. Nvidia's Q2 fiscal 2027 (which corresponds to Q2 calendar 2026) is typically reported in late August. Let's search when Nvidia's Q2 2026 earnings are expected or if they have been reported.

Nvidia reported Q1 fiscal 2027 in May 2026. This means their Q2 fiscal 2027 (which is Q2 calendar 2026) will likely be reported in late August 2026. Let's search for "NVIDIA Q2 fiscal 2027 earnings date" to find the exact date. Let's do that.

Ah! Nvidia schedules its Q2 fiscal 2027 earnings call for **August 26, 2026**. This is a dated catalyst! Let's register a `one_shot` watch for this dated catalyst. Wait, let's see what we can watch: "NVIDIA Q2 Fiscal 2027 Earnings Release (August 26, 2026)". This is a critical scheduled release. Let's register a watch for it. Let's check if there are any other specific updates we should do. Let's review the notes we need to write: 1. **CoreWeave's Q2 2026 Earnings & Debt Inversion** (`coreweave-liquidity-crisis-and-syndicated-debt`): - We will update the existing slug `coreweave-liquidity-crisis-and-syndicated-debt` to focus on the Q2 2026 results. - Key metrics: Revenue $2.58B (+112% YoY), Net Loss $626M (up from $290M), Operating Loss $49M, and crucially, **Interest Expense of $640M**, which exceeds their operating loss and completely wipes out operating profitability. - Total debt has soared to **$51.61 billion** on the balance sheet as of 2026-06-30 ($35 billion in funded debt). - Capital expenditures are projected at **$35 billion to $39 billion** for 2026, which is an extreme level of spending. - They closed a new **$2.6 billion DDTL 5.5 facility** on August 10, 2026, following the $3.1 billion DDTL 5.0 in May and $8.5 billion DDTL 4.0 in March. - Magnetar Financial (10% owner) has been aggressively selling shares and warrants, offloading **$213.8 million** in stock on August 13, 2026. - This highlights a profound structural risk: CoreWeave's massive capital buildout is funded by extreme high-yield leverage, and their interest payments are already outstripping their operating cash flow, forcing them into a continuous cycle of refinancing and debt accumulation. 2. **Nebius Group's Q2 2026 Earnings & Michael Burry's Short Thesis** (`nebius-meta-megadeal-and-financing`): - We will update the existing slug `nebius-meta-megadeal-and-financing`. - Key metrics: Q2 revenue of $582.3M (up 454% YoY), Q2 capex of **$5.7 billion** (H1 capex is $8.1 billion), with 2026 capex guidance of $20B-$25B. - They raised **$2.8 billion** by selling 12.7 million shares through their ATM program during Q2, and closed a **$775 million secured asset-backed facility** in July. - Michael Burry has disclosed a massive short position in Nebius (NBIS) at $247, alongside Micron (MU) and Oracle (ORCL). - Burry's thesis highlights two critical contradictions: - **Compute "Backwardation"**: Short-term (6-month) contracts command **$40M-$50M per MW**, while midterm (1-3 year) contracts are priced at **$20M-$25M per MW**. This is a classic commodities backwardation signal, meaning customers expect a massive glut/price drop in future compute. - **Depreciation Extension**: Despite this backwardation pointing to rapid economic obsolescence of chips, Nebius extended the depreciation useful life of its servers and network equipment from **four to five years** at the start of 2026. This aggressive accounting shift inflates current operating margins (group EBITDA margin of 41% and Nebius AI EBITDA margin of 50%) but masks the rapid decay of physical assets. - **Enron Comparison**: On August 12, Burry recommended rewatching the Enron documentary *"The Smartest Guys in the Room"*, warning that history is repeating. 3. **NVIDIA's Accounts Receivable Credit Concentration and Custom Silicon Substitution Risk** (`nvidia-circular-vendor-financing-loop`): - We will update the existing slug `nvidia-circular-vendor-financing-loop` (or `nvidia-circular-financing-loop` since both exist, wait! Let's check which one we should update. The read_notes returned two: `nvidia-circular-vendor-financing-loop` and `nvidia-circular-financing-loop`. Let's update `nvidia-circular-vendor-financing-loop` as it has a higher score in our prior findings search, or let's update both if they are identical or just update `nvidia-circular-vendor-financing-loop` and keep it as the main one). - Let's look at the metrics: Nvidia's customer concentration has reached critical levels, with **61% revenue concentration** among its top buyers. - **Accounts Receivable Concentration**: At year-end, the largest AR customer accounted for **25%** of Nvidia's $38.5 billion AR balance, which is roughly **$9.6 billion**! The second-largest customer accounted for another **18%** (roughly **$6.9 billion**). This means just two customers account for **43%** of Nvidia's total accounts receivable, exposing Nvidia to massive credit risk if either of these players experiences a cash crunch or defaults. - **Custom Silicon Substitution**: The largest buyers (hyperscalers like Amazon, Google, Meta, and Microsoft) are the ones aggressively building their own custom silicon alternatives at scale (e.g., Amazon AWS Trainium, Google TPU v7/v8, Meta Iris AI chip, Microsoft custom silicon). Over the past eight quarters, Nvidia's data center revenue has accounted for between 39% and 47% of total hyperscaler capex, peaking at 47% in Q1 2025 and recently standing at 45%. This shows that the share of incremental capex flowing to Nvidia is no longer expanding, signaling that custom silicon and other suppliers are beginning to capture a larger share of the capex pie. Let's review the guidelines for writing notes: - Body shape: No YAML frontmatter — the title is the first `# ` heading and the slug is the slug parameter you pass. - Every substantive finding must carry 1-2 REPRESENTATIVE VERBATIM QUOTES, copied exactly, each kept with the link it came from. - Name a slug for the DURABLE SUBJECT it tracks. - Ground every quantitative claim in the figures the view returned, and cite that view's `/markets` URL. Let's double-check the `/markets` URLs we pulled: - For Nebius: `/markets/NBIS/2026/08/13` - For CoreWeave: `/markets/CRWV/2026/08/17` - For Vistra: `/markets/VST/2026/08/17` Let's write the first note on CoreWeave. We will use the slug `coreweave-liquidity-crisis-and-syndicated-debt`. Let's verify the exact details and write it. Wait! Let's make sure we have the exact quotes and links. For CoreWeave: - CNBC article: `https://www.cnbc.com/2026/08/11/coreweave-crwv-q2-earnings-report-2026.html` - SEC press release: `https://www.sec.gov/Archives/edgar/data/1769628/000176962826000362/coreweave2q26earningspress.htm` Let's write the note.

Now let's write the second note on Nebius Group, Michael Burry's short thesis, compute "backwardation", and the depreciation extension accounting shift. We will use the slug `nebius-meta-megadeal-and-financing`. Let's double-check the sources: - Substack notes: `https://discussion.fool.com/t/my-nebius-q2-2026-notes/125967` - Seeking Alpha article: `https://seekingalpha.com/news/4632100-burry-adds-to-ai-shorts-says-nebius-is-what-the-top-of-a-boom-looks-like` - The Street article: `https://www.thestreet.com/investing/stocks/michael-burry-nebius-bearish-bet` - Nebius Group Market View: `/markets/NBIS/2026/08/13` Let's write the note.

This is spectacular! Now let's write the third note on Nvidia's accounts receivable concentration, customer concentration, and the custom silicon substitution risk. We will use the slug `nvidia-circular-vendor-financing-loop` to expand and update the existing note. Let's double-check the sources: - Daloopa analysis: `https://daloopa.com/blog/analyst-pov/nvidia-customer-concentration-a-big-4-earnings-preview` - Macrotrends receivables: `https://www.macrotrends.net/stocks/charts/NVDA/nvidia/receivables-total` Let's write the note.

This is fantastic! Let's check if there is any other note we should write or update. Wait, let's look at the other prior findings: - `nvidia-circular-financing-loop`: This slug is a duplicate or near-duplicate of `nvidia-circular-vendor-financing-loop`. Since we updated `nvidia-circular-vendor-financing-loop`, let's see if we should write a redirection or if we should just leave it. The instructions say: "Name a slug for the DURABLE SUBJECT it tracks... Before you write, scan your prior findings for the slug that already owns this subject and reuse it". We reused `nvidia-circular-vendor-financing-loop`, which is excellent. Let's check if we can resolve the thread `019feabd-a6f0-75a7-8e81-6af5f9b659bd`: "Monitor Q2 2026 earnings for CoreWeave (CRWV) and Nebius (NBIS) to track debt growth (expected to hit $33B for CoreWeave), revenue trajectory, and competitive pressure from SpaceX's entry into the neocloud market." Yes! This thread/watch has fired and we have written extensive findings on it. Let's resolve the thread `019feabd-a6f0-75a7-8e81-6af5f9b659bd` and record the event. Wait, the list_threads call returned: `{id: "019feabd-a6f0-75a7-8e81-6af5f9b659bd", kind: "watch", title: "CoreWeave and Nebius Q2 2026 Earnings Releases"}` Let's call `record_event` to mark that this watch has fired, and then `resolve_thread` to close the thread. Let's do both. First, let's call `record_event`.

That was successful. Now let's call `resolve_thread` to mark the thread as resolved. Let's do that.

Now let's register a forward-looking watch for NVIDIA's upcoming Q2 fiscal 2027 earnings release, which is scheduled for August 26, 2026. This is a critical scheduled release. Let's call `record_watch`.

This is perfect! Let's check if there are any other watches we should register. What about CoreWeave's next earnings or next DDTL facility? CoreWeave just closed DDTL 5.5. We could monitor the secondary market pricing of DDTL 5.0 (which is an ongoing watch `019f1273-ce52-7485-a89d-8b5383b6d175`). That is already registered, so we don't need to register it again. Let's review the memory. Is there any memory file we should write to? The prompt says: `- Long-term memory: /var/lib/hey-lefty/notes/019e89b4-58b3-7e3b-a0c2-6a68bd3242db/019e89cb-ed9e-7c95-8378-14e9dbf09660/memory/MEMORY.md (write important facts here)` Let's see if we should write any important facts to the memory file. Let's read it first as per the guidelines: "Before modifying a file, read it first. Do not assume files or directories exist." Let's search if there is a tool to read local files. Wait! There are no local file tools in our declarations! The only tools we have are `list_threads`, `record_thread`, `resolve_thread`, `record_watch`, `record_event`, `write_note`, `read_notes`, `market_view`, `web_search`, `web_fetch`, and `finish_cycle`. So we don't have local file tools. We only write findings via `write_note`. This is perfect. Let's do a quick final check of everything we did. - We checked existing threads and watches. - We fetched the market views for Nebius (NBIS), CoreWeave (CRWV), and Vistra (VST). - We researched CoreWeave's Q2 2026 earnings, its massive $51.61B total debt ($35B funded), its new $2.6B DDTL 5.5 facility, its $640M interest expense, and Magnetar's $213.8M stock sales. We updated the note `coreweave-liquidity-crisis-and-syndicated-debt`. - We researched Nebius's Q2 2026 earnings, its $5.7B capex, its ATM share sales of $2.8B, and Michael Burry's massive short position at $247. We detailed Burry's brilliant economic/accounting thesis: compute "backwardation" pricing anomaly (short-term pricing is double midterm pricing) and the aggressive depreciation useful life extension (from 4 to 5 years). We updated the note `nebius-meta-megadeal-and-financing`. - We researched Nvidia's structural concentration risks, including its 61% revenue concentration, its massive accounts receivable concentration (where just two customers hold 43% or $16.5B of its $38.5B total AR balance), and the custom silicon substitution scorecard (Google TPU v8, Amazon Trainium, Meta Iris, OpenAI custom chips). We updated the note `nvidia-circular-vendor-financing-loop`. - We recorded the event for the fired watch `019feabd-a6f0-75a7-8e81-6af5f9b659bd` (CoreWeave and Nebius Q2 2026 Earnings Releases) and resolved it. - We registered a new watch for NVIDIA's upcoming Q2 fiscal 2027 earnings release on August 26, 2026. Everything is perfectly aligned with the research cycle steps and guidelines. We are ready to call `finish_cycle`. Let's do that!

I have successfully completed the research cycle for topic `019e89cb-ed9e-7c95-8378-14e9dbf09660`. Here is a summary of the findings and actions taken: ### 1. **CoreWeave's Q2 2026 Earnings & Relentless Debt Accumulation** * **Finding Updated:** `[[coreweave-liquidity-crisis-and-syndicated-debt]]` * **Core Metrics:** CoreWeave reported Q2 2026 revenue of $2.58B (+112.5% YoY), but its **net loss widened to $626 million** (compared to $290 million in Q2 2025). * **The Interest Expense Trap:** Crucially, CoreWeave's quarterly **interest expense reached $640 million**, exceeding its operating loss of $49 million and completely devouring its operating cash flows. * **Leverage and Capex:** Total debt has skyrocketed to **$51.61 billion** on the balance sheet as of June 30, 2026 ($35 billion in funded debt), representing a **13.6x Debt-to-EBITDA ratio**. Despite this, CoreWeave raised its 2026 capex guidance to **$35 billion to $39 billion**. * **New Debt & Insider Sales:** CoreWeave closed a new **$2.6 billion DDTL 5.5 facility** on August 10, 2026. Following the earnings release, 10% owner Magnetar Financial aggressively offloaded **$213.8 million** in stock on August 13, 2026, alongside warrant sales. ### 2. **Nebius Group's Q2 2026 Earnings & Michael Burry's Short Thesis** * **Finding Updated:** `[[nebius-meta-megadeal-and-financing]]` * **Core Metrics:** Nebius reported Q2 revenue of $582.3 million (+454% YoY) and Adjusted EBITDA of $236.2 million (41% margin). However, its capital expenditures reached a massive **$5.7 billion** in Q2 alone, forcing the company to raise **$2.8 billion** by selling shares through its ATM program. * **The Michael Burry Bearish Bet:** Michael Burry (of "The Big Short") has disclosed a large short position in Nebius (NBIS) at $247, alongside shorts in Oracle (ORCL) and Micron (MU). His thesis highlights two critical structural contradictions: 1. **Compute "Backwardation":** Short-duration (6-month) contracts command **$40M-$50M per MW**, while midterm (1-3 year) contracts are priced at **$20M-$25M per MW**. This near-term spot premium is a classic backwardation signal, meaning customers expect a massive compute glut or rapid chip obsolescence in the near future. 2. **Depreciation Extension:** To flatten operating expenses and boost margins, Nebius extended its server depreciation useful life from **four to five years** at the start of 2026. Burry highlights the glaring contradiction: if future compute pricing is expected to drop sharply, assuming the hardware remains economically useful *longer* is highly aggressive and masks rapid asset decay. ### 3. **NVIDIA's Accounts Receivable Credit Concentration & Custom Silicon Threat** * **Finding Updated:** `[[nvidia-circular-vendor-financing-loop]]` * **Extreme Credit Concentration:** Nvidia's accounts receivable mix is dangerously concentrated. At the end of fiscal 2026, Nvidia's largest AR customer accounted for **25%** of its **$38.5 billion** total AR balance (~$9.6 billion), and the second-largest customer accounted for another **18%** (~$6.9 billion). Just two entities hold **43% ($16.5 billion)** of Nvidia's total accounts receivable, exposing the chipmaker to massive systemic credit risk. * **Custom Silicon Substitution:** Nvidia's "Big Four" hyperscaler buyers represent **61%** of its total revenue, but they are also the exact entities aggressively deploying custom alternatives to displace Nvidia: * *Google* TPU v8 (Sunfish/Zebrafish) is targeting TSMC 2nm and late 2027 deployment, with over half of Google's ML compute running on TPUs. * *Amazon* has secured a 10-year, $100B commitment from Anthropic to run on AWS Trainium custom chips. * *Meta* Iris AI chip is scheduled to enter manufacturing in September 2026. * *OpenAI* has committed to a 10 GW custom silicon program with Broadcom. * Over the past eight quarters, Nvidia's data center revenue share of total hyperscaler capex has compressed from a peak of 47% to **45%**, signaling that custom silicon is beginning to capture a larger slice of the capex pie. ### 4. **Resolved Threads and Registered Watches** * **Resolved Thread:** Closed `019feabd-a6f0-75a7-8e81-6af5f9b659bd` (CoreWeave and Nebius Q2 2026 Earnings Releases) after recording the event. * **Registered Watch:** Created a new `one_shot` watch for **NVIDIA's Q2 Fiscal 2027 Earnings Release on August 26, 2026** to track Blackwell shipment commentary, gross margins, and accounts receivable mix.