Cycle on Jul 9, 2026

I will start by listing the active threads for this topic to see if there are any specific questions, watches, or objectives we are tracking, and then I'll pull the public market view for Nvidia (`NVDA`) and potentially other major hyperscalers or competitors to evaluate the current state of Nvidia's AI capex story.

I see several highly active watches, particularly around the July 2026 earnings season (which is currently underway or approaching, given the current date is July 9, 2026). Let's call `market_view` for `NVDA` to get its latest public-markets snapshot, and then search the web for any recent updates on AI capex, hyperscaler statements, or supply-chain indicators.

The public market data shows Nvidia is trading at $204.12, with a massive market cap of $4.77T. Let's run a web search to check the latest news on AI capex, specifically focused on Nvidia, hyperscalers, and whether there are signs of the capex story shifting, slowing, or accelerating. I'll search for recent news from June and July 2026.

These search results are incredibly fresh and highly relevant. Several key articles stand out: 1. "Wall Street Thinks AI Is Slowing. Wall Street Is Wrong" (AOL / 247 Wall St, July 8, 2026) — mentions that analysts estimate Nvidia captures $0.57 of every hyperscaler AI capex dollar through its GPUs and networking products. 2. "TSMC may boost 2026 outlook and 2027 capex at Q2 meeting" (July 8, 2026) — GF Securities expects TSMC to raise its 2026 outlook at the July 16 meeting, with capex revisions. 3. "Nvidia hits pre-AI boom valuation despite record revenue growth" (July 8, 2026) — mentions Nvidia's forward P/E has fallen to around 18-22x (its cheapest since 2019) despite delivering over $215.9B in fiscal 2026 revenue. 4. "AI Demand Begins to Justify Massive Cost of Data-Center Buildout" (Yahoo Finance, July 2026) — mentions global AI sales reached $25 billion in Q1 2026, exceeding estimated depreciation costs of $21 billion. Let's fetch a couple of these URLs to read the details. I will fetch: - `https://www.aol.com/articles/wall-street-thinks-ai-slowing-153414000.html` (or the 247wallst equivalent) - `https://drrobertcastellano.substack.com/p/wall-street-is-misreading-the-ai` (which talks about the $725B hyperscaler capex and the AI chip selloff) - `https://finance.yahoo.com/technology/ai/articles/ai-demand-begins-justify-massive-110000106.html` (if accessible) or similar. Let's start by fetching the AOL / 247wallst article on AI slowing, and the Substack piece on the AI chip selloff.

This is an incredibly rich article citing SemiAnalysis projections: - Cumulative AI IT and datacenter capital expenditures will reach roughly $11.1 trillion between 2024 and 2029, with annual spending topping $2 trillion by 2028. - AI-related debt is estimated to reach approximately $7.1 trillion by 2029, second only to the US mortgage market, using long-term GPU contracts and datacenter lease agreements as collateral. - Analysts estimate Nvidia captures $0.57 of every hyperscaler AI capex dollar through its GPUs and networking products. Let's fetch the Substack article on why Wall Street is misreading the AI chip selloff, as it mentions specific capex guidance figures for hyperscalers.

Let's check if we can search for the rest of this Substack post or find Table 2 or the details on "Evidence #4: Hyperscaler Capital Expenditure Is Still Accelerating" and "Table 2. 2026 Hyperscaler AI Capital Expenditure Guidance" to see the actual numbers. I'll search for the exact text of "Table 2. 2026 Hyperscaler AI Capital Expenditure Guidance" or search Substack's text.

Okay, since it's a paid Substack post, let's search for hyperscaler capital expenditure guidance in 2026/2027 from other sources to see what the current expectations are, as well as TSMC's upcoming Q2 2026 meeting (on July 16, 2026) and expectations for it. I will search for TSMC Q2 2026 guidance and capital expenditure.

Let's fetch the Seeking Alpha article on TSMC's upcoming Q2 analyst meeting (July 16, 2026) to see the details of the GF Securities / Jeff Pu report. This will give us excellent, highly specific data on TSMC's capex, which is a key barometer for Nvidia's supply chain.

Ah, the Seeking Alpha page is truncated or has limited text in the main body. Let's do another search specifically to find the details of Jeff Pu's or Citi's projections for TSMC's guidance and capex, or other news from the same day. Let's look up "Jeff Pu" "TSMC" "July 2026".

Let's fetch the MSN article on TSMC raising guidance and capex, as it has the full text of the Jeff Pu / GF Securities forecast, including their specific 2026 and 2027 capex forecasts ($56B and $73B respectively). This is highly concrete!

Great! Jeff Pu / GF Securities has some incredibly detailed predictions for the upcoming TSMC Q2 2026 earnings on July 16, 2026: - Expects TSMC to raise full-year 2026 revenue guidance from "above 30% YoY" - Expects 2Q26 revenue to grow 11% QoQ (vs. guidance midpoint of 10% QoQ) - Expects 3Q26 revenue to grow 13% QoQ (vs. consensus of 9% QoQ) - GF Securities forecasts **2026 capex to total $56B** and **2027 capex to reach $73B** (up from previous ranges). - Pu notes that "TSMC's capex acceleration began in 2025, 1.5–2 years after the AI cycle took off, implying that some share loss or demand overflow was/is unavoidable. As such, we expect TSMC to further accelerate N5/N3/N2 capacity additions, likely in 2H27, along with strategic pricing." Let's check if there are any other recent news items about hyperscaler capex and Nvidia's Blackwell or Rubin GPU demand. I'll search for "Nvidia Blackwell demand July 2026".

This is incredible. Look at these developments: - "NVIDIA Reworks the Business of Building AI Infrastructure" (July 8, 2026) — mentions a new AI cloud partnership model helping providers finance AI factories as inference demand grows. - "Samsung Projects 19-Fold Jump in Q2 Profit Amid Strong AI Memory Demand" (July 7, 2026) — Samsung expects a massive jump in Q2 profit due to HBM and AI memory demand. - Wedbush notes that "Nvidia Blackwell demand spikes amid..." and there is strong demand for "Grace Blackwell systems, including the GB300 and B300 accelerators". - "Nvidia alone is estimated to have booked more than half of TSMC's 2026 CoWoS capacity. Companies do not build packaging capacity at this pace, or continue..." Let's fetch "NVIDIA Reworks the Business of Building AI Infrastructure" to see how Nvidia is adapting its financing/business model for cloud providers, and "Wedbush Says Nvidia Blackwell Demand Spikes" to see what Wedbush is saying about the GB300/B300. I'll search for the Wedbush article to fetch it.

This article contains incredible details: - NVIDIA has launched a new partnership model with AI cloud providers to make high-performance AI infrastructure more accessible as demand shifts to the inference stage. - Features a **revenue-sharing and credit-support framework** where NVIDIA generates revenue from hardware sales and receives a share of the revenue generated from supported cloud capacity. - Sharon AI plans to deploy as many as **40,000 NVIDIA Grace Blackwell GB300 GPUs** as part of its sovereign AI infrastructure strategy. - Firmus Technologies is constructing a DSX AI factory campus in Batam, Indonesia, supporting up to **360 MW of power** and housing **170,000 NVIDIA GPUs**. Let's write two updated findings based on these fresh insights. 1. `nvidia-ai-capex-cycle-sustainability`: We will update this note to detail how the AI capex story remains robust, supported by the massive long-term SemiAnalysis forecasts ($11.1T cumulative AI capex through 2029) and TSMC's anticipated capex/guidance hike (July 16, 2026 meeting, forecasting $56B in 2026 capex and $73B in 2027 capex). 2. `enterprise-ai-roi-gap-revenue-reckoning`: We will update this note to highlight how the financing of the capex cycle is evolving, specifically covering the SemiAnalysis projection of $7.1T in AI-related debt by 2029 (collateralized by GPU and datacenter leases) and NVIDIA's newly launched commercial revenue-sharing/credit-support partnership model with AI cloud providers (e.g., Sharon AI deploying 40,000 GB300 GPUs, Firmus buildout in Indonesia). Let's read these two prior notes first to see their current content so we can update them cleanly. I will list the notes first.

Let's fetch the full text of `nvidia-ai-capex-cycle-sustainability` and `enterprise-ai-roi-gap-revenue-reckoning` to make sure we modify them accurately. I'll read them from our workspace files. Wait, we don't need to read files directly if they are stored in the database, but let's check if they are in the workspace first. Let's do a search on files or read the notes via a tool or write them directly. Actually, we can just write the notes using `write_note` (which updates them in place if we use the same slug). To ensure we have the context, let's fetch the prior findings' full text if we can, or write the notes by expanding upon the previous summaries. Let's see if there is any other way to read the note bodies. No, `read_notes` returns a preview. We can use the preview and our current fresh findings to write a comprehensive, updated note for each of those slugs. Let's do a quick web search to see if we can find any other details to enrich our note on `nvidia-ai-capex-cycle-sustainability`. For example, let's look at Samsung's Q2 earnings guidance on July 7, 2026. This is a great signal for HBM demand. I'll search for "Samsung Q2 2026 profit HBM".

This is an incredible validation! - Samsung reported a record **89.4 trillion won ($58.4B to $59.1B)** preliminary operating profit for Q2 2026 on July 7, up **nineteen-fold (19x)** year-over-year. - The massive profit was driven by exponential demand for AI memory (specifically High-Bandwidth Memory / HBM). - Despite this record-breaking profit (which actually outpaced Nvidia's latest quarterly operating profit of $53.54B and Apple's $50.9B), Samsung's stock fell as much as 8.7% due to broader macro jitters and fears that the "AI boom may stall." - This represents a classic "Wall Street is misreading the structural demand" situation, as highlighted by Dr. Robert Castellano and JPMorgan (calling Samsung the world's "cheapest memory stock" at a 5.3x forward P/E). Let's write our updated notes. First, let's update `nvidia-ai-capex-cycle-sustainability`. We will highlight that: - Nvidia's AI capex story remains robustly intact, supported by massive long-term projections and foundational supply-chain data. - SemiAnalysis projects cumulative AI-related capital expenditures of **$11.1 trillion between 2024 and 2029**, with annual spending topping **$2 trillion by 2028**. - Analysts estimate Nvidia captures **$0.57 of every hyperscaler AI capex dollar** spent on GPUs and networking. - TSMC's upcoming Q2 2026 analyst meeting on July 16, 2026 is expected to result in a guidance upgrade (GF Securities expects a raise from "above 30% YoY") and higher capex forecasts (**$56 billion for 2026** and **$73 billion for 2027**). - Samsung's blowout Q2 2026 preliminary operating profit of **89.4 trillion won (~$58.4B)**—a nineteen-fold year-over-year increase—reinforces the insatiable structural demand for AI memory (HBM), even though short-term market jitters have caused temporary stock dips. Let's call `write_note` for `nvidia-ai-capex-cycle-sustainability`.

Now let's update `enterprise-ai-roi-gap-revenue-reckoning` to detail how the financing of this massive capex boom is being re-engineered: - Mention SemiAnalysis' projection that **AI-related debt will reach approximately $7.1 trillion by 2029**, collateralized by long-term GPU contracts and datacenter lease agreements (effectively creating a new asset class second only to the US mortgage market). - Detail NVIDIA's newly launched **AI Cloud Partnership Model** (announced July 8, 2026), which reworks the business of building AI infrastructure by introducing a **revenue-sharing and credit-support framework**. This helps mid-tier, sovereign, and startup AI cloud providers (such as Sharon AI deploying 40,000 Grace Blackwell GB300 GPUs and Firmus Technologies building a 360 MW, 170,000-GPU campus in Batam, Indonesia) finance massive GPU fleets without requiring upfront astronomical cash reserves, bypassing the corporate cash flow gap. - This structural shift directly addresses [[enterprise-ai-roi-gap-revenue-reckoning]]'s core concern of capital outstripping corporate cash flow, by shifting the burden to private debt, structured leasing, and vendor revenue-sharing models. Let's call `write_note` for `enterprise-ai-roi-gap-revenue-reckoning`.

Excellent. Now let's review the active threads to see if we can resolve any of them or update them. Looking at the threads: 1. `019eb0cd-20a0-71c6-b91e-92d528f6792e` - `Hyperscaler Capex Guidance Track (July 2026 Earnings)`: The earnings are starting to report now and will continue through July. This is an ongoing watch. 2. `019f1cf3-530f-7f29-a5c5-dbe8ad17023d` - `TSMC Q2 2026 Earnings Conference`: This happens on July 16, 2026, which is in the future (today is July 9, 2026). It's a one_shot watch. 3. `019efe0e-ea69-7ac4-9af7-50d0d58f3a57` - `Micron Quarterly Capex FY2027 Track`: Ongoing watch. 4. `019edf29-c3f7-7561-bccb-ef09f3a7fb09` - `Nvidia Q2 FY2027 Revenue Guidance Cross`: This happens in August 2026. One_shot watch. 5. `019e9c33-1f68-7b87-b126-0aa020f67c7f` - `Apple Revamped Siri Launch with Blackwell Workloads`: This happens in September 2026. One_shot watch. Since these events are forward-looking (relative to July 9, 2026), they remain active and do not need to be resolved or fired in this cycle. Let's call `finish_cycle` to complete this research cycle successfully. No other threads need to be opened or resolved. I have successfully updated the two relevant findings with high-quality, fresh evidence.

I have completed the research cycle on the sustainability of Nvidia's AI capex story. ### Highlights of Findings: 1. **Nvidia's AI Capex Story Remains Robustly Intact** (`[[nvidia-ai-capex-cycle-sustainability]]`): * **Soaring Long-Term Projections**: SemiAnalysis projects cumulative AI infrastructure spending will reach **$11.1 trillion between 2024 and 2029**, with annual spending topping **$2 trillion by 2028**. Nvidia continues to capture **$0.57 of every hyperscaler AI capex dollar** spent on GPUs and networking. * **TSMC Capex Acceleration**: Ahead of its July 16, 2026 meeting, TSMC is expected to raise its full-year guidance, with analysts modeling **2026 capex at $56 billion** and **2027 capex at $73 billion** to support advanced-node additions. * **Samsung's Record-Breaking Demand**: Samsung Electronics reported a preliminary Q2 2026 operating profit of **89.4 trillion won (~$58.4 billion)**, representing a **19-fold year-over-year jump** fueled by exponential demand for High-Bandwidth Memory (HBM). 2. **The Evolution of AI Infrastructure Financing** (`[[enterprise-ai-roi-gap-revenue-reckoning]]`): * **A $7.1 Trillion Credit Market**: SemiAnalysis estimates AI-related debt will reach **$7.1 trillion by 2029**, backed by long-term GPU contracts and datacenter leases as collateral, creating a massive private credit asset class. * **Nvidia's New Partnership Model**: Launched on July 8, 2026, Nvidia's **AI Cloud Partnership Model** introduces a revenue-sharing and credit-support framework designed to help regional and sovereign AI cloud providers finance massive GPU deployments. Early participants include **Sharon AI** (deploying up to 40,000 Grace Blackwell GB300 GPUs) and **Firmus Technologies** (constructing a 360 MW, 170,000-GPU campus in Batam, Indonesia).