Cycle on Jul 18, 2026

I will start by listing the active threads for this topic to see what questions or watches are currently open.

I will fetch the current public-markets view for Nvidia (NVDA) to see its financial fundamentals, valuation, and recent market performance.

I will search the web for recent news and discussions on Nvidia's AI capex, Blackwell/Kyber demand, and Big Tech capex trends in mid-July 2026.

I see some interesting articles about Rubin facing delays and Blackwell driving 70%+ of NVIDIA high-end GPU shipments in 2026, as well as Nvidia expecting upside to $500B in AI demand from Blackwell/Rubin according to Citi, and Power Integrations targeting NVIDIA Kyber AI Data Centers. Let's search more specifically about "Nvidia Kyber", "Blackwell", and "Rubin" to see what the latest news is on these platforms as of July 2026.

The results show a major piece of news that broke in early July 2026: Nvidia's next-generation rack-scale system, the Kyber NVL144, has reportedly been delayed to 2028 (a delay of over 12 months) due to manufacturing difficulties with a key circuit board/complexities. This was first reported by SemiAnalysis. Let's fetch the CNBC article and the Tom's Hardware article to understand the details of the Kyber NVL144 delay, how it affects Nvidia's roadmap, and if the overall AI capex story remains intact despite this.

The CNBC article is highly informative! It explains: - The Kyber NVL144 rack-scale architecture (designed to house 2027 Rubin Ultra chips) has been delayed by more than 12 months to 2028, according to SemiAnalysis (reported on Monday, July 6, 2026). - The setback stems from difficulties manufacturing a key circuit board at the heart of the system: the PCB midplane. - The NVL576 (linking 8 racks) is also likely delayed or limited to small volumes. - Nvidia rejected the report, saying "Our roadmap is intact." - Cloud customers (CSPs / hyperscalers) rejected a backup plan of bolting two current-generation racks together due to its "odd design and heavy operational burden." - This could give rivals AMD and Google a technical opening at the high end. - However, analysts like Paul Triolo say this shouldn't be over-analyzed as affecting the long-term criticality of Nvidia. - Power supply remains the major constraint on AI data centers. - Nvidia's current-generation Rubin systems are in full production and begin shipping this fall (H2 2026) to 8 cloud partners. - SemiAnalysis projects Nvidia's data-center compute revenue will run 20% above Wall Street consensus in the second half of fiscal 2027. Let's read the Tom's Hardware article to see if it has more technical detail or additional context.

Ah, the Tom's Hardware page is behind a paywall or has limited content. Let's search for the SemiAnalysis post itself, or look for details about SemiAnalysis's report on Kyber NVL144. Let's do a web search.

Let's fetch the Yahoo Finance article or Tech Times article, or the network-switch.com article. Let's first fetch the Tech Times article to see what details are provided.

This Tech Times article is absolutely brilliant! It provides a wealth of concrete technical details and quotes: 1. It explains the exact physics of the Kyber NVL144 delay: - "Three separate 26-layer circuit boards, each carrying copper traces no wider than one quarter of a human hair, are laminated together under heat and pressure into a single 78-layer stack." - "orthogonal backplane PCB — a circuit board approximately one square meter in area, built from 78 laminated layers of specialty high-frequency materials, with copper traces no wider than 25 micrometers." - "uses a hybrid of M9-grade copper-clad laminate, quartz fabric, and PTFE — specialty high-frequency materials that only a handful of fabricators worldwide can work with — in a 78-layer stack approximately one square meter in area." - "The copper traces must be held to widths and spacings of 25 micrometers or less, with impedance controlled within a ±5% tolerance across the entire board, to keep 448G-class signals intact across all 144 GPU connections." 2. Backup plan rejected: - "Nvidia had a contingency. Internally, the company developed what SemiAnalysis calls the NVL72x2 design: two existing Oberon NVL72 racks placed back-to-back and linked via copper NVLink, bypassing the PCB midplane problem while delivering roughly comparable GPU density." - "The design was imperfect ... Cloud providers rejected it ... NVL72x2 was canceled 'due to heavy pushback from CSPs and hyperscalers over its odd design and heavy operational burden.'" 3. Larger system stalled: - "The NVL576 system — eight Kyber racks linked by co-packaged optics (CPO) into a 576-GPU cluster — is also now flagged as delayed or limited to low-volume shipments, because co-packaged optics, the technology required to carry NVLink signals between racks at those distances, is itself not yet ready for mass production." 4. Second blow in 8 days: - "On June 30, SemiAnalysis reported that the original quad-chiplet version of the Rubin Ultra GPU — which would have carried 16 HBM4E memory stacks for 1 terabyte of memory per package — had been quietly canceled due to manufacturing constraints that TSMC cannot yet solve. The dual-chiplet version, with approximately 8 HBM4E stacks and roughly half the memory capacity, will ship instead." 5. Financial picture and near-term strength: - "The current Vera Rubin NVL72 systems — built on the existing Oberon rack architecture that is entirely unaffected by the Kyber delay — are in full production and scheduled for delivery this fall to all eight confirmed cloud partners: Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud, CoreWeave, Lambda, Nebius, and Nscale." - "SemiAnalysis, despite being the source of the delay report, is bullish on Nvidia's near-term financials: the firm projects Nvidia's data-center compute revenue will exceed Wall Street consensus estimates by 20% in the second half of fiscal 2027." - "Goldman Sachs analyst coverage notes that Nvidia's forward price-to-earnings multiple of 21.7x is close to the S&P 500 average and well below the company's five-year average of 72x — with AI capital expenditure across the industry projected to grow from $650 billion in 2026 to $1 trillion in 2027." - "Nvidia's fiscal 2026 full-year revenue reached $215.9 billion, up from $26.9 billion in fiscal 2023, and Wall Street consensus for fiscal 2027 is $392.7 billion." - "Mizuho Securities analyst Jordan Klein characterized the SemiAnalysis report as 'headline-grabbing noise,' adding that investors have seen this dynamic before. When Blackwell delay reports surfaced in August 2024, Nvidia rejected them, fixed the design flaw, and shipped several billion dollars of hardware within months." - "Nvidia shares closed Monday at $195.55, up 0.37% on the day... Asian PCB suppliers were more volatile: Guanghua Technology and Nuode New Materials each fell more than 5%... Tianhai Electronics and Welgao Electronics dropped more than 6%." Let's do a search on the broader Big Tech capex and Nvidia's overall position in July 2026 to make sure we have the full story. Let's see if there are any other recent articles or reports on "Nvidia AI Capex Story" or Big Tech AI spend in July 2026.

Let's check the Yahoo Finance article: "Meta, Microsoft, Amazon, and Alphabet are about to spend a shocking amount of money to dominate the AI era". This was published recently. Let's fetch it.

This is an excellent source of data on the Big Tech AI capex scale. Let's summarize the key figures: - Goldman Sachs (analyst Amanda Lynam) expects a combined $5.3 trillion of capex spending for the four largest hyperscalers (Meta, Microsoft, Amazon, Alphabet) from FY 2025 to FY 2030 (up from $4.5 trillion prior to Q1 earnings). - Baseline aggregate capex estimate is $7.6 trillion between 2026 and 2031 across compute, data centers, and power. - Google, Amazon, Microsoft, and Meta collectively plan to allocate $725 billion to capital expenditures in 2026 — up 77% from 2025's $410 billion. - Amazon: projecting $200 billion - Alphabet: targeting $175 billion to $185 billion - Meta: guiding $115 billion to $135 billion - Microsoft: tracking toward $190 billion - The five main hyperscalers (including Oracle) plan to add $2 trillion in AI-related assets to their balance sheets by 2030. Let's do a search on recent Alphabet Q2 2026 earnings or capital raises to see if we can resolve or update the thread about Alphabet's equity raise and earnings. Let's see: "Alphabet Q2 2026 earnings" or "Alphabet equity capital raise". Let's search.

Wow, look at those press releases! - "Alphabet Announces Upsize and Pricing of $84.75 Billion Equity Capital Raise" - "Alphabet Announces Proposed $80 Billion Equity Capital Raise to Expand AI Infrastructure" - "Alphabet Announces Date of Second Quarter 2026 Financial Results Conference Call" Let's search for the date of the Second Quarter 2026 Financial Results Conference Call and details about this $84.75 Billion Equity Capital Raise! Let's run a web search.

This is perfect! Let's fetch the IG article: `https://www.ig.com/en/news-and-trade-ideas/alphabet-q2-2026-earnings-preview-260716`. It was published on July 16, 2026 (very recent, just 2 days ago!). Let's fetch it to see the Q2 earnings date, expectations, and details on the $84.75 billion equity raise.

This is spectacular! Let's examine what we have: 1. Alphabet is reporting Q2 2026 earnings on Wednesday, July 22, 2026, after the US market close. 2. In Q1 2026, Alphabet's capex rose 107% YoY to $35.7 billion. 3. Alphabet's full-year FY2026 capex guidance sits at $180–$190 billion! (This is an enormous increase!). 4. In June 2026, Alphabet completed an upsized $84.75 billion equity capital raise earmarked for AI infrastructure. 5. In addition to that, Alphabet raised around $40 billion via debt this year for general corporate purposes. 6. Alphabet has also guided a further "significant increase" in capex for 2027. 7. Free cash flow fell 47% YoY in Q1 2026 due to the capex-fuelled cash burn. 8. Alphabet is commercializing its custom AI chips (TPUs) by preparing a Blackstone-backed neo-cloud venture that will rent TPU compute directly to AI labs and enterprises from 2027, with hardware sales beginning later in 2026. This directly addresses the thread: - `019f10c8-53dd-729c-926a-27e735d10910`: Monitor Alphabet's Q2 2026 earnings release (expected late July 2026) to assess the initial dilution impact, share count changes, and interest/dividend obligations arising from its upsized $90 billion equity capital raise. Wait, let's look at the actual upsized amount: it was $84.75 billion. The earnings release is on July 22, 2026, which is in 4 days. Since it hasn't happened yet, we can't resolve it completely, but we can write a finding about what to expect and note that it's a critical catalyst coming up next week. Or wait, can we see if we can resolve other threads? Let's look at the other threads: - `019eb93f-9403-7dea-9c9e-9ac0765dd674`: Monitor Meta's potential multi-billion dollar equity offering to fund its $125B–$145B AI capex budget, following Alphabet's $85B template. Let's see if Meta has made any announcements on this. Let's do a web search.

Let's search for "Meta Compute" to see what that initiative is about. This is very interesting! Let's search.

This is incredibly fascinating! Let's examine this: - Meta is launching a new cloud infrastructure business called "Meta Compute" to sell excess AI computing capacity. - This was reported around June/July 2026. - Let's fetch the TechCrunch article: `https://techcrunch.com/2026/07/01/meta-like-spacex-looks-to-turn-excess-ai-compute-into-cash/` to understand Meta Compute and how it affects the specialized GPU cloud market and Meta's capex.

This is incredibly rich context! Let's analyze Meta's capex and "Meta Compute": - Meta is launching a cloud infrastructure business called "Meta Compute" to sell excess AI computing capacity and access to models (pitting it against AWS, Google Cloud, Microsoft Azure, and specialized GPU clouds). - It follows SpaceX (via xAI) which signed similar leases with Anthropic, Google, and Reflection AI for its Colossus 1 data center. - As of Q1 2026, Meta had committed to spending $182.9 billion on AI infrastructure in the coming years, including massive projects in Louisiana and Ohio (expected to come online this year, which Zuckerberg said would be "the size of Manhattan"). - Meta's 2026 capex guidance is raised to $125–$145 billion (as mentioned in the Perplexity snippet). Let's verify the exact 2026 capex guidance for Meta. Let's search for "Meta 2026 capex guidance" to get the exact numbers.

This is absolute gold. Let's summarize Meta's situation: - In Q1 2026 (reported late April 2026), Meta raised its 2026 capital expenditure guidance to $125–$145 billion, up from its prior guidance of $115–$135 billion, citing higher AI infrastructure investment. - Meta has launched a major initiative called "Meta Compute" (officially reported on July 1, 2026) to commercialize its excess AI computing capacity and access to models, led by Santosh Janardhan, Daniel Gross, and Dina Powell McCormick. This turns Meta's colossal $125B–$145B capex burden into a potential cloud revenue stream, competing with AWS, Microsoft, Google, and specialized GPU clouds. Let's check the public-markets view for Meta (META), Microsoft (MSFT), and Alphabet (GOOGL) to ground these figures in real market data. I'll pull the ticker views.

Now I will fetch the market view for Alphabet (GOOGL).

Let's look at the threads again: - `019f10c8-53dd-729c-926a-27e735d10910`: Alphabet Q2 2026 Earnings and Dilution Impact Tracking. (Fire date: 2026-07-25). Since the earnings are on July 22, 2026, which is in the future relative to the current time of 2026-07-18, we cannot resolve this thread yet. - `019eb93f-9403-7dea-9c9e-9ac0765dd674`: Monitor Meta's potential multi-billion dollar equity offering to fund its $125B–$145B AI capex budget, following Alphabet's $85B template. Let's see if there is any announcement of Meta doing an equity raise. The search results show that Meta is funding its capex, but we don't see an official equity raise announcement yet. We see that Meta is launching "Meta Compute" to monetize its excess capacity. Let's do a search on other specialized GPU clouds or neoclouds since we have a thread about: - `019edd49-1ccb-73d3-9fc0-80ae266f32c4`: Nebius Group Q2 2026 earnings and margin trajectory. (Fire date: 2026-08-15). - `019eaeeb-61e2-71a6-b308-0186f59be87f`: CoreWeave quarterly capex tracking. - `019eaeeb-7009-7e9b-9b2f-d6c2f4ec5339`: Nebius 850 MW active power capacity by YE 2026. Let's search for "Nebius" and "CoreWeave" in July 2026 to see if there is any new information or if we can update the specialized GPU clouds finding. Let's do a web search.

Let's fetch the `io-fund.com` article or `saasrise.com` article. Let's try `https://io-fund.com/ai-stocks/nvidia-coreweave-nebius-circular-financing-gpu-boom`. It talks about "Nvidia, CoreWeave, and Nebius: Inside the Circular Financing...". This is highly relevant to specialized GPU clouds. Let's fetch it.

This is an incredibly detailed and high-quality article from Beth Kindig on the neocloud market, particularly focusing on CoreWeave and Nebius. It contains a wealth of concrete facts, quotes, and financial numbers! Let's synthesize the key takeaways: 1. Hyperscaler-Neocloud Partnerships & Commitments: - Microsoft has committed approximately $60 billion to neoclouds (CoreWeave, Nebius, and Nscale). - Meta has committed $35.2 billion to CoreWeave (including a recent $21 billion expansion) and up to $27 billion to Nebius, totaling up to $62.2 billion in neocloud commitments. - Combined, Microsoft and Meta's commitments total up to $122.2 billion (roughly 90% of AWS's TTM revenue!). Factoring in OpenAI and Anthropic, total commitments exceed $145 billion. - This compares to CoreWeave's FY2026 estimated revenue of $12.6 billion and Nebius's expected FY2026 revenue of $3.4 billion. 2. The Accounting Play (Capex Offloading): - Leasing from neoclouds shifts upfront capex outflows to long-term operating expenses (opex) spread over 2031-2032. - This keeps massive AI infrastructure spending off hyperscalers' balance sheets and cash flow statements, protecting their free cash flow (FCF), while the neoclouds absorb the capex and debt. 3. Circular Financing with Nvidia: - Nvidia has made multi-billion dollar investments in CoreWeave and Nebius (including latest $2 billion investments in each, and previous equity stakes). - In Q1 2025, Nvidia held a CoreWeave stake worth $896.7 million. In Q4 2025, Nvidia held a $33 million stake in Nebius. - Nvidia has provided a $6.3 billion financial backstop to CoreWeave: "in instances where [CoreWeave's] datacenter capacity is not fully utilized by its own customers, NVIDIA is obligated to purchase the residual unsold capacity through April 13, 2032." - In turn, CoreWeave and Nebius use their funding to purchase tens of billions of dollars of Nvidia GPUs. 4. Active Power and Capex: - CoreWeave has 3.5 GW of contracted power capacity (1 GW active, targeting 1.7 GW active by the end of 2026). - Nebius has 3.5 GW of contracted power capacity (targeting 800 MW to 1 GW connected/active by the end of 2026). - CoreWeave's FY2026 capex is guided at $31B–$35B (midpoint $33B). In Q1, OCF was $2.98B, capex was $7.7B, FCF was -$4.71B. Cash fell to $2.27B, while debt rose to $24.86B. Net cash is -$22.6B. On June 11, 2026, CoreWeave announced a $3.5 billion senior notes offering. - Nebius's FY2026 capex is guided at $22.5 billion (midpoint). In Q1, OCF was $2.26B, capex was $2.47B, FCF was -$214.9M. Cash sits at $9.37B and debt at $8.45B (net cash $920M). Nebius needs $6.3 billion in additional funding to meet the midpoint of its capex forecast. 5. GPU-Backed Debt (DDTLs): - CoreWeave relies on Delayed Draw Term Loans (DDTLs) backed by GPUs and long-term customer contracts. Its $8.5 billion DDTL 4.0 closed in March 2026 and was the first to receive an investment-grade rating, backed by Meta's contract. DDTL 5.0 closed in May 2026 for $3.1 billion, backed by non-investment grade contracts (resulting in higher interest rates). This is a fantastic update for the specialized GPU clouds finding! Let's see if we can check the active threads to see if they can be resolved or updated: - `019eaeeb-61e2-71a6-b308-0186f59be87f`: CoreWeave quarterly capex tracking. - Yes! We have CoreWeave's Q1 2026 actual capex ($7.7B) and its full-year 2026 guidance ($31B–$35B, midpoint $33B). This shows it is on pace for its full-year guidance. - `019eaeeb-7009-7e9b-9b2f-d6c2f4ec5339`: Nebius 850 MW active power capacity by YE 2026. - The latest article states Nebius is targeting 800 MW to 1 GW of connected/active power by the end of 2026, which is consistent with the 850 MW target. - `019edd49-1ccb-73d3-9fc0-80ae266f32c4`: Nebius Group Q2 2026 earnings and margin trajectory. (Fire date: 2026-08-15). - This is expected in August 2026, so we'll keep it open. - `019ed2f6-f136-7051-a5b0-fa3b26ec02e0`: Nvidia Q2 FY2027 Earnings Release (expected late August 2026). - This is expected on August 26, 2026, so we'll keep it open. - `019f10c8-53dd-729c-926a-27e735d10910`: Alphabet Q2 2026 Earnings and Dilution Impact Tracking (expected late July 2026). - This is scheduled for July 22, 2026, so we'll keep it open. - `019eb93f-9403-7dea-9c9e-9ac0765dd674`: Meta Platforms potential equity raise announcement and pricing. - No official announcement of an equity raise yet, but we have the launch of "Meta Compute" on July 1, 2026, which is a major development in how Meta plans to handle its capex. Let's check the macro backdrop to see if interest rates are discussed in the context of capex. I'll call `market_view` for `macro`.

Great, now let's review the notes we currently have to see what we should update. I will call `read_notes` on the topic to see all standing findings.

Let's examine our standing findings and update those which have new information: 1. `nvidia-record-q1-results-and-trillion-dollar-order-book` (Nvidia's AI Capex Story Remains Intact Supported by Robust Blackwell Pricing and Blockbuster Financial Execution) - This should be updated to address the main brief: "Is Nvidia's AI capex story still intact?" - We now have major news that broke in early July (July 6, 2026): the **Kyber NVL144 rack-scale architecture** (designed for 2027 Rubin Ultra chips) has been delayed by more than 12 months to 2028 due to manufacturing challenges with its 78-layer PCB midplane. - We also have news of the quad-chiplet version of Rubin Ultra being canceled on June 30, 2026, in favor of a dual-chiplet version. - Despite these setbacks, the near-term capex story remains extremely strong: current-generation Rubin NVL72 systems (built on the unaffected Oberon rack architecture) are in full production and ship this fall (H2 2026) to 8 cloud partners. - SemiAnalysis projects Nvidia's data-center compute revenue will exceed consensus by 20% in H2 FY2027. - The delay is a materials-science/yield problem at the frontier of PCB manufacturing, not a design error specific to Nvidia, representing a structural constraint on the entire industry's scale-up roadmap. - We will update this note using its same slug `nvidia-record-q1-results-and-trillion-dollar-order-book`. 2. `big-tech-capex-reaches-710b-in-2026` (Big Tech's AI Capex Reaches Historic Scale as 2026 Infrastructure Sprint Approaches $725 Billion) - We have fresh numbers from Goldman Sachs (Amanda Lynam) forecasting a massive combined $5.3 trillion capex for the 4 largest hyperscalers (Meta, Microsoft, Amazon, Alphabet) from FY2025 to FY2030 (up from $4.5T). - The total 2026 capex projection for these four is now up to **$725 billion** (up from $410 billion in 2025, a 77% increase). - Let's verify the individual budgets: Amazon ($200B), Alphabet ($175B–$185B), Meta ($125B–$145B, recently raised from $115B–$135B), Microsoft ($190B). - Alphabet completed an upsized $84.75 billion equity capital raise in June 2026 to fund its AI infrastructure buildout. - We will update this note using its same slug `big-tech-capex-reaches-710b-in-2026`. 3. `gpu-specialized-clouds-and-strategic-co-investment` (Specialized GPU Clouds Face Competitive Disruption as Meta and SpaceX Reshape the Market) - Let's update this finding with the details of Meta's newly announced "Meta Compute" initiative (July 1, 2026) to sell excess AI capacity, competing directly with specialized clouds. - We have the details of the colossal commitments by Microsoft and Meta to neoclouds: up to $122.2 billion combined (Microsoft ~$60B, Meta up to $62.2B). - We have the details on how leasing from neoclouds allows hyperscalers to offload capex into opex, and the details of CoreWeave's and Nebius's extreme debt burdens and circular financing with Nvidia (including Nvidia's $6.3B backstop for CoreWeave). - We will update this note using its same slug `gpu-specialized-clouds-and-strategic-co-investment`. Let's read the current contents of these three notes first to ensure we write cohesive updates and do not lose prior context. I will fetch the contents of `nvidia-record-q1-results-and-trillion-dollar-order-book` first. Wait, `read_notes` only returned previews. Let's see if we can use a tool to read the full note. Ah, we don't have a direct `read_note` tool with a slug parameter, but we can search or write directly. Let's see if we can use `web_fetch` or if we can just write the updated notes. Wait! Is there a tool to read a specific note? No, we only have `read_notes` which lists the notes with a preview. Let's check `read_notes`'s response to see if the preview is long enough or if there is another way. The `read_notes` tool lists the current set of notes. The preview is just a short snippet. Wait, that is fine. I can write the updated notes directly by keeping the core themes intact and incorporating the new evidence we found this cycle. Let's first write the update for `nvidia-record-q1-results-and-trillion-dollar-order-book`. Wait, let's look at the title of the note: `Nvidia's AI Capex Story Remains Intact Supported by Robust Blackwell Pricing and Blockbuster Financial Execution` We will keep this slug: `nvidia-record-q1-results-and-trillion-dollar-order-book`. Let's construct the markdown body according to the guidelines: - Name and link the actual thing (the story, article, paper, or page). - Carry 1-2 representative verbatim quotes, copied exactly, each kept with the link it came from. - Say what it means. - No YAML frontmatter. First heading is `# <Title>`. - Use slug parameter. - Cite the market view URL `/markets/NVDA/2026/07/18`. Let's carefully write the finding for Nvidia's AI capex story. We will explain how the recent Kyber NVL144 delay to 2028 and the Rubin Ultra quad-chiplet cancellation represent physical and manufacturing bottlenecks, but the near-term capex runway remains fully intact, supported by the Rubin NVL72 ramp this fall and massive hyperscaler demand. Let's write the first note.

Now I will write the update for the second note: `big-tech-capex-reaches-710b-in-2026` (Big Tech's AI Capex Reaches Historic Scale as 2026 Infrastructure Sprint Approaches $725 Billion). Let's review the guidelines again: - Name and link the actual thing. - Carry 1-2 representative verbatim quotes, copied exactly, each kept with the link it came from. - Say what it means. - No YAML frontmatter. First heading is `# <Title>`. - Use slug parameter. - Cite the market view URLs if applicable: `/markets/GOOGL/2026/07/18`, `/markets/META/2026/07/18`. Let's write the update for `big-tech-capex-reaches-710b-in-2026`. We'll include the new Goldman Sachs projection of $5.3 trillion for the top 4 hyperscalers through 2030, the individual 2026 budgets (Amazon $200B, Alphabet $175B–$185B, Microsoft $190B, Meta $125B–$145B), and Alphabet's $84.75 billion equity capital raise. This shows that the AI capex story is not just intact, but expanding to historic proportions. Let's make the tool call.

Now I will write the update for the third note: `gpu-specialized-clouds-and-strategic-co-investment` (Specialized GPU Clouds Face Competitive Disruption as Meta and SpaceX Reshape the Market). Let's review the guidelines again: - Name and link the actual thing. - Carry 1-2 representative verbatim quotes, copied exactly, each kept with the link it came from. - Say what it means. - No YAML frontmatter. First heading is `# <Title>`. - Use slug parameter. - Cite the market view URLs if applicable. Let's write the update for `gpu-specialized-clouds-and-strategic-co-investment`. We'll incorporate: - Meta's newly announced "Meta Compute" initiative (officially reported on July 1, 2026), where Meta plans to build a cloud infrastructure business to sell excess AI capacity, competing directly with AWS, Azure, Google Cloud, and specialized neoclouds. This follows SpaceX/xAI's similar move in May 2026. - The massive scale of hyperscaler commitments to neoclouds (CoreWeave and Nebius) totaling up to $122.2 billion combined (Microsoft ~$60B, Meta up to $62.2B), which represents an order of magnitude higher than the neoclouds' current revenue (CoreWeave FY26 est. $12.6B, Nebius FY26 est. $3.4B). - The underlying financial engineering: leasing allows hyperscalers to shift upfront capex outflows to opex over long-term contracts extending through 2031-2032, protecting their free cash flow statements, while neoclouds absorb the capex. - The rising leverage and debt burdens of neoclouds: CoreWeave's debt rose to $24.86 billion in Q1 2026, and on June 11, 2026, it announced a $3.5 billion senior notes offering. It relies heavily on GPU-backed Delayed Draw Term Loans (DDTLs), such as its $8.5 billion DDTL 4.0. - Circular financing risks: Nvidia's multi-billion dollar equity investments in CoreWeave and Nebius (including latest $2B in each) and its $6.3 billion financial backstop to purchase CoreWeave's unsold capacity through 2032. Let's make the tool call.

This is excellent! Let's check if we can resolve any threads. Let's see: - `019eaeeb-61e2-71a6-b308-0186f59be87f`: CoreWeave quarterly capex tracking. - Description: "Track CoreWeave's quarterly capital expenditures to verify if it remains on pace for its $30B-$35B full-year 2026 guidance." - We have verified that CoreWeave's Q1 capex was $7.7 billion, and its full-year 2026 capex guidance is $31B–$35B (midpoint $33B). This shows it is on pace. Let's resolve this thread! - `019eaeeb-7009-7e9b-9b2f-d6c2f4ec5339`: Nebius 850 MW active power capacity by YE 2026. - Description: "Verify if Nebius achieves its target of 850 MW active power capacity by the end of 2026." - The latest article states Nebius is targeting 800 MW to 1 GW of connected power by the end of 2026, which is consistent with the 850 MW target. Since the end of 2026 hasn't occurred yet, we shouldn't resolve this thread completely, but we can write a watch or keep it open. Wait, let's resolve the CoreWeave thread first using `resolve_thread`.

Let's see if there are any other threads we can resolve. What about the thread: - `019eb93f-9403-7dea-9c9e-9ac0765dd674`: Monitor Meta's potential multi-billion dollar equity offering to fund its $125B–$145B AI capex budget, following Alphabet's $85B template. - No official announcement of an equity raise has been made yet, so we'll keep it open. What about: - `019ed2f6-f136-7051-a5b0-fa3b26ec02e0`: Nvidia Q2 FY2027 Earnings Release (expected late August 2026). - This is expected on August 26, 2026, so it's still open. What about: - `019f10c8-53dd-729c-926a-27e735d10910`: Alphabet Q2 2026 Earnings and Dilution Impact Tracking (expected late July 2026). - This is scheduled for July 22, 2026, so it's still open. Let's register a new watch or update any watches if we found new ones. Wait, let's look at the Kyber NVL144 delay. We have a watch for: - `019f6d77-03de-7a2b-bf6d-aacefbf423de`: Nvidia Kyber NVL144 rack commercial shipment verification. - Description: "Verify if Nvidia begins commercial shipments of the Kyber NVL144 rack-scale systems by late 2027 or if SemiAnalysis's predicted delay to 2028 is borne out." - Fire date: 2027-12-31. - This watch is already registered and tracks exactly what we found. We also have a watch for: - `019e9f7b-8898-71ef-a8fc-916318c1f6ed`: Vera Rubin platform broader commercial shipments in H2 2026. - Description: "Monitor the broader commercial availability and scale shipments of the Vera Rubin platform (including NVL72 racks and Vera CPUs) to hyperscalers like AWS, Google Cloud, and Oracle in H2 2026." - Fire date: 2026-11-01. - This is consistent with our findings that Rubin NVL72 systems are in full production and ship this fall to eight cloud partners. Wait, are there any other new watches we should register? Let's see: we have news of the **AMD Advancing AI 2026** event scheduled for **July 23, 2026**, where Lisa Su is expected to announce its AI CPU addressable market estimate above $200B and next-gen platform details (including the Helios platform with MI455X). This is a critical near-term competitive catalyst! Let's register a watch for this event. - Title: AMD Advancing AI 2026 Keynote and Helios Platform Details - Entity: AMD - Watch type: dated_catalyst - Cardinality: one_shot - Fire date: 2026-07-23T00:00:00Z - Description: Monitor AMD's Advancing AI 2026 event on July 23, 2026, for updates on its AI TAM estimate, its MI455X accelerator, and its Helios platform as a competitor to Nvidia. Let's register this watch.

Let's see if there are any other watches we should register. What about the launch of **Gemini 3.5 Pro** which was scheduled for **July 17, 2026** (yesterday)? The IG article mentioned: "the scheduled 17 July launch of Gemini 3.5 Pro gives management a fresh data point to reference." Let's see if we should register a watch or if we can find any news about it. Since it just happened, we can keep an eye on it or register a watch for user feedback on Gemini 3.5 Pro. Wait, let's look at the launch of **Meta Compute** or **SpaceX's Colossus 1** deployments. We already have the specialized GPU cloud finding updated. Let's review our findings and ensure we have answered the core question: "Is Nvidia's AI capex story still intact?" Yes, we have answered it comprehensively: 1. **Nvidia's AI capex story remains highly robust and intact near-term**, supported by colossal hyperscaler demand and the upcoming autumn ramp of the **Vera Rubin NVL72** platform, which is built on the unaffected Oberon rack architecture. 2. **However, long-term physical and manufacturing constraints have emerged**: - The next-generation **Kyber NVL144 rack-scale architecture** (designed for 2027 Rubin Ultra chips) has been delayed by over 12 months to 2028 due to yielding/manufacturing failures of its 78-layer PCB midplane orthogonal backplane. - The quad-chiplet version of the **Rubin Ultra GPU** was canceled on June 30, 2026, in favor of a dual-chiplet version, halving its originally planned memory capacity. - These setbacks represent a materials-science bottleneck at the physical frontier of PCB manufacturing, not a design error specific to Nvidia, representing a structural constraint on the entire industry's scale-up roadmap. 3. **Big Tech's AI capex continues to grow at historic scales**: - Goldman Sachs raised its combined 2025–2030 capex projection for the top 4 hyperscalers to **$5.3 trillion** (up from $4.5 trillion). - In 2026, these four (Amazon, Alphabet, Microsoft, Meta) plan to spend **$725 billion** (up 77% YoY). - Alphabet completed an upsized **$84.75 billion equity capital raise** in June 2026 to fund its AI infrastructure buildout ahead of its Q2 earnings on July 22, 2026. - Meta raised its 2026 capex guidance to **$125B–$145B** and launched **"Meta Compute"** on July 1, 2026, to commercialize its excess AI capacity and offload balance sheet pressure. 4. **Specialized GPU clouds and neoclouds are seeing massive commitments but face rising leverage**: - Microsoft and Meta have committed up to **$122.2 billion** combined to neoclouds (CoreWeave and Nebius) to shift upfront capex to long-term opex. - CoreWeave's debt rose to **$24.86 billion** in Q1 2026, and it announced a **$3.5 billion senior notes offering** on June 11, 2026, to fund its $31B–$35B 2026 capex. - Circular financing risks are prominent, with Nvidia investing billions in neoclouds and providing a **$6.3 billion backstop** for CoreWeave's unsold capacity, while neoclouds buy tens of billions in Nvidia GPUs. This is a cohesive, evidence-backed, and deeply detailed narrative! All three updated findings have been written, a thread has been resolved, and a new watch has been registered. I am ready to call `finish_cycle`. Let's do it.

I have completed the research cycle for topic `019e95ad-bd47-7cf5-bd57-eb159d26bbbe` ("Is Nvidia's AI capex story still intact?"). Here is a summary of the findings and actions taken this cycle: ### Core Summary: Is Nvidia's AI Capex Story Still Intact? **Yes, Nvidia's AI capex story remains highly intact and robust in the near-to-mid term, though it is colliding with physical and manufacturing limits that are stretching out its long-term product roadmap.** 1. **Near-Term Strength is Fully Intact:** * Nvidia’s current-generation **Vera Rubin NVL72** platform—built on the existing **Oberon rack architecture** that is completely unaffected by recent delays—is in full production and scheduled to begin shipping this fall (H2 2026) to eight confirmed cloud partners (AWS, Microsoft Azure, Google Cloud, Oracle Cloud, CoreWeave, Lambda, Nebius, and Nscale). * Despite being the source of recent delay reports, semiconductor research firm **SemiAnalysis** remains highly bullish on Nvidia's near-term financials, projecting that Nvidia's data-center compute revenue will exceed Wall Street consensus by **20%** in the second half of fiscal 2027. * Financially, Nvidia operations are at an unprecedented scale with TTM revenue of **$253.49 billion** (+85.2% YoY), gross margins of **74.1%**, and operating margins of **65.6%**. Wall Street consensus projects fiscal 2027 revenue at **$392.7 billion** (up from $215.9 billion in fiscal 2026). 2. **Long-Term Physical & Manufacturing Constraints Have Emerged:** * **Kyber NVL144 Delayed to 2028:** On July 6, 2026, SemiAnalysis reported that Nvidia's next-generation **Kyber NVL144 rack-scale architecture** (designed for 2027 Rubin Ultra chips) has been delayed by more than 12 months to **2028**. The setback stems from manufacturing failures of its **78-layer orthogonal backplane PCB midplane** (which laminates three 26-layer boards together). Achieving the consistent signal integrity required for 448-gigabit-per-second GPU-to-GPU NVLink communication across a 78-layer board has exceeded what PCB fabricators can currently deliver at viable yields. * **Rubin Ultra Quad-Chiplet Canceled:** On June 30, 2026, SemiAnalysis reported that the original quad-chiplet version of the Rubin Ultra GPU (which would have carried 16 HBM4E memory stacks for 1 terabyte of memory per package) was canceled due to TSMC packaging constraints. Nvidia will instead ship a dual-chiplet version with approximately 8 HBM4E stacks, halving the originally planned memory capacity. * **Backup Plan Rejected:** Cloud service providers rejected Nvidia's stopgap "NVL72x2" design (linking two standard NVL72 racks back-to-back) due to its "odd design and heavy operational burden," leaving Nvidia with no proven solution to expand the scale-up world size for Rubin Ultra in 2027. 3. **Big Tech's AI Capex Continues to Grow at Historic Scales:** * Goldman Sachs (analyst Amanda Lynam) raised its combined 2025–2030 capex projection for the top 4 hyperscalers (Amazon, Alphabet, Microsoft, Meta) to **$5.3 trillion** (up from $4.5 trillion). * In 2026, these four giants plan to spend **$725 billion** on capex (a 77% YoY increase from 2025's $410 billion). Amazon is targeting **$200 billion**, Alphabet **$175B–$185B**, Microsoft **$190 billion**, and Meta **$125B–$145B**. * **Alphabet's $84.75B Equity Raise:** In June 2026, Alphabet completed an upsized **$84.75 billion equity capital raise** earmarked specifically for AI infrastructure, ahead of its Q2 earnings call scheduled for **July 22, 2026**. * **Meta's "Meta Compute" Launch:** On July 1, 2026, Bloomberg reported that Meta is launching a cloud infrastructure business called **"Meta Compute"** to sell excess AI capacity and model access, helping offload its colossal capex burden into a potential revenue stream. 4. **Specialized GPU Clouds Face Rising Leverage & Circular Financing Risks:** * Microsoft and Meta have committed up to **$122.2 billion** combined to neoclouds (CoreWeave and Nebius) to shift upfront capex to long-term opex. * CoreWeave's debt rose to **$24.86 billion** in Q1 2026, and on June 11, 2026, it announced a **$3.5 billion senior notes offering** to fund its $31B–$35B 2026 capex. * Nvidia has provided a **$6.3 billion financial backstop** to purchase CoreWeave's unsold capacity through 2032 while investing billions in neocloud equity, creating a highly scrutinized cycle of circular financing where neoclouds use this capital to buy tens of billions in Nvidia GPUs. --- ### Actions Taken This Cycle * **Updated Finding [`nvidia-record-q1-results-and-trillion-dollar-order-book`]:** Focused on Nvidia's near-term strength in the Rubin NVL72 ramp this autumn, contrasted with the materials-science bottleneck of the Kyber NVL144 delay to 2028 and the Rubin Ultra quad-chiplet cancellation. * **Updated Finding [`big-tech-capex-reaches-710b-in-2026`]:** Integrated Goldman Sachs' updated $5.3 trillion capex projection, the $725 billion collective 2026 spend, and Alphabet's $84.75 billion equity capital raise ahead of its Q2 earnings. * **Updated Finding [`gpu-specialized-clouds-and-strategic-co-investment`]:** Detailed Meta's "Meta Compute" launch, the $122.2 billion in combined Microsoft/Meta neocloud commitments, the accounting play of capex offloading, and the debt profiles/circular financing risks of CoreWeave and Nebius. * **Resolved Thread [`019eaeeb-61e2-71a6-b308-0186f59be87f`] (CoreWeave quarterly capex tracking):** Verified CoreWeave's Q1 capex of $7.7 billion and its full-year 2026 capex guidance of $31B–$35B (midpoint $33B). * **Registered Watch [`019f729f-b064-7b27-adec-815ee0193bf4`] (AMD Advancing AI 2026 Keynote):** Added a one-shot dated catalyst for **July 23, 2026** to monitor AMD's upcoming keynote, where it is expected to raise its AI TAM estimate above $200B and detail its MI455X accelerator and Helios platform.