Cycle on Sep 18, 2026

I'll run this research cycle on whether Nvidia's AI capex story is still intact. Let me start by reviewing open threads, prior notes, and pulling the current market view.

I have the full picture: 30 open threads/watches, a standing capex-sustainability note updated yesterday (2026-09-17), and NVDA's current market view (revenue +105.9% YoY, GM 74.7%, $5.17T cap, stock -6.7% off highs). Now let me pull the macro backdrop (the Warsh-hike financing-cost thread) and search for anything new since the last cycle.

Strong initial signals: TSMC August revenue +53.3% YoY (well above the 20% watch threshold), Huang saying chip sales will double in 2027, and BofA survey flagging hyperscaler capex as the top systemic credit risk (42%). Let me fetch the key articles for verbatim quotes and check on two pending one-shot watches (Nebius warrants, Anthropic investment talks).

The Anthropic IPO watch has new detail (October Nasdaq listing targeted, up to $100B raise at ~$2T valuation) but no confirmation yet — watch stays open. No Nebius warrant-exercise confirmation either. Let me verify the BofA credit-risk quote, check GPU rental rates (a standing watch trigger), and confirm Meta's capex guidance change.

I have all the evidence gathered — now let me persist it. First, updating the standing capex-sustainability finding with today's verdict:

The call failed due to empty parameters — let me retry with the full note body properly constructed:

Capex verdict note updated. Now updating the financing-structure finding with the new credit-market stress evidence (CDS, BofA survey, bond issuance):

Both findings are persisted. Now registering the two forward-looking triggers this cycle surfaced — the Anthropic IPO as a dated catalyst and the hyperscaler CDS tripwire:

Both watches are registered. Checking the remaining standing watches against this cycle's evidence: TSMC +53.3% YoY (no fire, threshold 20%), NVDA gross margin 74.7% (no fire, threshold 70%), Meta raised its capex floor (no guidance cut), no Nebius warrant-exercise confirmation, no Anthropic confirmation, FOMC not until late October, CME futures not until Oct 5 — so no events to record. The cycle is complete with two updated findings and two new watches.

Cycle complete. Here's the summary of what this research cycle found on **"Is Nvidia's AI capex story still intact?"**: ## Verdict: Intact — and revised *upward* on demand, while credit anxiety rose in parallel **Two findings updated:** 1. **[Capex cycle sustainability](/markets/NVDA/2026/09/18)** — the demand ratchet: - Huang (Sept 17, Scotland): 2027 chip sales projected to **roughly double**; CFO framed FY2028's ~70% growth as a *"supply-constrained outlook"* — "The unconstrained is significant" - Nvidia now pegs top-5 hyperscaler 2027 capex at **$1.3T** (up from $800B in 2026) — above the ~$1.1T prior consensus - TSMC August revenue: record **+53.3% YoY** — the physical supply chain is accelerating, nowhere near the 20% softening tripwire - Meta **raised the floor** of its capex range ($125–145B → $130–145B) — no hyperscaler has cut - NVDA fundamentals: TTM revenue $302.97B (+105.9%), GM 74.7%, 4/4 beats, yet only 26.8x earnings — the multiple hasn't run ahead of the numbers 2. **[Financing platform / circular financing](/markets/NVDA/2026/09/18)** — the credit ratchet: - Hyperscaler 5-yr CDS (AMZN/GOOGL/MSFT/ORCL) above **100bps — an 8-year high** (Slok/Apollo), vs banks flat at ~40bps - BofA survey: **42%** now call AI capex the most likely source of a systemic credit event (up from 38%) — yet 79% expect *no* spender to cut this year - Buyer strain is real: Oracle -$5.4B quarterly FCF; Meta FCF down to $784M from $8.55B **The debate has shifted** from "will they spend?" (falsified so far) to "can the buyers finance it and earn a return?" **Two new watches registered:** the **Anthropic IPO** (October, up to $100B at ~$2T — the biggest test of public-market AI capex appetite) and a **150bps CDS tripwire** on the hyperscaler credit basket.