Cycle on Oct 1, 2026

I'll run a research cycle on whether Nvidia's AI capex story is still intact. First, let me see the open threads and pull the current market view for NVDA.

I have the full watch list and NVDA's current market view. The verdict as of 2026-09-30 was "intact." Now let me check for anything new in the last day or two — IPO developments, rates, financing news — that could change that verdict.

Major developments since yesterday: the 10-year Treasury hit a 24-year high (~5.34%), there's a reported $250B Nvidia guarantee tied to an OpenAI Ohio campus, and Burry has moved to puts on NVDA. Let me dig into the biggest ones.

The Burry piece is substantive — puts struck mid-$100s, ~$1.1B notional, triggered by an Ares report. Now let me find the primary reporting on the $250B OpenAI guarantee and the Micron results.

Micron's results are strong second-order confirmation (HBM sold out, >$100B take-or-pay). Now let me get the details of the $250B OpenAI guarantee story and check whether Nvidia's 10-Q discloses existing guarantee obligations.

Key correction surfaced: the $250B guarantee story is from July — the SEC 8-K (Aug 17) actually capped the Ohio guarantee at $105B. That's context, not new. Let me now read the existing notes and verify the Anthropic S-1 numbers and the 10-year yield move.

The Reuters piece reconciles the Anthropic numbers: $42B net loss includes a ~$34B non-cash charge; operating loss was $8.06B on $4.6B revenue. Now let me verify the Anthropic–SpaceX $84.5B compute deal, check Nscale IPO status (a live watch), and try the White House AI summit piece.

Significant new development: the IPO window is closing for AI issuers — Oura postponed, Nscale's roadshow reportedly expected to slip, and OpenAI has delayed its IPO while seeking $30B privately. Let me get the details.

I have all the evidence gathered and verified. Now writing the four findings — one per theme — then closing the cycle.

Three notes landed; the Micron note call went out empty — resending it with the full body.

All four findings are written. No watches fired this cycle (10-yr at 5.34% is below the 5.5% tripwire; Nscale hasn't officially postponed; OpenAI's round hasn't priced or collapsed; Anthropic's public S-1 is still unreleased), and the IPO-window developments are already tracked by existing watches — so no new threads or watches are needed. Closing the cycle.

Cycle complete. Here's what this cycle found on **"Is Nvidia's AI capex story still intact?"**: **Verdict: intact — but the stress has migrated from demand to the price of money.** **Four findings written:** 1. **[[nvidia-ai-capex-cycle-sustainability]]** — Verdict unchanged (intact), but the pressure moved to the financing leg: the 10-year Treasury hit a 24-year high of ~5.34% (driven by energy inflation, Brent +42% from June lows), now ~16bp below the 5.5% "AI capex crack" tripwire this topic tracks. Michael Burry swapped shorts for ~$1.1B of puts (NVDA Sept-2027, struck mid-$100s vs $229), pulling his break timeline to 2027, citing an Ares report on cancellation clauses. Demand counters: NVDA's $302.97B TTM revenue, 74.7% GM, $150B buyback expansion. 2. **[[ai-ipo-window-equity-exit-channel]]** (new) — The AI IPO window shut in late September: OpenAI delayed its IPO to 2027 and is targeting a $30B private round at ~$1.4T (with $278B cumulative negative FCF projected through 2030); Oura pulled its $2.2B IPO the day before listing; SB Energy (Nvidia-backed, OpenAI-dependent) baulked; Nscale's roadshow is expected to slip; Anthropic moved Oct→Nov. SoftBank is the visible weak link (BB+ rated bonds funding its OpenAI tranche). 3. **[[anthropic-ipo-public-market-test]]** — The S-1 numbers reconciled: the $42B net loss is mostly a $34B non-cash financing charge; the operating loss was $8.06B on $4.6B revenue. New: the SpaceX compute deal doubled to $84.5B — terminable by either party on 90 days' notice, the concrete instance of the cancellation-clause risk. 4. **[[micron-ai-memory-pricing-power]]** — FQ4 confirmed the demand leg: $54.23B revenue (~4.8x YoY), 86%+ margins, HBM3E sold out deep into next year, >$100B take-or-pay contracts. **No watches fired** — the 10-yr tripwire (5.5%), Nscale IPO, OpenAI round, and Anthropic S-1/pricing watches all remain live for future cycles.