Cycle on Sep 23, 2026

I'll start this research cycle by checking the open threads, prior notes, and pulling the market data for Nvidia.

The threads and notes give me a clear picture: the verdict as of yesterday is "intact, but the debate is about financing." Today is Sept 23; the Trump-Xi dinner watch fires tomorrow (Sept 24). Let me pull the macro backdrop and scan for the freshest news since the last cycle.

Macro backdrop pulled (10-yr at 4.96%, easing off 5.01%) and I have several fresh leads: Nvidia's forward P/E at a decade low, bond investors demanding higher yields from Alphabet/Meta/Nvidia, and Burry's "70% of AI revenue is two companies" warning. Let me fetch the substantive pieces.

Two strong themes confirmed: a new "funding tax" in the bond market for AI debt, and Nvidia's forward P/E collapsing to a decade-low ~17x despite record fundamentals. Now let me check the watch items — TSMC monthly revenue and the Nebius warrant exercise — and get a citable source on the valuation story.

TSMC's August revenue came in at +53.3% YoY (watch is nowhere near firing — good counter-evidence), and the Nebius price-hike story is relevant. Now let me get the Burry "two companies" warning and the Morningstar bull rebuttal for balance.

The picture is rich. Before writing, let me verify the TSMC figure and the Morgan Stanley ROIC rebuttal with actual page fetches so every claim I cite is grounded.

All evidence is confirmed. No watches fired this cycle (TSMC is at +53.3% YoY vs the <20% trigger; the Trump-Xi dinner is tomorrow; the Anthropic S-1 isn't out). Now I'll write the findings — one per theme: the NVDA de-rate, the bond-market funding tax, the Eisman concentration warning, plus updates to the standing verdict and depreciation-debate notes.

My apologies — I sent an empty call. Let me write the notes properly now, starting with the two new subjects.

Both new subjects are recorded. Now the three updates to standing findings: the financing note (funding tax), the verdict note, and the depreciation-debate note (Nebius H100 price hikes).

Now the main verdict note update, synthesizing this cycle's layers.

All five findings are written. Now let me register the two forward-looking watches this cycle surfaced: the de-rate tripwire and the OpenAI funding-round catalyst.

The cycle is complete. Here's a summary of what I found: ## Cycle summary — Is Nvidia's AI capex story still intact? **Verdict: substance still intact, but the market's vote split this week.** Five findings written (2 new subjects, 3 updates): **New findings:** 1. **`nvda-derating-decade-low-multiple`** — NVDA trades under 17x forward earnings, a decade low and half its 2025 multiple, despite ~90%+ expected profit growth. It's up ~22% YTD while the SOX is up ~76% — near the bottom of the chip index. The market is discounting *durability* (memory-cost margin squeeze to <72% by Q4, custom silicon, financing circularity), not current earnings. Huang's response: "the world's first and only growth value stock." 2. **`ai-customer-concentration-openai-anthropic`** — Eisman's Sept 21 warning: "70% of AI revenue of the hyperscalers is just from OpenAI and Anthropic," with OpenAI "in trouble." The crux: the $250B OpenAI-Azure RPO behind Microsoft's capex (MSFT FCF -23% YoY). **Updates:** 3. **`nvidia-infrastructure-leasing-circular-financing`** — the bond market now charges an explicit "funding tax": Alphabet spreads 56→71bps as deal size scaled; META/NVDA $4B deals +15-19bps wider; hyperscaler debt +92% YoY vs +4% for the rest of IG; $5.5T total funding requirement vs ~$1.4T covered organically. 4. **`gpu-depreciation-debate-hyperscaler-earnings`** — the strongest rebuttal yet to Burry: Nebius is *raising* prices on four-year-old H100s in October, and CoreWeave is contracting at higher prices. 5. **`nvidia-ai-capex-cycle-sustainability`** — verdict refresh: TSMC August +53.3% YoY (4th straight record), ASE capex to $10.5B, 10-yr eased to 4.96%, NVDA fundamentals clean (TTM $303B, +105.9%, 4/4 beats). **Watches:** registered two new ones — NVDA forward P/E <14x tripwire, and OpenAI's next funding round pricing (or collapsing) at ~$1.5T talk. Nothing fired this cycle: TSMC is nowhere near the <20% trigger, the Trump-Xi dinner is tomorrow, and the Anthropic S-1 isn't out yet.

The research cycle is complete. The topic's findings now reflect the latest state as of Sept 23: **the capex story remains intact on every fundamental layer — demand, physical supply, and Nvidia's own financials — but this week produced the first hard evidence that markets are pricing skepticism rather than facts**: a decade-low forward multiple on the equity side, an explicit "funding tax" on the bond side, and a sharpened concentration debate around OpenAI's $250B Azure commitment. The two new watches (forward P/E <14x, OpenAI round pricing) will catch it if either market's skepticism proves justified in the next cycle.