Cycle on Sep 17, 2026

I'll start by pulling the current state: open threads, prior notes, and the market views for NVDA and the macro backdrop.

I have the state. Key observations: - **Watch to check first**: the FOMC Sept 16 decision (one_shot, fire_date was yesterday) — the macro squeeze was a central part of the prior cycle's verdict. - **Market data**: NVDA at $213.90, -9.1% off its high, with news of "this week's selloff" in AI infrastructure names followed by a rebound. TTM revenue $302.97B (+105.9%), gross margin 74.7%. - **Macro**: 10-yr at 5.0% — long rates elevated. Let me confirm what happened at the FOMC and what drove this week's AI selloff.

Two major developments to confirm and dig into: 1. **The Fed hiked** on Sept 16 (first since 2023, to 3.75–4.00%, unanimous, with more signaled) — this resolves the one_shot FOMC watch. 2. **Monday Sept 14's AI selloff** was triggered by AI leaders (Amodei et al.) calling for a frontier-AI slowdown — a direct escalation of the "pacing debate" from the prior cycle — followed by a rebound as analysts pushed back. Let me read the primary sources.

The picture is forming. The FOMC hiked (first since 2023, unanimous, with more signaled — and notably cited AI capex as an inflation driver), and Monday's selloff was the "pace the frontier" call hitting chip prices, followed by a rebound. There's also a revived **GPU depreciation debate** (Burry's $176B claim vs. Huang's rental-rate rebuttal) — a new angle on capex quality. Before writing, let me check the three outstanding one_shot watches: the Nvidia–Anthropic talks, the DOJ Groq probe, and whether any hyperscaler has guided down capex.

All three watches checked: - **Nvidia–Anthropic talks**: still unconfirmed — Reuters (Sept 11) reports talks for up to $10B anchor investment in a possible $100B IPO at ~$2T valuation; no official confirmation yet, so the watch stays open. - **DOJ Groq probe**: no escalation to enforcement this week — mostly legal analysis; stays open. - **Hyperscaler capex**: no guide-downs — 2027 capex tracking toward $1.1–1.3T with guidance raises across the board. Let me ground the rebound narrative and the hyperscaler capex numbers in full pages, and check the Nebius warrant watch (fire_date Sept 12 passed).

The Nebius warrant watch stays open (exercise "not confirmed" either way). I have the rebound narrative and the hyperscaler capex numbers grounded. One more fetch to properly cite Nvidia's Q3 guide and data-center figures, then I'll write.

All evidence is in hand. Now I'll write the two findings: the updated capex-sustainability verdict (reusing its slug) and a new note on the GPU depreciation debate.

The main note is updated. The second call went out empty — reissuing the GPU depreciation note properly.

Both findings are written. Now recording the FOMC watch event (it fired yesterday) and registering two new forward-looking watches.

The cycle is complete. Here's the summary: ## Cycle summary — Is Nvidia's AI capex story still intact? **Verdict: intact on every hard datapoint — but the market has started trading the *pace* question with real money, and the Fed has joined the debate.** **What happened this week:** 1. **The pacing debate hit the tape.** Amodei's "pace the frontier" essay (echoed by Musk and Altman) triggered Monday's selloff — NVDA -3.4%, Intel -7%, AMD -4.4%, the semiconductor ETF's worst day since July 1 — followed by a two-day rebound as Huang pushed back, Trump called AI concerns "a hoax," and analysts dismantled the "slower frontier AI = fewer chips" logic (inference, enterprise AI, and physical AI keep consuming compute). 2. **The Fed hiked — and named AI capex.** First hike since 2023 (to 3.75–4.00%, unanimous under Warsh), with 16 of 18 members signaling another by year-end. Reuters explicitly listed "capital spending from the artificial intelligence boom" among the inflation drivers. 10-yr Treasury at 5.0%. 3. **No demand cracks anywhere.** No hyperscaler has guided down — 2027 combined capex is tracking to ~$1.12T (Jefferies) vs ~$818B in 2026. Nvidia beat again (Q2 $96.2B, +106%), guided Q3 to ~$108B, gross margin 74.7%, and the stock's multiple compressed (PEG 0.448, -9.1% off high) even as estimates held. 4. **New attack surface: earnings quality.** Burry's revived depreciation critique ($176B understated through 2028; Oracle ~27%, Meta ~21% overstated) vs. the rental-market rebuttal (H100 rates +22% to $3.28/hr; ~95% contract rebooking) — written up as a new finding. **Housekeeping:** resolved the FOMC watch (fired as expected, hawkish follow-through); registered two new watches (the follow-up hike path, and the GPU useful-life/impairment tripwire); the Anthropic-IPO, DOJ-Groq, and Nebius-warrant watches remain open with nothing to fire on.

The research cycle is complete. Two findings were persisted: - **`nvidia-ai-capex-cycle-sustainability`** (updated) — the round-three verdict: fundamentals intact, macro now actively tightening, market trading pace not level - **`gpu-depreciation-debate-hyperscaler-earnings`** (new) — Burry's $176B depreciation attack vs. the rental-market rebuttal, with a standing tripwire on hyperscaler useful-life changes The FOMC watch was resolved as fired, and two new watches will keep the next cycle honest: the follow-up Fed hike and any hyperscaler depreciation-schedule revision.