Cycle on Sep 18, 2026

I'll run this research cycle on whether Nvidia's AI capex story is still intact. Starting by pulling the open threads, prior notes, and the current market views for NVDA and the macro backdrop.

I have the full picture: 40 open watches (mostly future-dated catalysts), standing notes through Sept 17, NVDA at $219.34 (+2.5%, $5.17T cap, 74.7% GM, TTM revenue $302.97B), and a restrictive macro backdrop (10-yr at 5.01%). The last cycle concluded demand is intact and the new test is rates. Now I'll hunt for the freshest evidence (Sept 17–18) on capex demand, the safety-pacing political front, and any cancellation signals.

Strong leads: Huang's "twice as many chips" quote at the King Charles III summit (Sept 17), the pending Trump–Xi state dinner (Sept 24) with Altman and Huang attending, and a Reuters piece on investor nerves over AI spending. Let me read the primary sources and check the capex-discipline side (Microsoft's disclosure, any cancellation signals).

The picture is forming: Huang's "double the chips" declaration (Sept 17) is a major demand-side reaffirmation, while Oracle's job cuts alongside a $95B capex reaffirmation and Microsoft's $175B calendar-2026 capex plan show the funding stress side. Let me check the new Nvidia/Google/Emerald AI energy alliance and the "first brake" market narrative before writing.

The AEMA launch (Sept 16) is a substantive new development on the power-bottleneck front — Nvidia, Google, and Emerald AI plus 18 partners including Anthropic, with the striking claim that power, not capital or silicon, is now the binding constraint. Let me grab one more primary citation for the Sept 24 state dinner attendance before writing up findings.

I have the core evidence. Before writing, I'll firm up the funding-stress numbers (Microsoft's $175B calendar-2026 capex plan, Oracle's $95B reaffirmation amid layoffs) and the CNN sidelines-meeting status so every claim rests on a page I've actually opened.

I have all the evidence I need from this cycle's fetches — three substantive updates to existing findings. Writing them now: (1) the capex story's demand-side reaffirmation plus the funding-stress picture, (2) the safety-slowdown debate's migration to the Sept 24 state dinner, (3) the AEMA power-flexibility coalition as the newest layer of Nvidia's power-bottleneck playbook.

Two notes saved. The third call went out empty — reissuing it properly now for the AEMA / power-bottleneck update.

All three findings are saved. Watch housekeeping: nothing fired this cycle — the FOMC decision lands late Oct/mid-Dec, the Trump–Xi visit is next week (Sept 21–25), the AI-leaders meeting remains unfinalized, and critically, the "first hyperscaler capex cut or order cancellation" tripwire is confirmed still untriggered per Reuters and GuruFocus. The existing watch set already covers every forward trigger this cycle surfaced, so no new threads or watches are needed.

Cycle complete. Here's where the **"Is Nvidia's AI capex story still intact?"** question stands after this cycle: **Verdict: intact on demand — the stress is migrating to financing and rates, not orders.** Three findings were updated (existing slugs, revised in place): 1. **[[big-tech-capex-reaches-710b-in-2026]]** — The demand side got its strongest reaffirmation yet: Huang at the King Charles III AI summit (Sept 17): *"I expect Nvidia to sell twice as many chips as this next year as we do this year."* NVDA closed +2.5% at $219.34 ($5.17T cap, TTM revenue $302.97B, +105.9%). The BofA capex line holds at $795B (2026) → $1.08T (2027). Meanwhile the funding stress is real but shows up as **layoffs and accounting, not cancellations**: Oracle cut 21,000 jobs (13% of headcount) while capex rose 162% to $55.7B with a –$23.7B FCF deficit; Microsoft plans $175B calendar-2026 capex and 38GW by 2032 while stretching leases from 15 to 25 years. GuruFocus: *"no disclosed evidence of major Nvidia customers cancelling chip orders."* The macro test (10-yr at 5.01%) remains the binding constraint. 2. **[[ai-safety-slowdown-debate]]** — Huang took the counter-case to Scotland (*"When a product is not safe, we should hold it back and keep engineering it"* — safety engineering, not pacing), while the political venue converged on the **Sept 24 Trump–Xi state dinner**: Altman and Huang both confirmed attending; a sidelines AI-executives meeting is contemplated but *"nothing has been finalized or scheduled."* Trump's "HOAX" position is unchanged. No pacing commitment has hardened into deferred compute orders. 3. **[[nvidia-lancium-power-investment-and-energy-bottleneck]]** — New development: Nvidia, Google, and Emerald AI launched the **AI Energy Management Alliance** (Sept 16, +18 partners including Anthropic and major utilities), declaring *"the primary limiting factor for deployment is no longer capital or silicon; it is power"* — with a claimed 100GW unlockable on the existing grid. This adds a policy/standards layer to Nvidia's power strategy (DSX Flex software, Lancium, the ~100MW Aurora factory in Manassas). **Watches:** none fired. The falsifier tripwire (first capex cut/order cancellation) is confirmed untriggered. Next catalysts on the calendar: Trump–Xi state dinner (Sept 21–25), then Nvidia's Q3 FY2027 print (Nov 17, $108B revenue guide, ~74% GM).