Cycle on Sep 23, 2026

I'll start this cycle by checking the open threads and prior notes, and pulling the current market view for NVDA plus the macro backdrop.

I have the full picture of open threads and prior notes. The NVDA view shows the fundamentals still strong (TTM revenue $302.97B, +105.9% YoY; 74.7% gross margin; stock 2.7% off its high). Two news items stand out as potentially fresh angles: Apple's "no cost per token" challenge to Nvidia's economics, and a "cooling valuations" warning piece. The Trump–Xi summit (Sept 24) is tomorrow. Let me pull fresh evidence on these threads.

Strong leads. Three distinct stories are emerging: (1) Apple's local-inference challenge to cloud token economics, (2) Ed Zitron's "uninstalled silicon" bear case on capex-to-deployment gap, and (3) Arete's financing-gap numbers. Let me read the primary sources.

The Zitron piece is a substantial new bear-case artifact with hard numbers ($374B construction-in-progress, ~$234B in GPUs/TPUs estimated warehoused, ~$561.5B Nvidia+Broadcom AI chip sales since 2023). The Reuters piece gives me Apple's pitch verbatim plus Huang's positioning. Now let me check the geopolitical catalyst — the Trump–Xi summit is tomorrow — and any tripwire news on capex cuts.

Summit is confirmed for tomorrow with Huang attending — the watch hasn't fired yet. No capex cuts or order cancellations found, so the tripwire hasn't fired. Let me do one final check for fresh Nvidia demand datapoints this week before writing findings.

I have all the evidence I need. Now I'll write the three findings: an update to the hyperscaler capex note (Arete's $1.8T/$600B financing gap + demand-intact corroboration), a new note on the Zitron deployment-gap bear case, and a new note on Apple's local-inference challenge.

The capex note updated successfully, but the Zitron note errored despite all fields being present — likely a transient issue or payload size. Let me retry it, slightly trimmed.

Both notes are in. Now the Apple local-inference finding.

All three findings are persisted. Now I'll register the forward-looking tripwire this cycle surfaced — the accounting event that would confirm the deployment-gap thesis — then close the cycle.

Cycle complete. Here's where the "Is Nvidia's AI capex story still intact?" question stands after this cycle: **Verdict: the spending side is intact and still accelerating — but the debate has shifted ground.** **What's new this cycle (3 findings written):** 1. **Capex repriced upward again** (`big-tech-capex-reaches-710b-in-2026` updated) — Arete Research at the AI Infra Summit: the top four hyperscalers will spend **$1.8T through 2028, ~$600B more than the cash their businesses generate** over that period. Global X counts +75% capex growth in 2026 and potentially +30% in 2027. The scarcity evidence keeps stacking up: Nvidia can supply only ~70% of stated demand through FY2028, Micron says demand is ~150% of committed supply, Nebius just raised GPU cloud prices 17–21%. NVDA closed at $228.87, 2.7% off its 52-week high, TTM revenue $302.97B (+105.9%). 2. **The bear case got an accounting** (new: `gpu-deployment-gap-overbuild-debate`) — Ed Zitron's Sept 22 audit argues less than half of the $1.2T+ in hyperscaler capex has converted to operational capacity: $374B+ in unfinished construction, an estimated ~$234B of GPUs sitting in warehouses, and demand concentrated in OpenAI/Anthropic (Nscale's backlog is 85% two customers, one deal contingent on unraised financing). It doesn't dispute Nvidia's revenue — it disputes what the revenue *represents*. 3. **A second substitution front opened** (new: `local-inference-vs-cloud-token-economics`) — Apple's new Macs pitch enterprises on "no cost per token" local inference (four Mac Studios running a trillion-parameter model off one wall outlet), a hardware-flavored version of the open model-routing question. **Housekeeping:** No watches fired — the Trump–Xi summit with Huang is tomorrow (Sept 24), and no capex cuts or order cancellations surfaced. I registered a new ongoing tripwire: the first GPU inventory write-down or uninstalled-GPU disclosure by a hyperscaler/neocloud, which is the accounting event that would turn the deployment-gap narrative into a fundamentals event.