AI Sentiment & Infrastructure.
American AI media coverage is normalizing while Chinese AI media coverage is ascendant — and U.S. AI data-center cancellations are the leading capex signal. Between January 2024 and May 2026, twelve named data-center projects were cancelled, delayed, or scaled back — Microsoft, Meta, AWS, Oracle, OpenAI / Stargate, CoreWeave, Crusoe — pulling back roughly 5.25 gigawatts of planned U.S. AI infrastructure. In the same window, Chinese AI capex, fundraises, and model launches landed every four to eight weeks. This paper measures both signals on a single dated scoreboard, runs a calibrated sentiment index across the U.S. and Chinese press of record, and logs falsifiable forecasts on the rolling 13-week delta. The current live prediction: the rolling 13-week U.S. AI sentiment index will fall to ≤ −0.075 by 6 July 2026.
What we've actually found.
U.S. hyperscalers have walked away from roughly five gigawatts of planned AI build.
Twelve publicly-disclosed Microsoft, Meta, AWS, Oracle, OpenAI / Stargate, CoreWeave, and Crusoe cancellations, delays, and scale-backs across two years — pulled into a single dated ledger with every row sourced to a primary disclosure or press of record.
Even a simple word-list scorer recovers the U.S.-normalizing, China-ascendant arc.
A 100-word bilingual lexicon agrees with human coders on about seven out of ten articles and reproduces the block ordering — U.S. peak bull through normalization through anxiety; Chinese slow-build through confidence through ascendant — on the calibration set.
Chinese AI capex, fundraises, and model launches land every four to eight weeks.
Fifteen disclosed events across the symmetric ledger — Alibaba's three-year ~USD 53 B cloud-and-AI commitment, ByteDance's reported USD 20 B 2025 spend, MOST East-Data-West-Compute II, the DeepSeek V3 / R1 / R2 cadence, and the provincial 14-FYP supplements.
The full U.S. DC cancellation ledger, the symmetric Chinese capex tracker, the rolling 13-week sentiment series with its three legible inflection points, the 60-article hand-coded calibration set, the live prediction with its falsification rule, the methodology, and the editorial guardrails all live inside the v0 founding paper.
Published pieces.
Each card is a standalone read — a substantive finding or a release. Methodology hygiene, audit notes, and engineering scaffolds live in the Changelog below.
A 200-word lexicon already sees the US-vs-China AI sentiment gap.
We ran a small bilingual word-list scorer end-to-end on the 60-article calibration set. It agrees with our human coders on roughly seven articles out of ten, and reproduces the headline arc — U.S. coverage normalizing, Chinese coverage ascendant — even before the bigger language-model scorer is wired up.
It pins down the headline thesis with a real, reproducible number rather than a narrative reading. The U.S. story is unmistakably normalizing; the Chinese story is unmistakably ascendant; and the cheapest possible model already sees both arcs cleanly.
Americans are losing faith in AI. The Chinese are not. Here's the scoreboard.
The founding paper of the tracker. We log every publicly-disclosed U.S. AI data-center cancellation since 2024, run the symmetric ledger of Chinese AI capex and model launches, score both press corpora on a rolling 13-week sentiment index, and publish the first dated, falsifiable forecast.
American AI media coverage is normalizing while Chinese AI media coverage is ascendant — and U.S. data-center cancellations are the leading capex signal. About 5.25 gigawatts of planned U.S. AI infrastructure has been pulled back across roughly two years. The tracker puts the asymmetry on a single, dated scoreboard.
Audits, scaffolds, engineering.
Smaller fixes — event reclassifications, evidence-pack pinning, scorer scaffolds — that aren't publishable on their own but are recorded here for traceability.
- 31 May 2026Details →Frame-aware LLM scorer wired up (run pending API key)Provider-agnostic LLM scorer is now scaffolded with a fixed JSON-mode prompt and a fallback chain across Anthropic, OpenAI, and Gemini. The 31 May run did not execute — pending API credentials — and a status stub was written rather than fabricated metrics.
- 31 May 2026Details →DC event reclassification + calibration scaffoldingThe 14 May 2025 Meta line was reclassified from a DC scale-back to a capex reshuffle into AI. Evidence packs were pinned for every DC event and every PRC capex row, and the 60-article hand-coded sentiment calibration set was built.