Americans are losing faith in AI. The Chinese are not. Here's the scoreboard.

AI sentiment is asymmetric. Capex is the proof.

In the eighteen months between the November 2022 ChatGPT launch and the May 2024 GPT-4o keynote, U.S. AI media coverage held a rolling sentiment above +0.65. By the 6 April 2026 anchor it prints near zero on the founding seed; on the same scale, Chinese AI media sentiment has climbed from +0.15 to +0.72. The gap between them has compressed by roughly 1.2 sentiment-units in twenty-one months — the largest such move in any twenty-four-month window since the index began. Whether U.S. AI sentiment is correct is not the indicator; the compression itself is.

Three legible events sit inside the move. The DeepSeek-V3 release of 26 December 2024 and DeepSeek-R1 of 20 January 2025⁴ broke the “Chinese AI is two generations behind” frame in mainstream US coverage inside of four weeks. Five weeks later the TD Cowen note on Microsoft's ~2 GW of canceled lease commitments converted a one-firm capability story into a sector-wide infrastructure story.¹By September 2025 the Stargate first-tranche slip extended it to the marquee buildout. Each is dated, each is marked on the chart in § V, each is sourced.

In the same window, Chinese AI sentiment moved the other way. The AI+ initiative was formalized in the March 2024 Government Work Report and extended in 2026; Alibaba committed ~USD 53 B in three-year AI+cloud capex on its FY25 Q3 call; ByteDance's 2025 AI capex was reported by the Financial Times at ~USD 20 B³; the East-Data-West-Compute phase-II buildout added another ~USD 12 B in state-direction commitments.²A major Chinese AI capex, fundraise, or model launch has landed every four to eight weeks across 2025 and 2026 H1. The US side, in the same period, has netted roughly five gigawatts of canceled, delayed, and scaled-back data-center capacity at the hyperscaler level — the running total compounds in §II below as the corpus is updated.

When Chinese AI sentiment is rising and US AI sentiment is normalizing, the structural assumption that US frontier labs dominate the next decade becomes contestable in real time. That asymmetric move is the program's leading signal.The remainder of this paper documents the cancellation record (§ II), runs the symmetric capex tracker (§ III), shows the sentiment series (§ IV), pins down the calibration baseline (§ V), logs the live prediction (§ VI), and specifies the methodology (§ VII).

Five moving parts. Each is versioned. Each ships.

The pipeline runs the Forward Indicators spine on a single corpus split four ways: U.S. AI media sentiment, Chinese AI media sentiment, U.S. AI data-center disclosures, and Chinese AI capex announcements. The founding release ships a deterministic seed scorer over hand-built anchors; subsequent releases swap the seed for a calibrated LLM scorer over the live corpus.

Per-article LLM scoring with structured-output prompts returns a sentiment score on the −1 to +1 scale and a coverage confidence between 0 and 1. The scorer is anchored to a hand-coded calibration set (sixty articles in the founding release; one hundred and twenty in a later release) labeled across 2024-Q1 through 2026-Q2. The thirteen-week rolling mean is the headline series; the window length follows the standard equity-research convention.

Every cancellation, delay, or scale-back enters as a structured event row with capacity in megawatts (rounded to fifty in the founding release) and a status verb. Quarterly review pulls from hyperscaler earnings transcripts, local utility interconnect filings, TD Cowen and SemiAnalysis sector notes, and major dispatches from Bloomberg, the Wall Street Journal, and The Information.

Symmetric on the Chinese side: each row carries a category (model launch, fundraise, state-led capex, provincial program, or talent), an actor, a U.S.-dollar figure, and a primary source. Sources: Caixin, 36Kr, Xinhua, releases from the Ministry of Industry and Information Technology and the Ministry of Science and Technology, and the published 14th- and 15th-Five-Year-Plan fascicles, cross-referenced with Brief II.

Three baselines ship alongside the LLM model: a last-observed-value baseline (the thirteen-weeks-ago reading), a random walk, and a feature-only logistic regression over the six founding features. The headline metric is the Brier score relative to baseline — a calibration score where lower is better; calibration-curve reliability is the secondary metric.

Brier score, Brier lift over baseline, calibration-curve reliability, and coverage of the 80% credible interval. A later release will ship a held-out evaluation against the expanded calibration set and a ten-fold cross-validation report.

What we will and will not publish.

Public sources only. Every data-center event and every capex announcement on the published scoreboard resolves to a hyperscaler earnings transcript, a local interconnect filing, a press-of-record dispatch (TD Cowen, Bloomberg, the Wall Street Journal, the Financial Times, The Information, Caixin, 36Kr), or a published model card or arXiv tech report.

Capacity rounded to fifty megawatts. Every data-center capacity figure is rounded to the nearest fifty megawatts. Triangulating across multiple overlapping press accounts of the same lease portfolio would otherwise introduce false precision.

Reported versus confirmed.Rows resting on secondary financial-press reporting rather than primary disclosure are flagged as reported — not confirmed — and do not carry full weight in the composite score.

No fabrication. The TD Cowen roughly two-gigawatt Microsoft headline is the founding citation; it is reported across multiple outlets but is not yet tied to a primary Microsoft disclosure. We do not infer additional capacity figures. The corpus is the source of truth.

Score cap. The composite confidence is held at 0.85 by construction to avoid false-precision overconfidence on the seed scorer.

Three things this program will not do.First, no fabricated sentiment scores: the founding release publishes ordinal seed anchors drawn from a careful reading of the press of record, marked as seed throughout, and subsequent releases publish LLM-scored values with the prompt, model name, and corpus snapshot logged alongside. Second, every U.S. data-center event on the scoreboard resolves to a primary disclosure or a press-of-record dispatch — if a row can't be traced, it does not ship. Third, no claims about non-disclosed Chinese capex: we publish what Caixin, 36Kr, Xinhua, the Ministry of Industry and Information Technology, the Ministry of Science and Technology, and provincial governments have themselves disclosed; we do not infer figures, do not extrapolate from rumor channels, and do not run satellite or shipping inferences as headline numbers.