China’s reported net additions of generation nameplate capacity stay at least three times the boundary-matched US total, every single year.
45%likely
Well sourcedNext check
Watching
Why it is different
A pure measurement of capability. The “so US AI gets throttled” half was cut because it repeated the transformer wall.
Why it matters
Electricity is the input to everything downstream: industry, AI, transport. A sustained gap compounds.
Next check · 31 Dec 2026
Do China’s new generating-capacity additions for 2026 come in at three times the US total or more?
How we scored it ·
from how often this kind of thing happens
If right, how big5/5How overlooked3/5Upside vs downside2/5Source trust5/5Independent sources3/5
The exact wording we score
For each calendar year 2026 through 2030, the year-over-year increase in China's NEA-reported total installed electricity-generation nameplate capacity is at least 3.0x the boundary-matched US increase computed from EIA final utility-scale net-summer capacity plus EIA's estimated small-scale solar capacity. Both sides include additions net of retirements; storage is excluded from both numerators because it is not generation. Resolves YES only if every annual ratio is at least 3.0x; NO on the first ratio below 3.0x. If either agency changes a boundary, the prior-year series must be restated on the new basis or the affected year is disputed rather than silently compared. Final adjudication occurs no later than 31 Mar 2031.
PROOF
Evidence and provenance
The links below are the evidence recorded when this forecast was scored. Each note states source quality or bias; a citation is not an endorsement.
Recorded sources · 4 citations / 3 independent origins
2026-07-28 | sell-side: US-China AI framing remains chip-led; mainstream press: US power constraints are now common but the comparative electricity-build ratio is not; priced market: no clean cross-country instrument identified; specialist: energy data supports the power-capability comparison while warning capacity is not generation | E3
Revision history4 entries
Probabilities and interpretations are never silently overwritten. This is the append-only record of what changed and why.
Domain review revision. The earlier NEA-versus-EIA wording compared a China series including distributed capacity with a US utility-scale figure and called the result “generation.” The outcome now includes EIA estimated small-scale solar, excludes storage on both sides, uses net nameplate capacity rather than energy supplied, and requires explicit restatement or dispute if an agency changes its boundary. P remains 45 because the correction changes measurement validity, not the already modelled compression path.
(2nd pass) — SPLIT. This signal was a compound claim and is now atomic. It previously asserted BOTH (i) China adds >=3x US net new generation each year AND (ii) the US buildout is throttled by power by 2030, scored as one probability of 77 — the highest in the book. Two defects followed. First, conjunct (ii) was strictly entailed by datacenter-grid-gear-bottleneck: that signal claims a third to a half of announced US DC capacity slips on gear, and 33% >= this file's 15% bar, so grid-gear TRUE implies conjunct (ii) TRUE. The same physical fact was being scored twice. Second, the conjunction arithmetic: P(ratio holds each year) ~0.45 given the ratio compressed from ~8x toward ~3.5x in a single year, times P(US throttled) ~0.85 ≈ 0.38-0.45, not 0.77. Conjunct (ii) is deleted (cross-reference grid-gear instead); P cut 77→45; the claim is now the ratio alone. Falsifier tightened: the old 'two CONSECUTIVE years below 3x' let a single sub-3x year make the claim false without falsifying it — worth ~15 points of probability by itself. Now falsified on the first sub-3.0x print. Left ai-buildout-wall: a claim about Chinese national capacity is not a leg of a thesis about US physical supply constraints; it was contributing to a diversification illusion. Its US-side indicators (NERC LTRA, PJM BRA) moved to grid-gear where they belong. Honesty note: this restructure mechanically lowers reported theme concentration — that is a side effect of correct labelling, NOT a reduction in real exposure, and must not be read as progress.
v2 rescore, P held at 77 on offsetting evidence. The China leg is compressing: NEA's H1-2026 release (2026-07-22) put total installed capacity at 4.04 TW, implying only ~150 GW added in H1-2026 versus ~296 GW in H1-2025, with solar additions down 66% YoY to 72.07 GW after the Document-136 pricing reform. The China:US ratio is falling from ~8x toward ~4x — the >=3x threshold still clears, but the margin is now thin enough that a second weak China year is a live falsification path. The US leg strengthened: Sightline Climate tracking (Apr–May 2026) puts 7–11 GW of the 12–16 GW of US data-center capacity planned for 2026 as delayed or cancelled — 30–50%, far above this claim's 15% bar.
created from research sweep; IEA leg spot-verified. Edge trimmed to 3. Highest-conviction seeded signal.
Method challenge
Adversarial review record
Admitted after revision
Structured internal role review tied to this frozen claim. It is not independent human peer review. Independent specialist review not yet performed
systems or deployment specialist
measurement and data reviewer
policy or geopolitical reviewer
forecasting-method reviewer
Strongest specialist challenge
The former ratio compared a Chinese headline series that includes distributed capacity with a US utility-scale figure and then described nameplate MW as electricity generated. That boundary mismatch could create the claimed gap without a physical capability difference.
Outside view
China begins from a very large deployment lead, but the annual ratio is compressing and must clear 3.0 in five consecutive years. The conditional-path calculation supports 45 percent rather than treating the 2025 ratio as independently repeated.
Causal rival
A Chinese slowdown, US tax or permitting acceleration, distributed solar measurement changes, or retirement timing can move the ratio. Capacity factor, curtailment, grid connection, and reliability can also diverge from nameplate deployment without affecting the binary.
Measurement risk
The revised method adds EIA estimated small-scale solar, excludes storage from both sides, and uses net nameplate changes. Any agency boundary revision requires a restated prior year or disputed disposition; silent splice or gross-versus-net comparison is prohibited.
Residual risk
NEA and EIA still differ in statistical systems, revisions, and publication lag, so the ratio is an administrative deployment comparison rather than a measure of usable energy or compute advantage. Public copy must keep that limitation visible.
COLDTELL