About

AI Gap Clock is a public, sourced scoreboard of measurable US–China AI capacity — infrastructure, routed demand, and research volume a reader can check against named publications. It is not an oracle for “who is winning AI.”

How the lanes connect. Chips, power, and memory are infrastructure — who can build and feed accelerators. Open-market token share is routed demand on one public router. Papers, notable models, and Arena are research and published preference rankings. Epoch ECI months are a modeled time-to-frontier yardstick, distinct from Arena; a current ECI point gap is listed beside that published average. Government packages are policy fuel, not a yearly budget. The lanes inform one another; they do not sum to a score.

The board is published as AI Gap Clock, a public scoreboard. There is no institute behind it, and no methodology paper to cite. The sources are the publications named on each tile.

The masthead clock is a representation of the latest board update. Hands are frozen at that date; digits do not tick. Each tile shows the last published figure from a named source. The board is not live and not real-time.

Primary is the checkable board: Measured counts and Official estimates a reader can open and verify. Estimates is the quieter row — modeled reconstructions, Epoch ECI months, or government packages that are hard to compare. Do not subtract papers from models. There is no unitless hero gap.

Notable-model counts are Epoch AI’s family as reported by the Stanford AI Index — one count, two publishers. The Index reports Epoch’s bar; Epoch curates the family. That is not a second independent census.

The frontier performance tile is the LMArena leaderboard-dataset (text, overall, raw ratings without style control; current vintage), not the R&D chapter’s paper and notable-model counts. The 2026 Index Technical Performance figure (March 2026: 2.7%) stays on the tile as an earlier figure. End-2023 gaps of 17.5–31.6 points on MMLU / MMMU / MATH / HumanEval are a different yardstick from the 2025 Index.

Open-market token share is routed demand from an OpenRouter top-models day sample — one public router. Unmapped or stealth models stay unmapped and are never counted as China. Attribute OpenRouter (CC BY 4.0).

The Epoch Capabilities Index (ECI) months tile is the published January 2026 Data Insight — mean China lag 7 months since 2023 (min 4, max 14). That is a modeled time-to-frontier yardstick, distinct from the Arena preference ranking. This board does not invent a current gap in months. A second reading on that tile’s companion is the current ECI point gap from eci_scores.csv — points, not months.

Every figure names a public primary and an as-of date. The board draws on Epoch AI, OpenRouter, Counterpoint, the IEA, the Stanford AI Index, NIST / CHIPS, and the other publications listed on Sources. If a figure cannot name a primary and a date, it stays off the board or is marked EXAMPLE.

Social posts, forums, blog rankings, and Wikipedia-only cites are not used.