Attention is narrowing, and it has changed subject.
The five most-covered subjects now take 35.1% of all named-subject events, against 24.8% in the preceding ninety days. OpenAI alone moved from 7.1% to 11.6%. Over the same span, policy gave up 8.9pp of the cycle while funding took +6.7pp.
- Events counted
- 4,886
- Named subjects
- 1,932
- Categories
- 17
- Months
- 17
Subjects
Share of named-subject events, trailing 90 days against the 90 before.
| Subject | Then | Now | Movement | ±pp |
|---|---|---|---|---|
| OpenAI | 7.1 | 11.6 | +4.5 | |
| 4.6 | 8.0 | +3.4 | ||
| Microsoft | 1.7 | 3.6 | +1.9 | |
| China | 0.5 | 1.3 | +0.8 | |
| Meta | 1.8 | 2.5 | +0.8 | |
| Cloudflare | 0.2 | 0.8 | +0.6 | |
| Losing ground | ||||
| Amazon | 1.6 | 0.4 | −1.2 | |
| United States | 1.9 | 0.8 | −1.0 | |
| Oracle | 0.6 | 0.0 | −0.6 | |
| Sam Altman | 0.8 | 0.2 | −0.6 | |
| Intel | 1.0 | 0.4 | −0.6 | |
Categories
Share of all events. Rotation shows here before it shows in headlines.
| Category | Then | Now | Movement | ±pp |
|---|---|---|---|---|
| policy | 17.7 | 8.7 | −8.9 | |
| funding | 3.5 | 10.2 | +6.7 | |
| security | 6.2 | 10.2 | +4.0 | |
| research | 10.1 | 13.2 | +3.1 | |
| opensource | 1.1 | 4.1 | +2.9 | |
| product | 21.8 | 19.5 | −2.3 | |
| release | 8.0 | 9.8 | +1.7 | |
| legislation | 3.4 | 1.8 | −1.5 |
Gone quiet · Oracle, UK Government, Tesla — each carried eight or more events in the archive and none in the last ninety days.
Listed vs not · 51.1% of coverage we can classify by listing status goes to companies that are not publicly traded — effectively unchanged from 50.1% the quarter before. OpenAI and Anthropic alone account for 20.9% of every named-subject event in the archive. Listing status is a maintained list, not archive data; it covers 46.7% of named-subject events.
Method
Every figure is a full count of the archive, not a sample. Coverage runs 25 Mar 2025 to 24 Jul 2026.
Shares are reported rather than totals because publishing volume is not constant — 492 events in the current window against 1,070 before it. Comparing raw counts across those windows would show a decline everywhere and mean nothing.
Subject labels are normalised onto head entities, so Google DeepMind and Google Cloud count as Google. Generic labels such as “AI” or “Big Tech” are excluded.
The same archive, read once a week by a human.
Issue 51 — Cursor's Zero-Day, Kimi K3, and Codex's Sealed Audit Trails. Two insights, each with the case against it. Tuesdays, 7:30am UK.
Read it on LinkedIn →