AI Daily · Series 01

Doom v Boom

One frozen edition a day: the previous 24 hours of AI data-centre, energy and sustainability news, scored boom or doom, every claim traceable to its source.

Latest · Fri 16 Oct 2026№ 24
Boom-leaning
TSMC explores Texas expansion as a selloff and rising rates test the AI data-center deal boomGartner forecast that 80% of neoclouds will lose their independence by 2030, and BlackRock's Larry Fink warned that a halt in AI data-center building could tip the US into recession. ASML raised its guidance for a third time, while Vicor shares fell 12% and the Trump administration doubled its equity stake in GlobalFoundries.85 stories · 8 papers

The last 24 mornings

Each ring is a morning: its green share is the boom share of the day’s weighted developments, tilted by the tracked stocks’ session, and the number is that share out of 100 — 50 is balanced. Select a ring to read the edition.

  1. 8123 Sep+0.62
  2. 8424 Sep+0.69
  3. 7325 Sep+0.46
  4. 7226 Sep+0.43
  5. 5827 Sep+0.17
  6. 8328 Sep+0.67
  7. 8329 Sep+0.66
  8. 8730 Sep+0.73
  9. 831 Oct+0.66
  10. 812 Oct+0.62
  11. 763 Oct+0.51
  12. 624 Oct+0.25
  13. 805 Oct+0.59
  14. 786 Oct+0.55
  15. 797 Oct+0.58
  16. 808 Oct+0.59
  17. 729 Oct+0.44
  18. 8010 Oct+0.61
  19. 3311 Oct−0.34
  20. 7312 Oct+0.45
  21. 8113 Oct+0.63
  22. 7514 Oct+0.51
  23. 7315 Oct+0.46
  24. 5616 Oct+0.12

Methodology

How a morning’s news becomes one Boom Score.

news = (Wboom − Wdoom) ÷ (Wboom + Wdoom)
market = tanh(layer-averaged move % ÷ scale)
reading = 0.75 × news + 0.25 × market
Boom Score = (reading + 1) ÷ 2 × 100

  1. 01

    The window

    Each edition covers 24 hours, 06:15 to 06:15 UTC: news about AI, data centres, chips, energy and sustainability, plus new arXiv research. A classifier keeps what is relevant and tags every report with its AI layer, place, the figures it states, and whether it points to boom, doom or neither.

  2. 02

    Events, not headlines

    Ten outlets covering one announcement is one development, not ten votes. Reports are grouped into events by what they describe — the same actors, place and figures — and each event takes the majority mood of its reports. An even split counts as neutral. (The earliest editions, before events, counted each report once.)

  3. 03

    Weighted by what matters

    Each event is weighted by relevance to the build-out (0.6–1×), by impact graded on its size (about 1× for the smallest to 6× for the largest; a plan or warning grades one step below the same thing done), and by coverage (up to 1.75× for widely reported events).

  4. 04

    The news reading

    The news reading is the boom weight minus the doom weight, over their sum — from −1 (all doom) to +1 (all boom), neutral events left out.

  5. 05

    The market

    The market is the previous day’s session for the US-listed AI-infrastructure stocks we track. Each close-to-close move is capped at ±10%; within each AI layer (data centres, hyperscalers, chips, chip equipment) the moves are weighted by market cap, and the four layers count equally, so the many chip names don’t outvote the rest. The result is read on the same −1 to +1 scale, with the scale calibrated on past days so the market swings about as much as the news. It then tilts the news reading rather than being averaged with it: a flat session leaves the score where the news put it, a strong one moves it by up to a quarter of a point. With no session in the window (weekends, holidays) the score is the news reading alone.

  6. 06

    The score

    The Boom Score is the blended reading as a share out of 100: 50 is balanced, 45–55 reads “Balanced”, 35–65 “leaning”, 20–80 “clearly”, and beyond that “decisively”. Editions before the market term scored news alone.

  7. 07

    Frozen and sourced

    The edition is composed each morning and published at 09:00 UTC, then never edited; a correction is a note in the next one. Every number is assembled in code from the stories it cites, and every note links back to them.

Every edition

October 2026