How a verdict is set, and what it is not
Nine conditions, not predictions
The nine trackers follow conditions that would have to hold for the AI build-out to keep being financed. They are conditions, not predictions. Each verdict is computed from dated, sourced signals. The nine questions on the home page are:
- Can they actually plug it all in?
- Is borrowing still cheap enough to build?
- Are governments slowing it down?
- Can anyone pay for all these data centers?
- Will one of the big AI labs go public?
- Are the AI labs actually making money?
- Is the rally still tied to earnings?
- Is AI making ordinary companies more productive?
- Is anyone actually using AI agents?
Thresholds
Each row names whoever stated the threshold and links to the source. Where a threshold was spoken rather than written, the quotation is transcribed from the linked recording with filler words removed; nothing inside quotation marks is paraphrased. Where no quotable figure was found, the row says the wording is the map’s own.
The 5% borrowing warning
On the row “Is borrowing still cheap enough to build?”, the 5% borrowing warning is the map’s own example, not a general rule.
What the labels mean
Cracking means a major dated event points against it. Tracking means qualifying evidence only points toward it. Not enough evidence means nothing has cleared the bar, and that is a real answer.
What never counts
Analyst estimates, reported internal projections and unconfirmed reports never count. A signal stops counting after 90 days.
Green, yellow and red
What green, yellow and red mean is in Signal status, the Pulse feed and the thesis tracker. What a verdict is, and what Mixed means.
Not a rating, a forecast, or a recommendation
A verdict is not a rating, a forecast, or a recommendation.
Briefings
Research and drafting are AI-assisted. Steven Dudley edits the weekly finding and the Pulse. Every claim links to a dated source. Briefings and publication dates.
How the AI Power Rankings work
The eight divisions
Power. Data centers. Chip equipment and manufacturing. AI chips and networking. Cloud and compute. AI models. AI software and apps. Robotics and physical AI.
Every company sits in exactly one division, based on its sector on the Acts of Evolution AI map. Space infrastructure, quantum computing and backers are not ranked.
How places are set
In AI models, each model's place comes mainly from how it ranks on independent public tests, with smaller capped nudges from its maker's backing, its use, its new releases and its sourced news.
In the other seven divisions, each company's place comes mainly from its size, measured by stock market value or latest private valuation and by revenue, with a small capped nudge from its sourced news. Revenue is taken from the company's own filings or results. Revenue in other currencies is converted to US dollars at the Federal Reserve rate on the last day of each company's financial year.
A company with a value but no published revenue is placed on its value alone and says so. A company with published revenue but no published value is placed on its revenue alone and says so. A company with no sourced size figure is not given a place yet. Large companies with many businesses are placed on their whole size and say so.
Division colour
Each company's colour comes from the Acts of Evolution AI map, which colours a company from its sourced news over the last 90 days. A company with no such news has no colour.
- Companies with no colour are left out. A division takes the most common colour among its coloured companies. Green is Healthy, yellow is Day-to-day and red is Injured list.
- If two or more colours are tied, the division shows yellow, Day-to-day.
- If no company in the division has a colour, the division shows Off-season.
The count covers every company in the division, placed or not. In AI models that is all 22 companies in the division. Counting only the 13 companies in the division that make a ranked model gives 3 coloured companies, and the division is still Day-to-day.
Colours as of 18 October 2026 (coloured companies out of all companies in the division):
Power: 5 of 18 coloured (5 green, 0 yellow, 0 red). Healthy.
Data centers: 5 of 23 coloured (5 green, 0 yellow, 0 red). Healthy.
Chip equipment and manufacturing: 10 of 54 coloured (10 green, 0 yellow, 0 red). Healthy.
AI chips and networking: 4 of 42 coloured (4 green, 0 yellow, 0 red). Healthy.
Cloud and compute: 6 of 12 coloured (5 green, 1 yellow, 0 red). Healthy.
AI models: 4 of 22 coloured (1 green, 2 yellow, 1 red). Day-to-day.
AI software and apps: 0 of 24 coloured. Off-season.
Robotics and physical AI: 0 of 48 coloured. Off-season.
Rankings show where a company stands in the AI economy. They say nothing about its shares.