The Power That Eats Itself
The AI electricity debate was always about scale. A quieter finding flips it: the violent, millisecond power swings of AI training are cracking turbines, killing batteries within weeks, and rattling the grid. The bottleneck may not be how much power exists, but how hard AI drinks it.
Read full story →Meta Opens the Weights
Zuckerberg gave away Meta's most powerful model and a 6,500-word manifesto to match, casting open AI as empowerment against rivals' "walled gardens." The subtext is sharper: Chinese open-weight models, cloud costs, and a bid to turn Meta's weakness in the model race into a strategy.
Read full story →The 7 Percent Problem
Only a handful of corporate AI projects can prove a return. The comfortable explanation is sloppy spending. The more unsettling one: firms are measuring AI like software, counting seats and licenses, when agents have started behaving like labor that does the work and can quietly do it badly.
Read full story →