Five pieces that sharpened the view on the AI economy this week — from the solo founder to Europe’s strategic position. Curated, commented, not aggregated.
The live case behind my sovereignty argument — and the must-read of the week. An export-control directive forced Anthropic to disable its two most capable models for every customer worldwide, eleven days after filing to go public. The lesson in one line: control follows the provider’s jurisdiction, not where the data sits — Frankfurt hosting wouldn’t have changed a thing.
The sharpest read of the week: a scenario-novella on Europe’s possible slide into AI irrelevance, grounded in real data. Its brutal number — 17.3 GW of US compute against Europe’s 1.4 GW, a 12.4× gap. Its core line could be mine: sovereignty is easier to announce than to build.
Satya Nadella’s thesis, the one I kept building on this week: a “frontier model” alone is unstable — value has to flow through an ecosystem, not condense into a few models. Every organisation needs its own learning loop as IP. This is the company-level version of my sovereignty argument: build, don’t just buy models.
The “super-individual” founder is no longer theory — one senior plus a swarm of agentic AI colleagues now does what used to need a team. I lived exactly this building SchulHeld. My operator’s footnote: what gets scarce is no longer execution capacity, but judgement, taste and distribution.
Marc Zao-Sanders’s annual reality check, third edition. The 2026 surprise: the fastest-rising uses aren’t productive but playful — fake reality TV, robot tarot, fan fiction. My take for industrial buyers: real adoption follows what people find useful or delightful, not the strategy deck.