The tension between AI's capital intensity and the political will to constrain it is sharpening. Anthropic just locked in $3.4 billion of gas-fired power generation to feed its data centers, a move that bypasses traditional energy suppliers and signals how vertically integrated the AI stack has become, while simultaneously European activists are mounting real resistance to Google's expansion and newly elected politicians are signaling they may tighten restrictions on US data center construction. Meanwhile, Nvidia's board is investigating Jensen Huang over a relationship with a suspected Chinese spy as Yale's Sonnenfeld estimates his tenure at 50-50, a distraction arriving precisely when the company's foundational role in the AI supply chain makes it a chokepoint for every player downstream. The infrastructure plays are getting more aggressive and more exposed, the regulatory pressure is shifting from framework documents to actual site-level obstruction, and the concentration of power in a few hands is creating both fragility and targets. Against this, the technical progress continues on parallel tracks: world models moving AI into 3D space, MIT's WorldParticle system learning how matter behaves across domains, and open-source security tooling like 0 emerging from smaller labs. The builders are still shipping, but the capital and political economy of scaling has become as contested as the models themselves.
Sloane Duvall