Chinese labs are shipping flash-class open models, with overlapping architectures and domestic-chip inference runs in the same window.

Coding-agent access limits, command gates, and multi-account agent tooling are shaping how builders ship constrained agent workflows.

Humanoid robotics evidence spans automaker deployment bets, dexterous-hand development, shared robot data, and physical-system security.

Courts are revisiting government AI supply-chain bans while safety diagnostics and AI-targeted cyber defenses draw institutional focus.

Large infrastructure funds and chip-leasing debt are flowing into AI compute while open-weight startups draw acquisition interest.

Chip roadmaps, GPU leasing debt, serving stacks, and local data-center limits keep compute supply and operating constraints active.