Layers

Capital availability for compute and infrastructure.

Speed at which enterprises and governments integrate AI capability.

Electricity generation, transmission, and interconnection capacity.

Labor-market disruption, cultural resistance, and political blowback.

Political and regulatory resistance (federal, state, international).

Governance and market-power concentration --- cloud primes, frontier labs, chip foundries, mission-software layers. Harm is the democratic and epistemic cost of few actors gatekeeping the compute, models, and deployment surface.

Labor welfare outcomes from AI-driven workforce reshaping --- wage compression, role hollowing, loss of meaning and skill-exercise, not only headcount loss. Distinct from friction_public (which tracks resistance as an obstacle); harm_displacement tracks the welfare harm itself if the intervention lands.

Silicon, rare-earth, cobalt, copper, and lithium supply-chain harm --- mine-site pollution, worker welfare in extraction economies, and geopolitical concentration of refining capacity.

Physical footprint of datacenters, transmission rights-of-way, and new generation siting --- including ecological-corridor fragmentation, agricultural land conversion, and siting in environmental-justice communities.

Institutional dependency on a single vendor, model family, or deployment stack --- procurement lock-in, curriculum lock-in, regulatory-frame lock-in. Harm is the loss of future optionality when the chosen path stops serving the original goal.

Datacenter and thermal-generation freshwater consumption --- direct cooling withdrawal plus indirect water embedded in power generation. Measured as withdrawal, consumption, and basin stress concentration.