Most AI coverage treats the industry as a collection of product launches, funding rounds, and model benchmark wars. This is like reading a map of traffic jams without understanding the underlying road network. The AI Power Atlas 10-Layer Framework is an attempt to draw that road network — to reveal the structural forces underneath the surface events.
The Framework
Each layer represents a distinct domain of structural control. Whoever controls a layer controls the terms on which others in that layer must operate.
L1 Compute
The physical substrate — GPUs, TPUs, custom silicon. Compute is the bottleneck through which all AI capability must pass. Control here creates leverage over every layer above it.
L2 Energy
Data centers consume massive amounts of electricity. As training runs scale, energy access becomes a competitive constraint. Geography, grid access, and nuclear/renewable policy determine who can afford to train at scale.
L3 Data
Training data is the raw material of model capability. Proprietary data advantages — web crawls, user interaction data, synthetic pipelines — create durable moats that compound with scale.
L4 Models
The trained models themselves — foundation models, fine-tunes, alignment layers. Model capability determines what the layer above (Platform) can offer, and to whom.
L5 Platform
The deployment and access layer — APIs, agent frameworks, fine-tuning services, enterprise integrations. Platform lock-in operates here: whoever controls the interface controls the relationship with developers and enterprises.
L6 Capital
Investment, valuation, and financial power. Capital gravity determines who can sustain the compute expenditure required for frontier model development and who cannot.
L7 Geopolitics
Nation-state competition for AI advantage — export controls, sovereign compute programs, allied tech policy, strategic partnerships. The geopolitical layer increasingly sets the rules within which all other layers operate.
L8 Regulation
Laws, standards, and compliance frameworks. Who writes the rules determines who benefits from them. Regulatory capture — where incumbents shape the regulations that govern them — is the defining risk at this layer.
L9 Labor
AI talent, workforce displacement, union dynamics, researcher mobility. The talent concentration in a handful of frontier labs mirrors the compute concentration — and creates similar structural leverage.
L10 Public Perception
Narrative power, media framing, public trust, and the legitimacy of AI development. Perception shapes regulation, talent flows, and political will. It is the slowest-moving layer but the most structurally destabilizing when it shifts.
How to Use the Framework
When any significant AI event occurs — a funding round, a model release, a policy announcement, an executive departure — the first question is: which layer does this primarily affect, and what does that imply for adjacent layers?
A model benchmark result (L4) matters primarily insofar as it shifts platform adoption (L5) and attracts capital (L6). An export control announcement (L7) matters primarily insofar as it constrains compute access (L1) for specific geographies. A talent migration (L9) matters insofar as it shifts model development capability (L4).
The framework does not replace judgment. It structures judgment — ensuring that analysis considers the full structural picture rather than the surface event.