THE TRANSMISSION CHAIN · WHERE THE SPIRAL FAILS

AI Inflation Stops at the API. Power Demand Does Not.

Scarce chips raise the cost of compute, but model efficiency and price competition interrupt the pass-through before it reaches software users.

01 · CHIPS

GPU scarcity

Demand surges faster than advanced packaging and memory supply can respond.

→

02 · COMPUTE

Capex explodes

Cloud providers finance servers, campuses, cooling and networking at historic scale.

→

03 · MODELS

Efficiency fights cost

Smaller models, better utilization and rival providers reduce cost per useful answer.

→

04 · APPS

Prices keep falling

Customers capture much of the efficiency unless an application owns distribution or workflow.

Software chain: brokenHardware inflation does not guarantee API pricing power. Competition can transfer savings directly to users.

Power chain: intactEvery accelerator still needs electricity, cooling and grid access. Those assets take years to add.

Trade the cycleGPUs, HBM and packaging while supply remains tight.

Own the bottleneckGeneration, grid equipment and approved rate-base growth.

Demand pricing powerSoftware must turn cheaper AI into revenue, not only lower cost.

Framework based on Nvidia results, official API pricing, Berkeley Lab demand scenarios, EIA retail-price analysis and PJM capacity-market materials. Educational, not an allocation recommendation.