Pillar

Compute Abundance

Learned models and efficient silicon as the second pillar.

5 min read·Updated 2026-04-12·editorial

Compute abundance is the second pillar of the Age of Abundance: the trajectory by which useful computation, and increasingly useful *cognition*, falls toward the cost of the electricity it consumes. Where energy makes atoms cheap to move and transform, compute makes intelligence cheap to apply — to design, to diagnose, to tutor, to coordinate.

From Moore's law to learning curves

For five decades the density of transistors on a chip roughly doubled every two years (Moore's law), and the cost per logic operation fell with it. As that geometric scaling slows, the frontier has shifted to specialization — GPUs, TPUs, and other accelerators — and to algorithmic efficiency, where the compute needed to reach a fixed capability has itself been falling rapidly. The compound effect is that the price of a unit of useful AI work keeps dropping even as raw transistor scaling plateaus.

Intelligence as an input good

When competent cognition is cheap and abundant, it stops being a scarce professional service and becomes an input to everything else — like electricity after the 1920s. Tutoring, legal triage, scientific literature review, and software creation move from rationed to ambient. The abundance framing treats this not as replacement of human judgment but as the removal of a bottleneck on how widely judgment can be applied.

The bottlenecks that remain

Compute abundance is gated less by physics than by energy, capital, and coordination: data-center power draw, fabrication concentration, and the governance of who controls frontier models. Cheap computation that is owned by few is the canonical way abundance turns into concentrated rents rather than shared capacity — which is why this pillar leans so heavily on the others.