Sublinear Scaling

In the AI Story

The pattern of sublinear scaling is one of the most elegant results in complexity science. When a biological organism doubles in size, it does not need twice as much food — it needs roughly 68% more. The cardiovascular system becomes more efficient because the branching geometry compounds: longer main vessels serve larger regions, and the energy required to pump blood per unit of tissue decreases. This is why elephants do not eat ten thousand times more than mice.

The same principle applies, with different constants, to cities and to companies. A city that grows from one million to two million residents does not need twice as many roads; it needs about 1.85 times as many. A company that grows from one thousand to ten thousand employees does not need ten times as much administrative overhead per unit of output; it needs less. In every case, growth produces efficiency.

But the efficiency comes with a cost that is invisible until it manifests. The same network geometry that produces sublinear scaling produces a characteristic developmental trajectory: rapid initial growth, gradual deceleration, plateau, and eventual decline. The sigmoid curve is not an accident of particular organisms or particular companies — it is the mathematical signature of any system whose scaling exponent is less than one. The corporate mortality curve that West documents follows the same mathematical form as biological mortality for this reason.

The mechanism is structural. A fractal branching network optimized for efficient delivery is also, by its geometry, hostile to novelty. Signals that do not fit the expected format must travel up through layers of aggregation, each layer filtering for relevance. The network is a signal-degradation machine. It delivers resources with increasing efficiency and destroys surprise with increasing efficiency, until the surprise-destruction outpaces the network's capacity for renewal. The system calcifies, stagnates, and dies.

This structural analysis is what makes sublinear scaling relevant to the AI transition. Organizations that adopt AI as a productivity accelerant without changing their hierarchical topology remain in the sublinear regime. They become faster mice, not bigger elephants — and emphatically not cities. Their increased metabolic rate, rather than extending their lifespan, compresses it.

Origin

Sublinear scaling was empirically documented in biology by Kleiber (1932) and later workers. Its theoretical explanation came with Kleiber's law and the 1997 derivation by West, Brown, and Enquist. The extension to cities and companies is West's own work with Luis Bettencourt and colleagues at the Santa Fe Institute, beginning in the mid-2000s.

Key Ideas

Exponent below 1.0. Output grows more slowly than input, producing increasing per-unit efficiency as the system grows.

Biological signature. Metabolism, heart rate, aorta diameter, and dozens of other biological variables scale sublinearly — always at multiples of one-quarter.

Urban infrastructure. Physical infrastructure in cities scales at approximately 0.85 — larger cities are more efficient in roads, cables, pipes per capita.

Corporate maturation. Companies scale sublinearly in innovation as they grow, becoming more efficient and less generative — the structural mechanism behind corporate mortality.

Sigmoid trajectory. Sublinear systems produce characteristic S-curves: rapid growth, deceleration, plateau, decline — the mathematical signature of finite lifespan.