Gordon Moore

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI identifies the imagination-to-artifact ratio—the distance between a human idea and its realization—as the quantity AI has compressed toward zero for a significant class of work. Moore’s framework supplies the mechanism: each successive halving of the cost of capable computation crosses a price threshold that releases stored pressure—the accumulated potential of unsatisfied needs that builds between compressions. ChatGPT’s two-month adoption to fifty million users was not a measure of how good the technology was; it was a measure of how long the dam had held. Moore watched exactly this pattern across half a century of semiconductor history and named it with precision: the technology does not create the demand, the cost reduction does.

His framework also supplies the most sober corrective to AI triumphalism the cycle deploys. Moore spent fifty years inside an exponential and never confused trajectory with destiny. The self-reinforcing cycle that drove Moore’s Law was not a law of nature; it was a social phenomenon sustained by economic incentives, and it persisted because the semiconductor ecosystem continuously found new dimensions when the old ones saturated. The AI scaling cycle faces the same test: the data wall, the energy wall, the economic wall of escalating training costs that must be matched by escalating revenue. Moore’s career says the walls will be encountered, the dimensional rotations will be necessary, and the response to them—not the current celebration—will determine whether the trajectory continues.

Moore also issued a warning the cycle amplifies: his was a one-dimensional measurement of a multidimensional phenomenon, and treating benchmark capability scores as though they captured intelligence with the same cleanness that transistor density captured processing capacity is the Panglossian error his own humility refused. In a 2008 IEEE Spectrum piece on the technological singularity, he argued that treating intelligence as a one-dimensional, quantifiable characteristic was naïve—this from the man whose name was synonymous with quantifying progress on a single axis. The AI benchmarks that now drive investment are, by his own light, as incomplete as transistor counts were for measuring the value the chips would create.

Origin

Gordon Earle Moore was born in 1929 in San Francisco and trained as a chemist, earning his doctorate at Caltech. He was among the eight engineers who left Shockley Semiconductor in 1957 to found Fairchild Semiconductor—the “Traitorous Eight” whose departure seeded Silicon Valley—and later co-founded Intel in 1968 with Robert Noyce. The 1965 paper in Electronics magazine that contained the observation later named Moore’s Law was a commissioned piece on the future of integrated circuits, not a formal academic publication. Moore extended his trend line forward ten years, assumed it would hold that long at most, and moved on.

The observation acquired the force it did because the industry organized around it. Semiconductor companies planned research to meet the doubling timeline; equipment manufacturers designed lithography tools for the next generation; software developers wrote programs that would require the next generation’s processing power. The prediction became a target, and the target became a coordinating mechanism for an entire ecosystem. Moore spent the rest of his career inside the curve he had identified, watching it hold through mechanisms—photolithography refinements, new materials, three-dimensional architectures, multi-core parallelism—none of which he had anticipated in 1965. He died in 2023 at ninety-four.

Key Ideas

Cost, not capability, is the primary variable. The 1965 paper was not about faster computers; it was about cheaper computation. Moore always insisted that the doubling of transistors mattered because it meant lower cost per transistor, and cheaper computation creates new classes of users who were previously priced out. The home computer, the automobile control system, the smartphone—none arrived because a capability threshold was crossed; each arrived because a cost threshold was crossed. Applied to AI, the hundred-dollar subscription is more predictive than the benchmark score.

Self-fulfilling prophecies are social, not physical, phenomena. Moore’s Law held because an industry organized itself to make it hold, not because physics demanded it. The AI scaling laws—Kaplan (2020), Chinchilla (2022)—are acquiring the same self-fulfilling character: companies invest hundreds of billions on the assumption that the next doubling is coming, and the investment helps ensure it does. But a self-fulfilling prophecy breaks the moment the economic incentives that sustain it do.

Walls and dimensional rotations. Every exponential curve in technology hits a physical limit, and the semiconductor industry’s response—rotating to a new dimension of growth rather than abandoning the trajectory—is the structural signature Moore’s career documents. Clock speed hit a thermal wall in 2003; the industry rotated to multi-core parallelism. AI training faces a data wall (the finite supply of quality text), an energy wall, and an economic wall. The rotations that follow—synthetic data, efficiency improvements, new architectures, on-device inference—will preserve or break the trajectory depending on how quickly they are executed.

Exponential growth is invisible from inside. Each doubling, experienced in real time, felt to the engineers producing it like a manageable, incremental next step. The cumulative effect—the second half of the chessboard—exceeded all intuition. Moore’s practical corrective was the industry roadmap: a formal document that organized the present around a trajectory too steep for individual intuition to grasp. AI has no equivalent roadmap, and its practitioners are operating on the second half of the computational chessboard with first-half intuitions.