Cost Over Capability

In the AI Story

The Intel 4004, released in 1971, was the clearest illustration. Operating at 740 kilohertz and executing roughly sixty thousand instructions per second, it was by any capability measure inferior to the mainframes of the era. What it was, decisively, was cheap: a single chip costing a few dollars in volume replaced a circuit board costing hundreds or thousands. The cost collapse created users who had never existed before — calculator manufacturers, traffic-light controllers, microwave-oven designers — none of whom had been 'computer users' in the mainframe era. The microprocessor created the category by crossing a cost threshold.

Each subsequent halving of cost per transistor crossed another threshold and created another category of user. The personal computer made the individual knowledge worker a user. The smartphone made the global consumer a user. IoT sensors made the physical environment itself a user. The number of computing devices went from thousands to millions to billions to what is now projected as trillions — and none of these transitions was driven by a capability breakthrough. Every one was driven by a cost breakthrough.

Applied to AI, the framework reframes the conversation. The imagination-to-artifact ratio Segal describes has not compressed because a brilliant model was invented. It has compressed because the cost of intelligent computation has dropped to the point where describing what you want in plain language is cheaper than hiring someone to build it. Large language models existed before ChatGPT; GPT-3 was available through an API in 2020. The stored pressure was not released because the release mechanism was too expensive for the people carrying the pressure.

The hundred-dollar-per-month price point is a waypoint on a cost curve that will continue declining. At fifty dollars, the student in a middle-income country crosses the threshold. At twenty-five dollars, the micro-entrepreneur in a low-income country. At effectively zero — subsidized by advertising or data economics — the last barrier between imagination and artifact disappears for the end user. Moore's framework predicts that the AI equivalent of the musical greeting card will be built, and it will seem, in retrospect, as trivial and as transformative as the microprocessor-in-a-greeting-card seemed in 1995. The cost curve does not care about prestige.

Origin

The principle is implicit throughout Moore's 1965 paper, which frames the coming revolution in economic rather than performance terms. 'Integrated circuits will lead us to such wonders as home computers — or at least terminals connected to a central computer — automatic controls for automobiles, and personal portable communications equipment.' Every prediction came true, not because the chips became powerful but because they became cheap enough to embed in products that could not have justified the previous cost structure.

Key Ideas

The 'who' matters more than the 'what.' Capability determines what is possible; cost determines what is actual — and the gap between them is where the history of technology actually happens.

Thresholds create users. Each halving of cost crosses a price threshold that makes a new category of user economically viable, and the new users generate applications the previous users could not imagine.

Applications follow cost, not capability. The greeting card that plays music, the toy that responds to voice commands — none would have survived a capability-based cost-benefit analysis, yet all became possible when the cost crossed the threshold.

The hundred-dollar price point is a waypoint. Each subsequent halving will create new cohorts of users, and the cumulative effect will dwarf the current moment of the orange pill moment.

The cost curve does not care about prestige. The most consequential applications of AI will be, by current standards, unpredictable and unprestigious — exactly as the musical greeting card was unpredictable and unprestigious to the engineers who built ENIAC.