Essay Argues AI Defies the Usual Rules of Tech Adoption

A newsletter essay published on 12gramsofcarbon.com under the title "AI is not a normal technology" is pushing a claim that cuts against the assumptions baked into most corporate AI strategies: that artificial intelligence will spread through the economy the way electricity, the car and the internet did. The piece holds that AI's cost structure, capability profile and deployment behavior all deviate from what a standard diffusion model would forecast. That argument matters now because forecasts and budgets across the enterprise software world still rest on those historical analogies.

The essay points to the classic pattern in which a general-purpose technology starts in a narrow niche, then spreads outward until it is absorbed across factories, offices and households. Electrification followed that path over decades; so did the internet. According to the summary accompanying the link, the author contends AI does not track that curve, and that its economics and rollout patterns look different enough to invalidate the comparison.

The piece arrives without fresh data or a named institutional sponsor, which caps how much evidentiary weight it can carry by itself. Its significance is as a signal of a widening disagreement. Through 2025, researchers at schools including MIT and Stanford have published findings indicating that gains from AI adoption cluster in a limited set of companies and job tasks rather than dispersing evenly, the way factory electrification did in the early 1900s.

Executives at several major AI labs have pushed back on that reading, arguing the technology is simply too early in its deployment cycle for the analogy to be judged either way. The dispute is not academic. Enterprises are writing checks at a pace that presumes AI will reach broadly across functions, from customer support to software engineering, and produce returns in each. If the 12gramsofcarbon.com thesis is right — that AI behaves more like a concentrated, capital-hungry system than a ubiquitous utility — then a sizable share of that spending rests on a model that may not hold.

What to watch is whether the argument moves from newsletter contrarianism into the forecasting work that shapes budgets. If analysts and executives begin discounting the diffusion analogy, the assumptions underneath 2026 planning — headcount, tooling, infrastructure — would need rethinking. The open question is what replaces the old curve: a concentrated buildout model, a task-specific one, or something without a useful precedent.