The Apex of the Cloud

Big-iron frontier AI might represent both the peak of platform concentration and the beginning of technologies that undermine cloud dependence.

Model & Argument · Initial edition, November 2026

Today's frontier AI requires immense concentrations of hardware, infrastructure, talent and capital. This looks like a culminating expression of the cloud-platform era, in which a small number of providers can serve millions of people from enormous central systems.

One hypothesis is that we are witnessing the apex of the cloud: maximum concentration shortly before some of the capabilities built there become inexpensive enough to run elsewhere.

How the paradox might work

Frontier research can be extremely expensive while inference for already trained models becomes progressively more efficient. Smaller models, better algorithms and cheaper hardware might move useful intelligence to personal computers, phones and robots.

If local intelligence becomes sufficient for most ordinary activities, users may stop depending so heavily on cloud services even when cloud frontier models remain much more capable.

The analogy is not necessarily a victory of weaker models over stronger models. A sufficiently capable local model may offer a different combination of value: privacy, independence and control.

The case against the hypothesis

Frontier models could continue improving much faster than local systems. New capabilities may create demand that never settles at a fixed level. Advanced robotics may require cloud infrastructure or expensive local hardware. Large providers may enjoy lower costs per task at scale.

And the meaning of cloud dependence can evolve: personal agents might control their own memory and decisions while selectively renting remote inference.

The unresolved research

We need trends in capability per watt, memory requirements, model compression, hardware prices, energy and bandwidth use, and the proportion of common activities that local models can perform satisfactorily.

The outcome might be continued cloud domination, a two-tier ecosystem, or widespread local sufficiency. The model is intentionally undecided. See The Local Intelligence Sufficiency Hypothesis.

View page as JSON