From the Cloud Era to the AI Era
Why AI could alter the foundations of the global internet and cloud-platform era—and why the end state remains unclear.
Model & Argument · Initial edition, November 2026
The global internet and cloud era combined cheap communications with centralised data centres, services and platforms. Those systems profoundly changed how people communicate, trade and organise work.
Artificial intelligence may be another industrial transition, not merely an additional feature of existing applications.
A strange beginning
Much frontier AI is capital-intensive. Training advanced models and serving demanding workloads often require large concentrations of specialised hardware, energy, expertise and finance. In that sense, big-iron AI appears to extend the cloud era to a new peak of concentration.
Yet the technology being developed is intelligence that can be copied, compressed, adapted and potentially run near the point of use. Model efficiency and local hardware may therefore create pressures in the opposite direction.
The unresolved fork
There is no established outcome here. Frontier systems might continue advancing much faster than local models, preserving a compelling cloud advantage. Alternatively, locally executable intelligence might reach adequate capability for a large portion of practical human activity while frontier models remain technically superior.
We might have both: a powerful cloud frontier and widely useful local intelligence. Or progress may plateau, changing the comparison again.
The crucial research question is practical sufficiency, not model-ranking parity. What do people need a system to do, and how much additional value comes from greater general capability?
Control matters
Cloud AI also comes with dependence on providers, including terms of service, information handling, service availability and potentially conflicting commercial interests. Local intelligence can offer privacy and control, although local systems carry their own security and maintenance burdens.
These are not marginal preferences when AI becomes deeply involved in personal life and physical machinery.
The Apex of the Cloud explores the concentration hypothesis. The Future of Personal Intelligence develops the sufficiency and control arguments. We need real research into capabilities, memory and compute economics before assigning the likely outcome.