The mark of the infrastructure moment is the quiet disappearance of the technology's name from the description of the systems that use it. In 2015 a company would announce "we're integrating AI into our customer support"; in 2025 it announces the support product and the AI is understood. The 2015 announcement was about a capability; the 2025 announcement is about a consequence. The same pattern played out for databases ("database-driven website" was a meaningful category in 1998 and is a meaningless category in 2024), for the internet ("internet-enabled product" meant something distinct in 1996), for electricity ("electric appliance" was an innovation claim in 1910 and a redundant phrase by 1960).
Clarke's Extra-Terrestrial Relays paper is the cleanest historical exemplar of a successful infrastructure-moment forecast. In October 1945 Clarke described an orbit at 36,000 km where a satellite's period would match Earth's rotation, proposed that three such satellites could provide global communications coverage, and predicted the use case would be compelling enough to justify the launch cost within decades. The first geostationary communications satellite, Syncom 3, launched nineteen years later; global telephone traffic moved onto the satellite network through the 1960s and 1970s; the Clarke orbit is now so dense with hardware that launch slots are assigned like radio frequencies. The forecast was correct in timing, in mechanism, and in scale.
For AI the moment is more compressed and more visible. Between November 2022 (ChatGPT's release) and the present, AI has moved from a research demo to a platform that major industries are reorganizing around. Software engineering is the clearest case: AI-assisted coding tools are standard issue in most engineering organizations, and a meaningful fraction of production code is now AI-generated. Customer service, translation, image generation, video generation, legal-document review, medical-record abstraction, and programming education are in parallel transitions. The infrastructure moment is not a single event; it is many simultaneous moments in many sectors, each running on its own clock.
The governance question is what regulatory, educational, and economic adjustments accompany the moment. Electricity's infrastructure moment produced utility commissions, electrical codes, grid-operator professions, and universal-service mandates. The internet's moment produced net neutrality debates, data-protection law, and the ISP-as-common-carrier question. AI's moment is producing export controls on compute, safety-evaluation regimes, sectoral regulation in medicine and finance, and — more slowly than the analogies would suggest — labor-market adjustments. The governance lag is expected; every infrastructure moment has produced it. The specific question for AI is whether the lag is long enough to produce avoidable harm at scale before it closes.
The sociology of infrastructure has a substantial academic literature (Star and Ruhleder 1996, Edwards 2003) that identifies durable features: invisibility in use, visibility only in breakdown, embeddedness in standards, and the relational character of what counts as infrastructure. Clarke's 1945 paper is cited in this entry as a practitioner's forecast; the general theory came later and is now the standard frame for analyzing transitions like the present one.
The infrastructure moment is recognizable by a change in rhetoric. When the technology's name disappears from descriptions of the systems that use it, the moment has arrived.
Forecasting the moment is possible. Clarke did it for satellites; the failure in the AI case would be one of governance response, not recognition.
Governance lags by design. Rules and institutions are built for the pre-moment landscape; they update afterward, which means the early post-moment period has unusually high stakes.
The moment is sectoral, not singular. Industries cross the threshold at different times; a country or company can be mid-moment in some sectors and pre-moment in others.